{"id": "2506.13506", "title": "Stimulus Motion Perception Studies Imply Specific Neural Computations in Human Visual Stabilization", "abstract": "Even during fixation the human eye is constantly in low amplitude motion, jittering over small angles in random directions at up to 100Hz. This motion results in all features of the image on the retina constantly traversing a number of cones, yet objects which are stable in the world are perceived to be stable, and any object which is moving in the world is perceived to be moving. A series of experiments carried out over a dozen years revealed the psychophysics of visual stabilization to be more nuanced than might be assumed, say, from the mechanics of stabilization of camera images, or what might be assumed to be the simplest solution from an evolutionary perspective. The psychophysics revealed by the experiments strongly implies a specific set of operations on retinal signals resulting in the observed stabilization behavior. The presentation is in two levels. First is a functional description of the action of the mechanism that is very likely responsible for the experimentally observed behavior. Second is a more speculative proposal of circuit-level neural elements that might implement the functional behavior.", "date": "2025-06-16", "categories": []}
{"id": "2606.09770", "title": "Discovering Functionally Selective Brain Regions with a Deep Topographic Multimodal Model", "abstract": "Nearby neurons in cortex share similar response profiles, producing systematic spatial organization across sensory and cognitive systems. Recent topographic models reproduce aspects of this structure but remain unimodal and spatially constrain each layer separately, yielding fragmented maps that capture neither the contiguity of cortical processing streams nor their integration across modalities. We introduce Topo-Omni, a topographic multimodal model in which visual, auditory, and language/cognitive processing share a single contiguous in-silico sheet. Built by fine-tuning a pretrained foundation model with a spatial smoothness objective, this architecture develops clusters across modalities that are consistent with human neuroimaging, from sensory to cognitive systems. Driving or suppressing a cluster selectively biases or impairs perception, paralleling human intervention studies. Finally, we use our model to screen for novel clusters in-silico and discover new natural landscape and animal networks which we validate in human data. A single spatial principle thus organizes representations across modalities and processing stages, yielding testable hypotheses about cortical organization.", "date": "2026-06-08", "categories": []}
{"id": "2511.05505", "title": "Rewiring Human Brain Networks via Lightweight Dynamic Connectivity Framework: An EEG-Based Stress Validation", "abstract": "In recent years, Electroencephalographic analysis has gained prominence in stress research when combined with AI and Machine Learning models for validation. In this study, a lightweight dynamic brain connectivity framework based on Time Varying Directed Transfer Function is proposed, where TV DTF features were validated through ML based stress classification. TV DTF estimates the directional information flow between brain regions across distinct EEG frequency bands, thereby capturing temporal and causal influences that are often overlooked by static functional connectivity measures. EEG recordings from the 32 channel SAM 40 dataset were employed, focusing on mental arithmetic task trials. The dynamic EEG-based TV-DTF features were validated through ML classifiers such as Support Vector Machine, Random Forest, Gradient Boosting, Adaptive Boosting, and Extreme Gradient Boosting. Experimental results show that alpha-TV-DTF provided the strongest discriminative power, with SVM achieving 89.73% accuracy in 3-class classification and with XGBoost achieving 93.69% accuracy in 2 class classification. Relative to absolute power and phase locking based functional connectivity features, alpha TV DTF and beta TV DTF achieved higher performance across the ML models, highlighting the advantages of dynamic over static measures. Feature importance analysis further highlighted dominant long-range frontal parietal and frontal occipital informational influences, emphasizing the regulatory role of frontal regions under stress. These findings validate the lightweight TV-DTF as a robust framework, revealing spatiotemporal brain dynamics and directional influences across different stress levels.", "date": "2025-10-17", "categories": []}
{"id": "2602.17997", "title": "Whole-Brain Connectomic Graph Model Enables Whole-Body Locomotion Control in Fruit Fly", "abstract": "Animals perform coordinated whole-body movements under the control of neural systems shaped by brain-wide connectivity. The mapping of the whole-brain neural connections, or the connectomes, provides a natural graph for modeling sensorimotor information flow, yet its potential as a neural controller for embodied agents remains largely unexplored. Here, we introduce the Fly-connectomic Graph Model, which directly instantiates the whole-brain connectome of an adult Drosophila as a graph-structured neural controller for movements of a simulated biomechanical fruit fly via deep reinforcement learning. We achieve stable performance across diverse locomotion tasks, as well as better sample efficiency compared to both graph and non-graph baselines. Our results demonstrate a biologically informed way towards effective control policy design by translating whole-brain wiring principles into actionable architectural priors, while also improving the interpretability through dynamic information flow. This work also highlights the potential to bridge neuromechanics with embodied intelligence by providing a computational platform for investigating the sensorimotor transformation underlying animal behavior and a paradigm to advance the development of more nature-aligned intelligent systems.", "date": "2026-02-20", "categories": []}
{"id": "2606.16294", "title": "Sex-based Network-Specific Differences in Connectomes: A Krakencoder-Based Analysis", "abstract": "This study examines how deficiencies in one brain connectome modality propagate to the other, using the Krakencoder as a simulation framework. Structural and functional connectomes from 702 healthy participants in the Human Connectome Project were analyzed, with the impact of each of the Yeo-7 functional networks assessed separately. Seven scenarios were considered, each involving the removal of a single network while the remaining networks were preserved. The resulting perturbations in cross-modal predictions were quantified using three complementary metrics: KL divergence on eigenvalue spectra, Frobenius norm, and Wasserstein distance. In addition, the persistence of sex-specific information within the predicted connectomes was evaluated. Across all metrics and both prediction directions, the Default Mode Network produced the largest perturbations, whereas the Somatomotor network yielded the smallest. Sex differences in network-level perturbation signatures were subtle, with the best result being an accuracy of 66.09% from connectomes predicted under network-removal conditions. In contrast, connectomes predicted from intact inputs achieved substantially higher sex classification accuracy, reaching up to 84.76%. These findings confirm that full predicted connectomes retain considerably more sex-discriminative information than perturbation-derived signatures alone.", "date": "2026-06-15", "categories": []}
{"id": "2606.16693", "title": "Learning Hybrid Biophysical Neuron Models with Neural ODEs", "abstract": "Biophysical neuron models link measurements of neural activity to underlying cellular mechanisms. Yet, a central challenge is that the kinetics of many ion channels are poorly characterized, and practical simplifications -- omitting channels or reducing morphological detail -- introduce systematic gaps between model and biology. Bridging these gaps requires approaches that can flexibly discover unmodeled dynamics while preserving mechanistic interpretability. Here, we introduce a hybrid modeling framework that embeds neural ordinary differential equations into conductance-based biophysical models to capture unknown currents or mis-specified channel kinetics. By parameterizing the neural ODE in terms of voltage-dependent steady-state and time-constant functions, we recover interpretable gating dynamics directly from voltage recordings without assuming a functional form. We show that the hybrid model fits the gating kinetics of 2400 ion channel models and recovers unknown gating dynamics from single current-clamp recordings, generalizing to out-of-distribution stimulus regimes under realistic inputs and parameter misspecification. We also use our method to reduce a multicompartment model of a cortical neuron into a single-compartment hybrid model with a learned axial current, yielding up to an order of magnitude lower computational cost. Together, our results establish a plug-and-play framework for selectively replacing unknown components of conductance-based models with neural ODEs while preserving their mechanistic structure.", "date": "2026-06-15", "categories": []}
{"id": "2507.09011", "title": "From dots to faces: Individual differences in visual imagery capacity predict the content of Ganzflicker-induced hallucinations", "abstract": "A rapidly alternating red and black display known as Ganzflicker induces visual hallucinations that reflect the generative capacity of the visual system. Individuals vary in their degree of visual imagery, ranging from absent to vivid imagery. Recent proposals suggest that differences in the visual system along this imagery spectrum should also influence the complexity of other internally generated visual experiences. Here, we used tools from natural language processing to analyze free-text descriptions of hallucinations from over 4,000 participants, asking whether people with different imagery phenotypes see different things in their mind's eye during Ganzflicker-induced hallucinations. Topic modeling of descriptions revealed that strong imagers described complex, naturalistic content, while weak imagers reported simple geometric patterns. Using crowd-sourced sensorimotor norms, we also found that participants with stronger imagery used language with richer perceptual associations. These findings may reflect individual variation in coordination between early visual areas and higher-order regions relevant for the imagery spectrum.", "date": "2025-07-11", "categories": []}
{"id": "2604.24942", "title": "Independent-Component-Based Encoding Models of Brain Activity During Story Comprehension", "abstract": "Encoding models provide a powerful framework for linking continuous stimulus features to neural activity; however, traditional voxelwise approaches are limited by measurement noise, inter-subject variability, and redundancy arising from spatially correlated voxels encoding overlapping neural signals. Here, we propose an independent component (IC)-based encoding framework that dissociates stimulus-driven and noise-driven signals in fMRI data. We decompose continuous fMRI data from naturalistic story listening into ICs using one subset of the data, and train encoding models on independent data to predict IC time series from large language model representations of linguistic input. Across subjects, a subset of ICs exhibited consistently high predictivity. These ICs were spatially and temporally consistent across subjects and included cognitive networks known to respond during story listening (auditory and language). Auditory component time series were strongly correlated with acoustic stimulus features, highlighting the interpretability of identified component time series. Components identified as noise or motion-related artifacts by ICA-AROMA showed uniformly poor predictive performance, confirming that highly predicted components reflect genuine stimulus-related neural signals rather than confounds. Overall, IC-based encoding models enable analyses at the level of functional networks, accommodating the variability in network locations across individuals and providing interpretable results that are easy to compare across subjects. Code provided at: https://github.com/kamyahari/IC-Encoding-Models.git", "date": "2026-04-27", "categories": []}
{"id": "2509.22050", "title": "BrainPro: Towards Large-scale Brain State-aware EEG Representation Learning", "abstract": "Electroencephalography (EEG) reflects underlying brain states, whose activities are distributed across brain regions and manifest as spatial patterns on the scalp. Learning these spatially structured, state-related patterns requires consistent spatial representations across datasets. However, existing EEG foundation models are typically based on self-attention, which does not preserve location-specific information and struggles to align signals recorded with different channel configurations. Moreover, brain states contain both shared and state-specific regional activity, suggesting that learning neurophysiologically plausible, state-aware representations can complement the shared representations targeted by current models and improve downstream decoding. To address these limitations, we propose BrainPro, a large EEG model that combines a retrieval-based spatial learning mechanism for cross-layout spatial alignment with a brain state-decoupling module that learns both shared and state-specific representations through parallel encoders and region-aware reconstruction. Pre-trained on a large EEG corpus, BrainPro achieves state-of-the-art performance across nine public BCI datasets spanning emotion, motor, speech, stress, mental disease, and attention tasks. Analyses of spatial filters, channel-drop robustness, and encoder contributions further validate the effectiveness of its spatial alignment and state-aware pathways. These results show that BrainPro achieves improved interpretability of learned spatial patterns and produces representations that benefit diverse EEG decoding tasks.", "date": "2025-09-26", "categories": []}
{"id": "2510.03699", "title": "Dissecting Larval Zebrafish Hunting using Deep Reinforcement Learning Trained RNN Agents", "abstract": "Larval zebrafish hunting provides a tractable setting to study how ecological and energetic constraints shape adaptive behavior in both biological brains and artificial agents. Here we develop a minimal agent-based model, training recurrent policies with deep reinforcement learning in a bout-based zebrafish simulator. Despite its simplicity, the model reproduces hallmark hunting behaviors -- including eye vergence-linked pursuit, speed modulation, and stereotyped approach trajectories -- that closely match real larval zebrafish. Quantitative trajectory analyses show that pursuit bouts systematically reduce prey angle by roughly half before strike, consistent with measurements. Virtual experiments and parameter sweeps vary ecological and energetic constraints, bout kinematics (coupled vs. uncoupled turns and forward motion), and environmental factors such as food density, food speed, and vergence limits. These manipulations reveal how constraints and environments shape pursuit dynamics, strike success, and abort rates, yielding falsifiable predictions for neuroscience experiments. These sweeps identify a compact set of constraints -- binocular sensing, the coupling of forward speed and turning in bout kinematics, and modest energetic costs on locomotion and vergence -- that are sufficient for zebrafish-like hunting to emerge. Strikingly, these behaviors arise in minimal agents without detailed biomechanics, fluid dynamics, circuit realism, or imitation learning from real zebrafish data. Taken together, this work provides a normative account of zebrafish hunting as the optimal balance between energetic cost and sensory benefit, highlighting the trade-offs that structure vergence and trajectory dynamics. We establish a virtual lab that narrows the experimental search space and generates falsifiable predictions about behavior and neural coding.", "date": "2025-10-04", "categories": []}
{"id": "2605.29355", "title": "Neural-Behavioral Representation of Natural Whole-body Movement in Monkeys", "abstract": "Understanding how cortical activity represents natural whole-body behaviors in primates remains challenging. Limited by the diversity of movements and inaccessibility of large-scale neural representation of whole-body kinematics, previous motor decoding studies focused on constrained tasks and limited limb movements. Here, we present a neural-behavioral recording and modeling framework for freely moving monkeys, combining large-scale epidural cortical signals from distributed sensory- and motor-related areas with synchronized multi-view motion capture through a custom-made data collection platform. We reconstructed whole-body monkey kinematics and learned a compact behavior prior using an autoregressive encoder-decoder model. Conditioned on neural signals, the model decoded accurate and realistic whole-body movement without explicit physical constraints. Our results provide a novel proof-of-concept approach for decoding natural whole-body movements in primates using large-scale intracranial neural activity.", "date": "2026-05-28", "categories": []}
{"id": "2606.11833", "title": "Flow Matching with In-Context Priors for Out-of-Distribution Brain Dynamics", "abstract": "Flow matching and diffusion models enable conditional generation across domains ranging from images to proteins, with recent extensions to out-of-distribution contexts. Yet generative models of neural time series have largely remained restricted to categorical conditioning, precluding compositional and zero-shot generalization. In this work, we propose a per-timestep conditioned diffusion transformer for generating realistic fMRI brain dynamics during unseen cognitive tasks by injecting both compositional language and optional spatial priors in-context. Such zero-shot generation could enable counterfactual neuroscience by supporting in-silico design and evaluation of novel cognitive experiments before empirical validation. Leveraging this model, we evaluate across hundreds of held-out task conditions and characterize predictive performance in relation to the training manifold. From language alone, the model recovers region-specific recruitment across tasks and held-out spatial activation patterns. Spatial priors, when available, complement the text pathway by anchoring generation in regions of task space where language alone degrades, while retaining the compositional structure needed for counterfactual task specification. To our knowledge this is the first generative model of whole-cortex fMRI dynamics for unseen cognitive tasks, advancing counterfactual neuroscience and data-driven experimental design.", "date": "2026-06-10", "categories": []}
{"id": "2606.10238", "title": "Hyperbolic Neural Population Geometry Benefits Computation", "abstract": "Neural population geometry shapes downstream computation. Recent empirical findings in neurobiology suggest that a hyperbolic structure underlies population activity in the hippocampus. Here we provide a theoretical framework for this phenomenon. First, we propose a plausible construction of hippocampal tuning curves that statistically induces hyperbolic geometry. Next, we establish a connection between neural decoding and associative memory by demonstrating that the Modern Hopfield Network update rule computes the minimum mean-squared-error (MMSE) estimator. Finally, we introduce a novel associative memory model defined in hyperbolic space that yields significantly larger capacity than leading models. Our results suggest that animals encode spatial information as a latent hyperbolic cognitive map, improving both memory capacity and decoding accuracy.", "date": "2026-06-08", "categories": []}
{"id": "2606.10530", "title": "Machine Learning Methods for Studying Latent Neural Activity Dynamics", "abstract": "Recent developments in brain recording are driving a demand for machine learning tools capable of decoding the latent structure of large populations of neurons. In this paper, we provide a comprehensive survey that outlines the trajectory of Latent Variable Models (LVMs) from early state-space models to more recent deep generative models. We organize the literature into three closely related domains: (1) Single-Region Latent Dynamics, which includes models such as linear dynamical systems to more complex dynamics represented by Recurrent Neural Networks (RNNs) and Neural Ordinary Differential Equations (ODEs); (2) Multi-Region Communication, which employs probabilistic as well as subspace methods to study how information is transferred across different brain areas considering synaptic propagation delays and network connectivity; and (3) Behavior-Aligned Modeling, which seeks to disentangle neural activity related to task performance from other internal states via supervised or contrastive learning. This survey also includes large-scale neural foundation models, such as Transformers and diffusion models, that rely on large-scale pre-training for optimal performance across subjects. Finally, we conclude and discuss benchmarks, evaluation criteria, and open challenges, such as the ability to identify causal links or directionality of communication, to facilitate future research for bridging interpretable brain dynamics with reliable neural decoding.", "date": "2026-06-09", "categories": []}
{"id": "2606.11066", "title": "GRAFT: Gain-Recalibrated Adapters for Transformer-Based Neural Population Activity Modeling", "abstract": "Neural population activity models can recover rich temporal structure from binned spikes, but their read-in and readout layers often remain tied to a fixed set of recorded neurons. This coupling limits reuse in long-term brain-computer interfaces, where recorded neuron identities, counts, and response statistics can change across days. We introduce GRAFT, a Transformer-based neural population activity model that separates reusable temporal dynamics from a recalibratable neuron interface. The neuron interface controls how recorded neurons enter and leave the shared backbone, and auxiliary gain and positional mechanisms support neural activity modeling inside the Transformer. On MC Maze under the standard NLB'21 protocol, GRAFT reaches 0.3866 co-bps as an ensemble, setting a new state of the art on the primary co-bps metric among public and reported NLB'21 results. In a cross-day protocol constructed from the NLB'21 MC Maze dataset series, GRAFT recalibrates from MC Maze to the scaled MC Maze datasets (Large/Medium/Small) by updating only 9.21% of parameters, reaching 0.3749, 0.3112, and 0.3152 co-bps with restricted target-day support sets. These results show that the same interface-backbone separation supports both strong Transformer-based neural population activity modeling and data-efficient cross-day recalibration.", "date": "2026-06-09", "categories": []}
{"id": "2606.06857", "title": "Interpreting Brain Responses to Language with Sparse Features from Language Models", "abstract": "A central goal of cognitive neuroscience is to characterize the features that are represented by human language cortex. Artificial language models (LMs) have emerged as a powerful tool to address this challenge, but studies relating biological and artificial representations are often criticized as relating one black box to another. The present work introduces Augmented Sparse Encoding Models, an encoding framework that replaces dense LM hidden states with hierarchically-organized sparse autoencoder (SAE) features, while explicitly including surprisal as a predictor. Using this approach, we (i) produce interpretations of neural responses and (ii) test whether model-brain alignment reflects primary or idiosyncratic variation in LM representations. Using a high-field 7T fMRI dataset of eight participants listening to 200 linguistically diverse sentences, we first validate our modeling framework by recovering previous interpretations of voxel populations tuned to processing difficulty and meaning abstractness. We then interpret a previously-uncharacterized (but reliable) voxel population and find that it is tuned to people-related content. Next, we show that the fronto-temporal human language network is predicted by a common set of features across its constituent regions, but find that frontal regions are relatively well-explained by surprisal alone, even in the absence of LM-based features. Finally, we show that brain responses during language processing are not merely predictable from an arbitrary set of LM features. Rather, brain responses are best explained by the features that tend to capture the most general information encoded in LM representations, suggesting a nontrivial correspondence between brain and LM language representation.", "date": "2026-06-05", "categories": []}
{"id": "2606.05870", "title": "Cross-scale spatially-aware generative modeling of transcriptomic programs underlying neurodegenerative brain organization", "abstract": "Neurodegenerative disorders such as Alzheimer's disease exhibit highly organized patterns of regional brain vulnerability, yet the biological mechanisms underlying this spatial selectivity remain incompletely understood. Existing imaging-transcriptomic studies have largely relied on correlation-based analyses between gene expression and neuroimaging phenotypes, limiting their ability to model how molecular organization gives rise to neurodegeneration. Here, we introduce a cross-scale spatially-aware generative framework for modeling transcriptomic programs underlying cortical neurodegeneration. Regional transcriptomic profiles were derived from the Allen Human Brain Atlas using 910 landmark genes across 68 cortical regions. Neurodegenerative vulnerability maps were constructed from ADNI FreeSurfer cortical thickness measurements by computing regional cortical thinning differences between cognitively normal controls (NC = 926) and Alzheimer's disease subjects (AD = 426). A variational generative architecture was used to learn latent biological programs linking regional gene-expression organization to cortical degeneration while incorporating graph-based spatial smoothness regularization to preserve cortical organization. The proposed framework achieved strong prediction of regional neurodegenerative vulnerability, yielding an explained variance of 0.8604 and a significant spatial correlation between predicted and observed cortical degeneration profiles (r = 0.9439, p < 0.001). The learned latent representations revealed structured transcriptomic organization associated with distributed disease susceptibility. These findings demonstrate that biologically constrained generative modeling can bridge microscale molecular organization with macroscale neurodegeneration, providing a foundation for spatially-aware generative neurobiology and computational neuroscience.", "date": "2026-06-04", "categories": []}
{"id": "2606.06094", "title": "Integrating Mechanistic and Data-Driven Models for Neurological Disorders through Differentiable Programming", "abstract": "Advances in computational modeling, neuroimaging, and artificial intelligence are revolutionizing the modeling of neurological disorders for improved diagnostics, prognosis, and treatment planning. Mechanistic models provide valuable scientific insight into the disorders, but in practice they are often simplified with assumptions or computationally expensive and slow to solve. However, while purely data driven approaches provide speed and scalability, they require large, high quality data to train and generally suffer from interpretability and generalization issues. This perspective paper presents a structured overview of hybrid modeling strategies, which combine deep learning models with physics based solvers, and are categorized into parallel, series, and parallel-series architectures. Three main approaches that have been emphasized are residual modeling for missing or incomplete physics, Neural Ordinary Differential Equations (NODEs) for continuous time dynamics approximation, and solver in the loop that accelerates traditional solvers with neural approximations. These hybrid models integrate the governing differential equation based formulations and deep learning to characterize the evolution of neurological disorders, and promise advanced personalized neurological modeling. In addition, the study explores and proposes different hybrid configurations to improve diagnosis accuracy, predict disease progression, and inform treatment strategies across a range of neurological disorders. These capabilities outperform standalone mechanistic or purely data driven approaches, making hybrid modeling a powerful tool, especially in applications involving modeling the progression and treatment responses in neurological conditions such as brain tumors, Alzheimer's disease, and stroke.", "date": "2026-06-04", "categories": []}
{"id": "2606.04010", "title": "The Variance Brain Foundation Models Forgot: Third-Order Statistics Predict Cognition Where Billion-Parameter Models Fail", "abstract": "Brain foundation models (BFMs) are self-supervised Transformers pretrained on fMRI data. We posit that these models should capture each subject's cognitive performance from their fMRI signal. Yet across three state-of-the-art BFMs and every readout we test, they predict cognition worse than a linear regression from the $\\sim$80K parameters of the functional connectivity matrix (FC). The gap widens with scale: BrainLM's 650M model predicts cognition worse than its 111M. We attribute this to a \\textbf{variance allocation problem}: BFM pretraining captures the variance components that dominate fMRI but not the higher-order structure that predicts cognition. Our per-cumulant analysis of the reconstructed signal shows that the second-order covariance is partially preserved, while the third-order co-skewness tensor is largely destroyed. To recover what BFMs lose, we design a linear pipeline that projects the fMRI signal into the subspace that best preserves its co-skewness and computes FC there. This \\textbf{exceeds raw FC and every pretrained BFM} on every dataset and parcellation we test, outperforming prior state-of-the-art under controlled evaluation \\textbf{with no pretraining and no GPU}. We \\textbf{recover the raw-FC ceiling on BrainLM's forward pass} by finetuning with a loss targeted at this same subspace. This shows that the bottleneck is the pretraining objective, not the architecture or the model size.", "date": "2026-05-29", "categories": []}
{"id": "2606.04339", "title": "Literature-Guided Minimax Optimization of Virtual Epilepsy Neurostimulation", "abstract": "Computational models of epilepsy promise patient-specific treatment design, but most optimization workflows still search for parameters that perform well on average. In neuromodulation, this is a weak target: a protocol that improves the mean response can still fail in the patient whose network is least tolerant to stimulation. We present a literature-guided minimax pipeline that couples PubMed-scale hypothesis extraction, The Virtual Brain (TVB) Epileptor simulations, and large-language-model-guided black-box optimization. The optimizer proposes either intrinsic model-control parameters or clinically interpretable external-stimulation protocols; TVB evaluates each proposal across sampled virtual patients; and the objective maximizes worst-case reward, defined as the negative variance of simulated seizure activity. In the intrinsic model-control experiment, the best archived parameter set improved worst-case reward from -0.5285 to -0.3182, a 39.8% gain over baseline. The clinical-style external-stimulation search produced a much smaller worst-case improvement (1.7%), and a 20-patient virtual cohort showed no aggregate benefit (p=0.9019), despite a 55% responder rate and a positive temporal-lobe subgroup signal. The study should be read as an in silico proof of concept for robust, literature-aware neurostimulation design, not as clinical evidence.", "date": "2026-06-03", "categories": []}
{"id": "2606.04772", "title": "Coarse-to-fine Hierarchical Architecture with Sequential Mamba for Brain Reconstruction", "abstract": "Understanding the relationship between deep visual representations and the human visual system is a fundamental challenge in computational neuroscience. While modern vision models achieve strong performance in image recognition, their correspondence with the hierarchical organization of the human visual cortex remains an open question. In this study, we propose CHASMBrain, a novel hierarchical two-stage framework for image-to-fMRI encoding. Our architecture leverages a dual-stream Mamba design to explicitly separate and process global semantic tokens and local spatial patches, motivated by the functional organization of the visual cortex. A coarse-to-fine strategy is employed: Stage 1 predicts denoised ROI-level activations, while Stage 2 refines these coarse responses into full voxel-level predictions using a Mamba-VAE. Experiments on the Natural Scenes Dataset (NSD) demonstrate that our method achieves a Pearson correlation of 0.429 and an MSE of 0.261, outperforming all evaluated baselines including ridge regression and DINOv2 linear probes. Beyond predictive performance, causal branch-ablation experiments reveal an asymmetric specialization: the patch stream is specifically locked to early visual cortex (retinotopic regions), while the CLS stream contributes broader semantic context to higher-order areas -- a correspondence that holds causally, not merely correlationally. Cross-subject transfer experiments further show that the learned backbone generalizes across individuals with minimal per-subject adaptation, suggesting the model captures a shared, subject-agnostic visual representation.", "date": "2026-06-03", "categories": []}
{"id": "2511.13899", "title": "A Factorized Low-Rank RNN Framework for Uncovering Independent Neural Latent Dynamics and Connectivity", "abstract": "Low-rank recurrent neural networks (lrRNNs) are a class of models that uncover low-dimensional latent dynamics underlying neural population activity. Although their functional connectivity is low-rank, it lacks independence interpretations, making it difficult to assign distinct computational roles to different latent dimensions. To address this, we propose the Factored Recurrent Neural Network (FacRNN), a generative lrRNN framework that assumes group-wise independence among latent dynamics while allowing flexible within-group entanglement. These independent latent groups allow latent dynamics to evolve separately, but are internally rich for complex computation. We reformulate the lrRNN under a variational autoencoder (VAE) framework, enabling us to introduce a partial correlation penalty that encourages independence between groups of latent dimensions. Experiments on synthetic, monkey M1, and mouse voltage imaging data show that FacRNN consistently improves the disentanglement and interpretability of learned neural latent trajectories in low-dimensional space and low-rank connectivity over baseline lrRNNs that do not encourage group-wise independence.", "date": "2025-11-17", "categories": []}
{"id": "2604.04958", "title": "CalM: A Self-Supervised Foundation Model for Population Dynamics in Calcium Imaging Data", "abstract": "Recent work suggests that large-scale, multi-animal modeling can significantly improve neural recording analysis. However, for functional calcium traces, existing approaches remain task-specific, limiting transfer across common neuroscience objectives. To address this challenge, we propose \\textbf{CalM}, a self-supervised neural foundation model trained solely on neuronal calcium traces and adaptable to multiple downstream tasks, including forecasting and decoding. Our key contribution is a pretraining framework, composed of a high-performance tokenizer mapping single-neuron traces into a shared discrete vocabulary, and a dual-axis autoregressive transformer modeling dependencies along both the neural and the temporal axis. We evaluate CalM on a large-scale, multi-animal, multi-session dataset. On the neural population dynamics forecasting task, CalM achieves competitive performance against strong specialized baselines after pretraining. With a task-specific head, CalM further adapts to the behavior decoding task and achieves superior results compared with supervised decoding models. Moreover, linear analyses of CalM representations reveal interpretable functional structures beyond predictive accuracy. Taken together, we propose a novel and effective self-supervised pretraining paradigm for foundation models based on calcium traces, paving the way for scalable pretraining and broad applications in functional neural analysis. Code is released at https://github.com/TSuXinH/CalM.", "date": "2026-04-03", "categories": []}
{"id": "2606.00496", "title": "Torus Graphs for Large Scale Neural Phase Analysis", "abstract": "Oscillatory neural signals such as electroencephalography (EEG) and local field potentials (LFPs) show phase relationships that coordinate communication across brain regions. Modern recordings capture hundreds of channels across many frequency bins, yet standard phase analyses are restricted to only a few variables. The Torus Graph (TG) model, an exponential-family distribution over phases whose univariate and pairwise potentials generalize von Mises distributions, infers principled structure among oscillations but models only static, undirected dependencies and is limited to $\\sim \\! 100$ variables because its score matching inference scales as $\\mathcal{O}(d^{6})$. We introduce a stochastic score matching procedure that reduces the per-iteration cost to $\\mathcal{O}(d^{2})$, enabling inference on datasets with thousands of variables. This scalable foundation supports analyses of 1,860 frequency-phase features from multi-electrode LFPs and enables two extensions previously inaccessible to TGs or classical circular statistics: (i) a TG Hidden Markov Model capturing state-dependent phase-coupling changes (e.g., spindle-related states during sleep) and (ii) an autoregressive TG inferring directional interactions via transfer-entropy estimation. Applied to LFP recordings, these models reveal state-dependent phase-interaction patterns between wakefulness and NREM sleep. Together, they enable systematic, large-scale mapping of dynamic and directional phase relationships across brain and cognitive states.", "date": "2026-05-30", "categories": []}
{"id": "2606.00667", "title": "Cortex and subcortex play distinct roles over learning when cortical memory is limited", "abstract": "It has been proposed that the brain integrates flexible, computationally expensive cortical processing with simpler, lower-cost subcortical mechanisms to achieve resource-efficient performance greater than that of either system alone. Despite the allure of this perspective, satisfying theoretical frameworks that explore this hypothesis are still limited. We extend existing frameworks in which a model-based module and model-free module learn in tandem by explicitly constraining the memory resources of the model-based module, and investigate the impact of this constraint in a simple decision-making setting. Memory constraints naturally give rise to strategies for allocating memory resources. We evaluate the performance of different strategies in different situations and demonstrate that when the rewarded states change often, it can be advantageous for the model-based module to focus its memory resources not on exploiting the current reward, but on capturing general structure of the environment. This work provides a theoretical foundation for a functional dissociation between cortical and subcortical systems during learning: the cortex supports general structure learning, while subcortical circuits specialize in reward-based learning. We further detail how these hypotheses can be tested on experimental data.", "date": "2026-05-30", "categories": []}
{"id": "2606.01468", "title": "Computation-Aware Kalman Filtering with Model Selection for Neural Dynamics", "abstract": "Due to their explicit priors and ability to model uncertainty, Bayesian methods have played a major role in dynamical latent variable modeling of single-cell neural recordings. However, modern-sized datasets have made overparameterized deep networks the preferred methods of choice due to their predictive power and favorable computational scaling. While many posterior approximations exist, all incur approximation errors. Recent work accounts for this error in the form of computational uncertainty but comes at the cost of quadratic complexity and assumes fixed model hyperparameters. Here we extend this development to model selection, including a novel training loss and optimization scheme, which yields tractable inference in large state-spaces. We introduce a framework, the Computation-Aware State-Space Model (CASSM), specifically designed for the scale-imbalanced regime, where the number of trials is significantly lower than the number of recorded neurons. In this regime, for both synthetic and real data, we show that our method is competitive with data-hungry deep networks, with significantly improved uncertainty calibration over previous attempts to scale Bayesian methods. Our experiments provide a roadmap to neuroscience researchers in choosing from a host of potential dynamical latent variable models given key dataset properties and constraints.", "date": "2026-05-31", "categories": []}
{"id": "2410.19153", "title": "Learning Coupled Subspaces for Multi-Condition Spike Data", "abstract": "In neuroscience, numerous studies conduct sensory or behavioral experiments under multiple conditions to acquire neural responses in the form of high-dimensional spike train datasets. Analyzing high-dimensional spike data is a challenging statistical problem. To this end, Gaussian process factor analysis (GPFA), a popular class of latent variable models, has been proposed for data collected under a single experimental condition. GPFA extracts smooth, low-dimensional latent trajectories that summarize highdimensional spike datasets. However, standard GPFA infers these trajectories independently for each experimental condition, not accounting for how the underlying activity varies across the condition space. This poses limitations on both accuracy and the interpretability of the latent representation. To address these limitations, we propose Coupled Subspaces GPFA (CS-GPFA), a Bayesian model that jointly learns latent representations, characterizing how the neural activity varies over the condition space. Building on this, we further develop an active-learning algorithm for adaptively selecting conditions. Experiments on both synthetic and real neural datasets demonstrate that CS-GPFA achieves superior performance compared to existing approaches. Moreover, our active learning results show that CS-GPFA can efficiently guide experiment design in practical settings.", "date": "2024-10-24", "categories": []}
{"id": "2510.16658", "title": "Large-Scale AI and Foundation Models for Neuroscience: A Comprehensive Review", "abstract": "The development of large-scale artificial intelligence (AI) models is influencing neuroscience research by enabling end-to-end learning from raw brain signals and neural data. In this paper, we review applications of large-scale AI models across four major neuroscience domains: neuroimaging and data processing, brain-computer interfaces and neural decoding, clinical decision support and translational frameworks, and disease-specific applications across neurological and psychiatric disorders. These models show potential to address major computational neuroscience challenges, including multimodal neural data integration, spatiotemporal pattern interpretation, and the development of translational frameworks for clinical research. Moreover, the interaction between neuroscience and AI has become increasingly reciprocal, as biologically informed architectural constraints are now incorporated to develop more interpretable and computationally efficient models. This review highlights both the promise of such technologies and critical implementation considerations, with particular emphasis on rigorous evaluation frameworks, effective integration of domain knowledge, prospective clinical validation, and comprehensive ethical guidelines. Finally, a systematic listing of critical neuroscience datasets used to develop and evaluate large-scale AI models across diverse research applications is provided.", "date": "2025-10-18", "categories": []}
{"id": "2605.29591", "title": "Mind-Omni: A Unified Multi-Task Framework for Brain-Vision-Language Modeling via Discrete Diffusion", "abstract": "Modeling the interplay between external stimuli and internal neural representations is a pivotal research area for Brain-Computer Interfaces (BCIs). A major limitation of prior work is the prevailing paradigm of specialized, single-task models, which curtails versatility and neglects inter-task synergies. To address this, we propose Mind-Omni, the first versatile framework that unifies seven distinct encoding and decoding tasks through a discrete diffusion paradigm. At its core is a novel Brain Tokenizer that transforms heterogeneous, continuous brain signals into standardized, discrete tokens. This enables direct, token-level interactions for mutual understanding and generation between any two or more modalities within a shared semantic space. To unlock advanced reasoning capabilities, we further curate a specialized Brain Question Answering (BQA) instruction-tuning dataset. Our model not only establishes a new state-of-the-art among multi-task unified frameworks but also provides strong evidence for multi-task synergy. By demonstrating performance competitive with, and at times superior to, larger specialized models, our work offers a powerful new paradigm for neural modeling and paves the way for foundation models of neural activity. The code is publicly available at https://github.com/ReedOnePeck/Mind-Omni.", "date": "2026-05-28", "categories": []}
{"id": "2605.29850", "title": "MIRAGE: Adaptive Multimodal Gating for Whole-Brain fMRI Encoding", "abstract": "Recent progress in task-optimized neural networks has established encoding models as a powerful tool for predicting brain responses to naturalistic stimuli, yet most existing approaches rely on unimodal representations. The emergence of omni-modal foundation models and rich multimodal neural datasets enables encoding models that jointly integrate visual, auditory, and linguistic information across subjects. We introduce MIRAGE, a brain encoding framework for predicting whole-brain fMRI responses to naturalistic audiovisual stimuli. MIRAGE achieves state-of-the-art performance via a native multimodal backbone and adaptive feature gating across layers. These representations are then combined with a transformer-based brain encoder and a subject-specific linear head over the cortical parcels. Controlled comparisons show that natively multimodal features consistently outperform post-hoc aggregation of independent unimodal features, across architectural levels and backbones. Beyond predictive accuracy, the learned attention weights are directly inspectable to interpret the modality-specific gating profile over the backbone, and each modality traces a distinct anatomical pattern across cortex. Together, these results propose adaptive layer-wise aggregation of natively multimodal features as a generalizable, interpretable, and accurate approach for whole-brain encoding.", "date": "2026-05-28", "categories": []}
{"id": "2605.27929", "title": "Exploratory Experience Shapes the Geometry of Predictive Representations", "abstract": "Active sensing links behavior and learning through an action-perception loop: actions determine the observations used to update internal predictive models of perception, which subsequently guide the next actions. Predictive-coding frameworks provide a natural way to model this process, since internal representations are continuously updated to predict future observations. Here, we ask how exploratory and exploitative behavioral strategies shape these internal predictive representations. We build an online learning agent in a tree-like maze with a controllable parameter regulating the balance between exploratory and exploitative regimes. The agent updates a predictive-coding-based perception model from experience generated by its own behavior. The model predicts both future maze states and reward probability, allowing the agent to select actions either by expected information gain during exploration or by predicted reward during exploitation. We show that the resulting internal predictive representations depend strongly on the agent's behavioral regime. Exploratory agents develop representations that are more spatially organized and better preserve the structure of maze transitions in latent space. In contrast, exploitative agents learn less organized representations. We then train this predictive model on natural trajectories of water-deprived mice navigating the same maze and compare the resulting representations with those learned from agent trajectories. More exploratory mice show representational geometries that closely match those of exploratory agents, whereas mice with more restricted visitation patterns resemble reward-driven, exploitative agents. Together, these findings suggest that exploration enables predictive models to form generalized internal representations by organizing latent space around both spatial location and transition context in artificial agents and animals.", "date": "2026-05-27", "categories": []}
{"id": "2406.12179", "title": "The Wisdom of a Crowd of Brains: A Universal Brain Encoder", "abstract": "Image-to-fMRI encoding is important for both neuroscience research and practical applications. However, such \"Brain-Encoders\" have been typically trained per-subject and per fMRI-dataset, thus restricted to very limited training data. In this paper we propose a Universal Brain-Encoder, which can be trained jointly on data from many different subjects/datasets/machines. What makes this possible is our new voxel-centric Encoder architecture, which learns a unique \"voxel-embedding\" per brain-voxel. Our Encoder trains to predict the response of each brain-voxel on every image, by directly computing the cross-attention between the brain-voxel embedding and multi-level deep image features. This voxel-centric architecture allows the functional role of each brain-voxel to naturally emerge from the voxel-image cross-attention. We show the power of this approach to (i) combine data from multiple different subjects (a \"Crowd of Brains\") to improve each individual brain-encoding, (ii) quick & effective Transfer-Learning across subjects, datasets, and machines (e.g., 3-Tesla, 7-Tesla), with few training examples, and (iii) use the learned voxel-embeddings as a powerful tool to explore brain functionality (e.g., what is encoded where in the brain).", "date": "2024-06-18", "categories": []}
{"id": "2509.12783", "title": "Fast reconstruction of degenerate populations of conductance-based neuron models from spike times", "abstract": "Inferring the biophysical parameters of conductance-based models (CBMs) from experimentally accessible recordings remains a central challenge in computational neuroscience. Spike times are the most widely available data, yet they reveal little about which combinations of ion channel conductances generate the observed activity. This inverse problem is further complicated by neuronal degeneracy, where multiple distinct conductance sets yield similar spiking patterns. We introduce a method that addresses this challenge by combining deep learning with Dynamic Input Conductances (DICs), a theoretical framework that reduces complex CBMs to three interpretable feedback components governing excitability and firing patterns. Our approach first maps spike times to DIC densities at threshold using a neural network that learns a low-dimensional representation of neuronal activity. The predicted DIC values are then used to generate degenerate CBM populations via an iterative compensation algorithm, ensuring compatibility with the intermediate target DICs, and thereby reproducing the corresponding firing patterns, even in high-dimensional models. Applied to two models, this algorithmic pipeline reconstructs spiking and bursting regimes with high accuracy and robustness to variability, including spike trains generated under noisy current injection mimicking physiological stochasticity. It produces diverse degenerate populations within milliseconds on standard hardware, enabling scalable and efficient inference from spike recordings alone. Together, this work positions DICs as a practical and interpretable link between experimentally observed activity and mechanistic models. By enabling fast and scalable reconstruction of degenerate populations directly from spike times, our approach provides a powerful way to investigate how neurons exploit conductance variability to achieve reliable computation.", "date": "2025-09-16", "categories": []}
{"id": "2512.19097", "title": "DIVER-1: Scaling Intracranial EEG Foundation Models for Transferable Representations", "abstract": "Intracranial EEG (iEEG) provides direct, millisecond-scale recordings of human neural activity, but reusable representation learning is difficult because electrode layouts, anatomical coverage, referencing schemes, and recording conditions vary across patients and centers. We introduce DIVER-1, a self-supervised iEEG foundation model for variable-input recordings that combines any-variate electrode-time attention, spatio-temporal resampling, input-conditioned positional embeddings, and multi-domain masked reconstruction without assuming a fixed electrode montage. We pretrain two variants, DIVER-1-0.1s and DIVER-1-1s, on 5,310 hours of ECoG and SEEG spanning 352k channel-hours, roughly 54x the BrainTreeBank-based pretraining volume. We evaluate DIVER-1 on two held-out benchmarks: Neuroprobe for naturalistic cognitive decoding and MAYO for seizure detection. On leakage-aware Neuroprobe, DIVER-1-0.1s outperforms prior evaluated iEEG foundation models despite using no BrainTreeBank recordings, the corpus underlying Neuroprobe, during pretraining; it also exceeds the linear spectrogram decoder in mean AUROC and remains competitive with stronger nonlinear baselines, a level prior evaluated iEEG foundation models did not reach. DIVER-1-1s also achieves the top AUROC on MAYO seizure detection. Finally, we conduct, to our knowledge, the first controlled compute-aware scaling study for self-supervised iEEG pretraining, sweeping data scale, subject count, training duration, and model size up to 1.8B parameters. Our results indicate a data-constrained regime: expanding unique recordings and training sufficiently long are more reliable scaling axes than increasing parameter count alone. Code is available at link.", "date": "2025-12-22", "categories": []}
{"id": "2605.16953", "title": "How do Humans Process AI-generated Hallucination Contents: a Neuroimaging Study", "abstract": "While AI-generated hallucinations pose considerable risks, the underlying cognitive mechanisms by which humans can successfully recognize or be misled by these hallucinations remain unclear. To address this problem, this paper explores humans' neural dynamics to characterize how the brain processes hallucinated content. We record EEG signals from 27 participants while they are performing a verification task to judge the correctness of image descriptions generated by a multi-modal large language model (MLLM). Based on an averaged event-related potential (ERP) study, we reveal that multiple cognitive processes, e.g., semantic integration, inferential processing, memory retrieval, and cognitive load, exhibit distinct patterns when humans process hallucinated versus non-hallucinated content. Notably, neural responses to hallucinations that were misjudged versus correctly judged by human participants showed significant differences. This indicates that misjudged AI-generated hallucinations failed to trigger the standard neurocognitive fact verification pathway.", "date": "2026-05-16", "categories": []}
{"id": "2605.23996", "title": "Brain-to-Image Retrieval and Reconstruction via Multimodal EEG Alignment", "abstract": "We present a brain-to-image system that decodes visual stimuli from EEG signals recorded during natural image viewing. Our system addresses two tasks: (1) EEG-to-image retrieval, which ranks the correct stimulus image among 200 candidates given an EEG segment, and (2) EEG-to-image reconstruction, which generates an image consistent with the perceived stimulus. For retrieval, we implement a multi-level blurring approach improved with biologically inspired EVNet features and trained with the InfoNCE loss. Evaluated over 10 random seeds for a single subject, the retrieval model achieves a mean final-epoch Top-1 accuracy of 86.30% and Top-5 accuracy of 98.55%. For reconstruction, we implement CognitionCapturerPro, which aligns EEG representations to multi-modal CLIP embeddings, including image, text, depth, and edge embeddings, and synthesizes images with SDXL-Turbo conditioned via IP-Adapter. Averaged over 10 seeds, the reconstruction model achieves a CLIP score of 0.903 using ViT-H-14, a CLIP score of 0.870 using ViT-L/14, and an SSIM of 0.409. These results demonstrate the feasibility of decoding rich visual representations from EEG signals using modern multi-modal alignment and generative modeling techniques.", "date": "2026-05-18", "categories": []}
{"id": "2605.24679", "title": "MindAdapter: Few-Shot Parameter-Efficient Residual Calibration of Cross-Subject Brain-to-Visual Decoding Models", "abstract": "Cross-subject brain-to-visual decoding remains a core challenge in brain-computer interfaces due to severe inter-individual variability that induces systematic subject-specific functional misalignment. To address this issue, we propose MindAdapter, a parameter-efficient few-shot calibration framework for pretrained brain-to-visual decoding models. MindAdapter adopts a decoupled linear-residual cascade alignment paradigm by freezing a pretrained explicit brain functional alignment backbone (coarse) and introducing a lightweight nonlinear residual adapter (fine), thereby disentangling global cross-subject correspondence from subject-specific residual corrections for fine-grained spatial and semantic calibration. To further preserve global representational stability, we design a topology-anchored dual-stream manifold constraint, where a small set of shared stimuli serves as topological pins with voxel-level paired supervision, while a semantic stream enforces consistency through a frozen vision-language decoder on unpaired brain data. Together, MindAdapter efficiently injects subject-specific corrections while maintaining the global representational geometry learned during pretraining. Experiments on the Natural Scenes Dataset (NSD) demonstrate that MindAdapter substantially improves cross-subject visual reconstruction and retrieval accuracy using only a few shared stimuli, offering a practical and data-efficient solution for personalized brain-to-visual decoding.", "date": "2026-05-23", "categories": []}
{"id": "2605.15652", "title": "Bridging Silicon and the Hippocampus: Algebro-Deterministic Memory \"VaCoAl\" as a Substrate for Vector-HaSH and TEM", "abstract": "Vector-HaSH and the Tolman-Eichenbaum Machine (TEM) propose the hippocampal-entorhinal circuit factorizes memory via a grid-cell scaffold for compositional replay. Concurrently, human iEEG shows sharp-wave ripples gate recall and multi-hop replay fidelity decays multiplicatively. Yet, these fields lack a shared algebraic foundation. We introduce VaCoAl, an algebro-deterministic hyperdimensional memory architecture built on Galois-field linear-feedback shift registers. Its deterministic Galois-field diffusion offers a substrate-level alternative to Vector-HaSH's random projections, matching quasi-orthogonality while ensuring bit-exact reproducibility. Furthermore, the path-integral Confidence Ratio CR2 provides an algebraically tractable model for the empirically observed multiplicative replay decay. Biologically, VaCoAl's two operating regimes align with the EC-CA3 direct and EC-DG-CA3 trisynaptic pathways, explaining their 520-Myr conservation. Independent cellular evidence supports that the DG-CA3 pathway implements a biophysical homologue of Galois-field arithmetic. We also link this framework to Judea Pearl's Ladder of Causation. Reversible GF(2) binding provides the surgical algebra for the do-operator (Rung 2), and VaCoAl's dual-orthogonalizer architecture supplies the parallel substrate required for counterfactual reasoning (Rung 3). Ultimately, we prove these formal correspondences and derive testable iEEG predictions, uniting computational neuroscience, electrophysiology, and hyperdimensional computing.", "date": "2026-05-15", "categories": []}
{"id": "2605.23035", "title": "Sparse Autoencoders Map Brain-LLM Alignment onto Cortical Semantic Topography", "abstract": "Intermediate layers of large language models (LLMs) best predict human brain responses to language, one of the most robust findings in computational neurolinguistics, yet why remains mechanistically unexplained. We address this gap by bridging sparse autoencoders (SAEs) from mechanistic interpretability with neural encoding models, decomposing GPT-2 XL and Llama-3.1-8B into 16K-32K interpretable features per layer. A human-validated taxonomy ($\\kappa \\geq 0.74$) reveals that semantic features alone recover 94% of peak encoding performance ($r=0.285$), substantially exceeding variance-matched baselines ($p<0.001$, $d=1.31$). Beyond this aggregate dominance, we test a novel cortical topography prediction: five semantic subcategories derived a priori from three independent neuroscience programs should map onto distinct brain regions. A formal convergence test confirms this alignment (Spearman $\\rho=0.72$, $p<0.001$; hypergeometric $p=0.007$), demonstrating that SAE-discovered features recapitulate known cortical semantic organization at a granularity inaccessible to prior methods. SAE features further predict human reading times beyond lexical controls ($\\Delta\\mathrm{logLik}=38.4$, $p<0.001$), and an exploratory prediction-error analysis provides preliminary evidence that the brain additionally encodes unexpected semantic content. Results generalize across English, Chinese, and French.", "date": "2026-05-21", "categories": []}
{"id": "2605.23895", "title": "From Activation to Causality: Discovery of Causal Visual Representations in the Human Brain", "abstract": "Identifying which brain regions represent a visual concept in the human brain is a central challenge in neuroscience. Existing approaches have localized coarse functional regions (e.g., faces, places) through activation maximization, identifying regions that activate strongly for a target concept relative to other concepts. Yet strong activation alone does not establish that a region represents the concept itself, as responses may instead be driven by correlated visual or semantic cues. We introduce BrainCause, an automated framework that combines generative and brain models to synthesize controlled stimuli and validate neural representations through targeted causal testing. Given a query specifying a concept of interest, our framework constructs targeted stimulus sets comprising concept images, counterfactual edits that remove the target concept while preserving other image content, and images with candidate correlated distractors. It then uses an image-to-fMRI encoding model to predict brain responses and searches for representations that respond specifically to the target concept over correlated alternatives. BrainCause returns validated candidate representations and proposes follow-up fMRI experiments to further test or extend its discoveries. Our approach successfully recovers known functional localizations and identifies new candidate representations across dozens of concepts, validated on both predicted and measured fMRI data. Critically, we show that without causal validation, a large fraction of localizations would be false positives, confirming that activation alone is insufficient evidence of representation.", "date": "2026-05-22", "categories": []}
{"id": "2605.19965", "title": "Normative Networks for Source Separation via Local Plasticity and Dendritic Computation", "abstract": "Blind source separation (BSS) is a natural framework for studying how latent causes may be recovered from sensory mixtures, but deriving online and biologically plausible algorithms for structured (i.e., constrained to known domains) and potentially correlated sources remains challenging. Recent work has derived neural networks for BSS from maximization of an entropy measure, yet its online implementations involve complex and nonlocal recurrent dynamics. Motivated by this perspective, we propose Predictive Entropy Maximization, which achieves competitive performance in BSS, using only local weight updates. The method employs a close approximation of an entropy measure, yielding an objective function with easily interpretable components. Minimizing this objective leads to a predictive neural architecture in which feedforward synapses follow an error-driven rule (that can be realized through dendritic mechanisms), lateral inhibitory connections are learned with local Hebbian plasticity, and source-domain constraints are enforced through simple output nonlinearities. We derive explicit spectral bounds on the surrogate error, characterizing when the approximation is accurate. Empirically, Predictive Entropy Maximization remains robust under increasing source correlation and observation noise, outperforms biologically plausible algorithms that rely on stronger independence or decorrelation assumptions, and remains competitive with exact determinant- and correlative-information-based baselines. These results show how local plasticity and adaptive lateral inhibition can emerge from maximizing a regularized second-order entropy over structured source domains. Our implementation code is available at https://github.com/BariscanBozkurt/Predictive-Entropy-Maximization.", "date": "2026-05-19", "categories": []}
{"id": "2605.22206", "title": "Temporal Coding as a Substrate for Sensorimotor Object Inference: A Spiking Reinterpretation of Thousand Brains Architecture", "abstract": "The Thousand Brains Theory (TBT) and its open-source Monty framework model object recognition through sensorimotor inference -- identifying objects by actively moving a sensor across their surface and building evidence contact by contact. The current implementation encodes each contact as a dense floating-point vector. While Monty tracks inter-step displacement and accumulates evidence across contacts, it treats the feature activation pattern at each contact as an unordered set - the directional sequence in which features are encountered carries no representational weight. In TBT, the sequence of contacts carries spatial meaning: knowing that feature A was felt before feature B during a left-to-right sweep tells you something about where A and B sit on the object. Dense vectors discard this ordering. We propose replacing dense vectors with rank-order spike packets: each contact produces a brief burst of neural events where the most strongly activated neuron fires first. The time gap between successive bursts implicitly encodes sensor displacement without explicit coordinate calculations. A biologically motivated learning rule (STDP) encodes traversal direction into synaptic weights. A learnable parameter lambda adjusts reliance on earlier versus recent contacts, adapting to each object's geometry. We derive three testable predictions and specify an implementation of four components in approximately 450 lines of NumPy. Three synthetic experiments confirm the core claims: temporal coding achieves perfect discrimination accuracy on objects with identical features in different spatial arrangements, where dense accumulation performs at chance; temporal coding maintains a 30-50 percentage point advantage across all tested noise levels; the adaptive lambda converges to distinct values, reflecting object geometric complexity. End-to-end evaluation on Monty's YCB benchmark is left for future work.", "date": "2026-05-21", "categories": []}
{"id": "2605.22401", "title": "Cross-Species RSA Reveals Conserved Early Visual Alignment but Divergent Higher-Area Rankings Across Human fMRI and Macaque Electrophysiology", "abstract": "Does the relationship between learning rules and brain alignment generalize across species? We extend our prior finding that untrained CNNs match backpropagation at human V1 by testing the same five learning rules against macaque electrophysiology. The rules are backpropagation (BP), feedback alignment (FA), predictive coding (PC), spike-timing-dependent plasticity (STDP), and an untrained random-weights baseline. The macaque data come from two datasets: MajajHong2015 (V4/IT, 3,200 stimulus presentations, 88/168 neurons) and FreemanZiemba2013 (V1/V2, 135 stimuli, 102/103 neurons). Using RSA with identical model weights from our human study, we find: (1) all models achieve higher alignment with macaque early visual cortex (rho = 0.15-0.30 at V1/V2) than with human fMRI (rho = 0.01-0.08), consistent with the higher signal-to-noise ratio of electrophysiology; (2) STDP and PC produce the highest macaque V1/V2 alignment (rho ~ 0.30 and 0.28), consistent with their leading position among trained rules in human V1; (3) at IT, learning rule rankings show no detectable correlation across species (Kendall's tau = 0.00, p = 1.00), though this null result is expected given that n = 5 provides power only at tau = +/-1.0, and is further confounded by stimulus set differences; (4) a pretrained ResNet-50 (ImageNet) achieves rho = 0.25 at macaque IT, substantially above all custom CNN conditions (rho = 0.07-0.14), suggesting IT alignment is limited by model capacity and training data rather than by the learning rule. Noise ceilings, multi-seed variability (5 seeds), and a stimulus-control analysis are reported. These results demonstrate that early visual alignment is robust across species, while higher-area alignment is modulated by model capacity and stimulus domain.", "date": "2026-05-21", "categories": []}
{"id": "2506.08277", "title": "Task-conditioned probing of instruction-tuned multimodal LLMs: Region-specific brain alignment patterns under naturalistic stimuli", "abstract": "Recent voxel-wise multimodal brain encoding studies have shown that multimodal large language models (MLLMs) exhibit a higher degree of brain alignment compared to unimodal models. More recently, instruction-tuned multimodal (IT) models have been shown to generate task-specific representations that align strongly with brain activity, yet most prior evaluations focus on unimodal stimuli or non-instruction-tuned models under multimodal stimuli. We still lack a clear understanding of whether instruction-tuning is associated with IT-MLLMs organizing their representations around functional task demands or if they simply reflect surface semantics. To address this, we estimate brain alignment by predicting fMRI responses recorded during naturalistic movie watching (video with audio) from MLLM representations. Using instruction-specific embeddings from six video and two audio IT-MLLMs, across 13 video task instructions, we find that instruction-tuned video MLLMs show higher brain alignment than in-context learning (ICL) multimodal models (~9%), non-instruction-tuned multimodal models (~15%), and unimodal baselines (~20%). Our evaluation of MLLMs across video and audio tasks, and language-guided probing produces distinct task-specific MLLM representations that vary across brain regions. We also find that ICL models show strong semantic organization (r=0.78), while IT models show weak coupling to instruction-text semantics (r=0.14), consistent with task-conditioned subspaces associated with higher brain alignment. These findings are consistent with an association between task-specific instructions and stronger brain-MLLM alignment, and open new avenues for mapping joint information processing in both systems. We make the code publicly available [https://github.com/subbareddy248/mllm_videos].", "date": "2025-06-09", "categories": []}
{"id": "2605.20496", "title": "Platonic Representations in the Human Brain: Unsupervised Recovery of Universal Geometry", "abstract": "The Strong Platonic Representation Hypothesis suggests that representational convergence in artificial neural networks can be harnessed constructively: embeddings can be translated across models through a universal latent space without paired data. We ask whether an analogous geometry can be recovered across human brains. Using fMRI data from the Natural Scenes Dataset, we propose a self-supervised encoder that learns subject-specific embeddings from brain data alone by exploiting repeated stimulus presentations. We show that these independently learned spaces can be translated across subjects using unsupervised orthogonal rotations, without paired cross-subject samples or intermediate model representations. Synchronizing pairwise rotations into a single shared latent space further improves cross-subject retrieval, indicating that subject-specific spaces are mutually compatible with a common coordinate system. These results provide evidence for a shared neural geometry in the human visual cortex: subject-specific fMRI representations are approximately isometric across individuals and can be translated through purely geometric transformations.", "date": "2026-05-19", "categories": []}
{"id": "2605.21324", "title": "Stimulus symmetries can confound representational similarity analyses", "abstract": "What can representational similarity matrices (RSMs) tell us about a neural code? As the popularity of these summary statistics grows, so too does the need for a more complete characterization of their properties. Here, we show that symmetries in network inputs can confound RSM-based analyses. Stimulus symmetries render many representations functionally equivalent, but these different configurations can lead to different RSMs. These different RSMs reflect qualitatively different representational geometries. We show that stochastic gradient descent or energetic regularization can generate sparse, drifting codes, leading in turn to drifting RSMs. Moreover, we demonstrate that these phenomena are present in networks trained to encode image data, where the symmetry is latent. Our results illustrate the challenges inherent in comparing nonlinear neural codes, when functionally-equivalent representations are not related by a simple rotation.", "date": "2026-05-20", "categories": []}
{"id": "2602.07570", "title": "How does longer temporal context enhance multimodal narrative video processing in the brain?", "abstract": "Understanding how humans and artificial intelligence systems process complex narrative videos is a fundamental challenge at the intersection of neuroscience and machine learning. This study investigates how the temporal context length of video clips (3--24 s clips) and the narrative-task prompting shape brain-model alignment during naturalistic movie watching. Using fMRI recordings from participants viewing full-length movies, we examine how brain regions sensitive to narrative context dynamically represent information over varying timescales and how these neural patterns align with model-derived features. We find that increasing clip duration substantially improves brain alignment for multimodal large language models (MLLMs), whereas unimodal video models show little to no gain. Further, shorter temporal windows align with perceptual and early language regions, while longer windows preferentially align higher-order integrative regions, mirrored by a layer-to-cortex hierarchy in MLLMs. Finally, experiments with four narrative-task prompts show that they elicit task-specific, region-dependent brain alignment patterns and context-dependent shifts in clip-level tuning in higher-order regions. Our work positions long-form narrative movies as a principled testbed for studying long-timescale temporal integration in long-context MLLMs and its relationship to cortical responses during narrative comprehension.", "date": "2026-02-07", "categories": []}
{"id": "2605.19155", "title": "Efficient coding along the visual hierarchy", "abstract": "Biological visual systems learn from limited experience, unlike deep learning models that rely on millions of training images. What learning principles make this possible? We tested whether efficient coding, the idea that neural representations capture the statistical structure of natural inputs, can build a hierarchy of human-aligned visual features from limited data. We developed an unsupervised learning procedure in which each layer of a deep network compresses its inputs onto the dominant modes of variation in natural images, using only local statistics and no labels, tasks, or backpropagation. This unsupervised procedure yields features that progress from edges and colors to textures and shapes. The features of this deep efficient coding model are readily recognized by human observers and are predictive of image-evoked fMRI responses in human visual cortex. Furthermore, a hybrid learning procedure that combines efficient coding with supervised fine-tuning yields better brain alignment in low-data settings and more rapid category learning. These findings suggest that efficient coding may shape representations across the entire visual hierarchy and help explain the data efficiency of biological vision.", "date": "2026-05-18", "categories": []}
{"id": "2605.20127", "title": "Beyond Prediction Accuracy: Target-Space Recovery Profiles for Evaluating Model-Brain Alignment", "abstract": "Artificial vision models are often evaluated against the human visual cortex by measuring how accurately their internal representations predict brain responses. However, prediction accuracy alone does not indicate which dimensions of the target brain's response space are recovered. Here, we introduce a unified framework for evaluating both model-brain and brain-brain alignment by identifying the response dimensions recovered by prediction. Using repeated fMRI measurements, we first identify target-brain response dimensions that can be reproducibly predicted across independent trial splits. We then predict target-brain responses from either another subject's brain responses or a vision model's internal representations, and quantify how strongly each of these reproducible response dimensions is recovered. Applying this framework to a subset of the Natural Scenes Dataset, in which eight subjects viewed the same natural images during fMRI, we find that the early-to-intermediate visual-cortex responses contain a low-dimensional set of reproducible dimensions. Brain-to-brain comparisons identify which of these dimensions are consistently recoverable from other subjects' brains, providing a diagnostic human reference rather than only a scalar benchmark. In some cases, pretrained and randomly initialized models achieve similar prediction accuracy while showing distinct recovery profiles across these response dimensions. These results show that prediction accuracy alone can mask model-brain mismatches. By making explicit which reproducible brain response dimensions are recovered by prediction, our framework provides a more diagnostic evaluation of alignment between artificial vision models and the human visual cortex.", "date": "2026-05-19", "categories": []}
{"id": "2603.03190", "title": "Expectation and Acoustic Neural Network Representations Enhance Music Identification from Brain Activity", "abstract": "During music listening, cortical activity encodes both acoustic and expectation-related information. Prior work has shown that ANN representations resemble cortical representations and can serve as supervisory signals for EEG recognition. Here we show that distinguishing acoustic and expectation-related ANN representations as teacher targets improves EEG-based music identification. Models pretrained to predict either representation outperform non-pretrained baselines, and combining them yields complementary gains that exceed strong seed ensembles formed by varying random initializations. These findings show that teacher representation type shapes downstream performance and that representation learning can be guided by neural encoding. This work points toward advances in predictive music cognition and neural decoding. Our expectation representation, computed directly from raw signals without manual labels, reflects predictive structure beyond onset or pitch, enabling investigation of multilayer predictive encoding across diverse stimuli. Its scalability to large, diverse datasets further suggests potential for developing general-purpose EEG models grounded in cortical encoding principles.", "date": "2026-03-03", "categories": []}
{"id": "2605.16418", "title": "Neural Visual Decoding via Cognitive guided Adaptive Blurring and Information Constrained Alignment", "abstract": "EEG-based visual decoding aims to establish a mapping between neural signals and visual semantics. However, it remains constrained by the dual challenges of severe information granularity mismatch and the low signal-to-noise ratio (SNR) of EEG signals. Existing approaches typically treat static visual features, ignoring the dynamic selectivity of human vision and the frequency specificity of neural oscillations. To bridge this gap, we propose CAIA, a Cognitive-guided Adaptive blurring with Information-Constrained Alignment framework for Neural-Visual decoding. On the visual side, it simulates selective attention to adaptively reduce redundancy. Meanwhile, on the EEG side, it leverages neural oscillation priors and the information bottleneck mechanism to enhance SNR. Specifically, we devise a cognitive-dynamics-based adaptive blurring mechanism that dynamically integrates center-biased and saliency-guided visual cues via cross-modal attention. Furthermore, we introduce a distribution-aware boundary calibration loss to robustly rectify alignment bias caused by outlier samples. Moreover, a cognitively-guided information-screening method is proposed to select task-relevant EEG oscillations. Extensive experiments demonstrate that CAIA improves both subject-dependent and subject-independent average Top-1 and Top-5 accuracy in zero-shot brain-to-image retrieval, significantly outperforming prior methods. Our work validates that optimizing visual information density to match neural granularity offers a more interpretable and robust pathway for neural decoding.", "date": "2026-05-14", "categories": []}
{"id": "2605.16468", "title": "Mechanistically Interpretable Neural Encoding Reveals Fine-Grained Functional Selectivity in Human Visual Cortex", "abstract": "A central goal in understanding human vision is to uncover the visual features that drive neuronal activity. A growing body of work has used artificial neural networks as encoding models to predict cortical responses to natural images, revealing the visual content that activates category-selective regions. However, existing approaches are largely correlational and treat the encoder as a black box, leaving open which image features drive each voxel's response. We introduce Mechanistically Interpretable Neural Encoding (MINE), a framework that opens this black box by applying mechanistic-interpretability tools to localize the features within natural images that drive millimeter-scale (voxel-level) activity. MINE predicts each voxel's response using language-aligned image representations, and produces semantically interpretable descriptions of the features critical for the voxel's activation. We further generalize these per-image features into per-voxel functional profiles. To validate the per-image descriptions, we show they are sufficient to generate images that elicit voxel responses matching the responses to the original images, more accurately than images generated from random or low-attribution controls. Moreover, counterfactually inserting or removing the predicted features from images shifts activation in the expected direction, providing causal evidence. Counterfactual editing guided by the per-voxel activation profiles produces even stronger activation shifts, indicating that the profiles faithfully capture each voxel's selectivity. Finally, we apply MINE to well-studied category-selective brain regions, showing it recovers their known categorical preferences while revealing fine-grained unique voxel structure within each region. Overall, our results establish mechanistic interpretability as a path to discover and causally validate fine-grained hypotheses about neural function.", "date": "2026-05-15", "categories": []}
{"id": "2605.17562", "title": "Beyond Accuracy: Robustness, Interpretability and Expressiveness of EEG Foundation Models", "abstract": "EEG foundation models (EEG-FMs) have been evaluated predominantly on clean, in-distribution accuracy, leaving their robustness, interpretability and representational quality largely unexamined. This study addresses these gaps by benchmarking six EEG-FMs against a baseline deep learning model across eight datasets. Beyond clean accuracy, we conduct three layers of analysis: (i) Robustness: we apply test-time perturbations including additive noise, random and region-based channel dropout and region-specific noise injection. Our analyses show that no single model dominates all failure modes. The most noise-robust model is among the most fragile under channel dropout and much of the dropout fragility disappears when channels are removed rather than zero-padded. (ii) Interpretability: we present the first application of Attention-Aware Layer-Wise Relevance Propagation (AttnLRP) to EEG-FMs and show that models broadly concentrate relevance on task-appropriate brain regions consistent with known neurophysiology. However, attribution maps remain spatially stable under perturbation while predictions degrade, suggesting that the models attend to the correct brain regions but decode corrupted content. (iii) Expressiveness: With block-wise probing we show that late blocks are repurposed during fine-tuning, while early blocks already hold task-related information. Furthermore, we demonstrate that the poor head-only performance previously attributed to low-quality pre-trained representations is largely explained by pooling and that EEG-FMs possess sufficient representational capacity when their token-level embeddings are preserved. Together, these findings provide the first systematic assessment of robustness, interpretability and expressiveness for EEG-FMs and highlight critical considerations for their development.", "date": "2026-05-17", "categories": []}
{"id": "2605.18251", "title": "Subject-Specific Analysis of Self-Initiated Attention Shifts from EEG with Controlled Internal and External Attention Conditions", "abstract": "Self-initiated attention shifts play a critical role in voluntary behavior but are difficult to study due to the absence of explicit temporal markers. While previous studies have examined their neural correlates, it remains unclear how multi-dimensional electroencephalography (EEG) features contribute to their characterization within an interpretable computational framework. In this study, we build on an experimental paradigm developed in our previous work, which enables controlled comparison between task-constrained self-initiated shifts and externally instructed shifts under identical visual stimulation. Within this setting, we investigate whether preparatory EEG activity can distinguish these two types of attention shifts. We adopt a machine learning-based approach and conduct two complementary analyses: (1) a performance-oriented assessment of frequency-specific topographic patterns, and (2) a model-based feature attribution analysis using SHapley Additive exPlanations (SHAP). These analyses provide a structured view of how spectral features across regions of interest contribute to model behavior. Our results demonstrate reliable within-subject classification performance, indicating that preparatory EEG activity contains subject-specific discriminative information within this paradigm. The analysis shows that higher-frequency bands and frontal regions contribute strongly to model decisions, although such contributions should be interpreted cautiously due to the potential influence of non-neural artifacts in high-frequency EEG signals. Overall, this work highlights the value of interpretable machine learning for analyzing subject-specific EEG signal patterns in a controlled experimental setting, with potential applications in personalized and asynchronous brain-machine interface systems.", "date": "2026-05-18", "categories": []}
{"id": "2512.04745", "title": "Neural Policy Composition from Free Energy Minimization", "abstract": "The ability to flexibly compose previously acquired skills to execute intelligent behaviors is a hallmark of natural intelligence. Such compositional flexibility is often attributed to context-dependent gating mechanisms that determine how multiple policies or behavioral primitives are combined. Yet, despite remarkable efforts, the normative objective from which such gating rules should arise, and the neural computations capable of implementing them, remain unclear. Existing approaches typically rely on prespecified design choices for the gating rules, and remain tied to specific architectures, learning paradigms, or datasets. Here, we introduce a normative framework in which policy composition emerges from the minimization of a variational free energy, providing a principled and broadly applicable objective for gating. Based on this framework, we derive a continuous-time gradient flow whose trajectories are guaranteed to converge, with explicit rate, to the optimal composition of primitives. We further show that this dynamics admits a mechanistic neural implementation as a soft-competitive recurrent circuit with context-sensitive local interactions. We evaluate the model on emerging flocking behaviors in multi-agent systems, human decision-making in bandit tasks, and control benchmarks in layered architectures. Across these settings, the model provides interpretable mechanistic accounts of policy composition, reproduces key behavioral signatures, yields insights into data, and matches or outperforms established models.", "date": "2025-12-04", "categories": []}
{"id": "2602.23410", "title": "Brain-OF: An Omnifunctional Foundation Model for fMRI, EEG and MEG", "abstract": "Brain foundation models have achieved remarkable advances across a wide range of neuroscience tasks. However, most existing models are limited to a single functional modality, restricting their ability to exploit complementary spatiotemporal dynamics and the collective data scale across different neuroimaging techniques. This limitation largely arises from severe semantic heterogeneity and resolution discrepancies among modalities. To address these challenges, we propose Brain-OF, an omnifunctional brain foundation model jointly pretrained on fMRI, EEG and MEG, capable of handling both unimodal and multimodal inputs within a unified framework. To reconcile heterogeneous spatiotemporal resolutions, we introduce the Any-Resolution Neural Signal Sampler, which projects diverse brain signals into a shared semantic space. To further manage semantic shifts, the Brain-OF backbone integrates DINT attention with a Sparse Mixture of Experts, where shared experts capture modality-invariant representations and routed experts specialize in modality-specific semantics. Furthermore, to explicitly internalize the characteristics of neural activity through self-supervised learning, we propose Masked Temporal-Frequency Modeling, a dual-domain pretraining objective that jointly reconstructs brain signals in both the time and frequency domains. Brain-OF is pretrained on a large-scale corpus comprising around 40 datasets and demonstrates superior performance across diverse downstream tasks, highlighting the benefits of joint multimodal integration and dual-domain pretraining.", "date": "2026-02-26", "categories": []}
{"id": "2605.11885", "title": "From Clever Hans to Scientific Discovery: Interpreting EEG Foundational Transformers with LRP", "abstract": "Emerging foundation models (FMs) in electroencephalography (EEG) promise a path to scale deep learning in diagnostics and brain-computer interfaces despite data scarcity, yet their opaque nature remains a barrier to wider adoption. We investigate attention-aware Layer-wise relevance propagation (LRP) as a post-hoc attribution method for EEG-FMs, extending LRP's use on convolutional neural network (CNN)-based EEG models to the Transformer architectures that current FMs are based on. We find that LRP can both verify EEG-FM decisions and surface novel, biologically plausible hypotheses from them. In motor imagery, it unmasks 'Clever Hans' behavior where models prioritize task correlated ocular signals over the intended motor correlates. In a naturalistic paradigm for affect prediction, it reveals a recurring reliance on a central electrode cluster, suggesting a candidate sensorimotor signature of arousal. Though heatmap interpretation remains ambiguous in this complex domain, the results position LRP as a tool for both verification and exploration of EEG-FMs, a role that will grow in both importance and discovery potential as the underlying models mature.", "date": "2026-05-12", "categories": []}
{"id": "2605.15733", "title": "Structure Abstraction and Generalization in a Hippocampal-Entorhinal Inspired World Model", "abstract": "Humans abstract experiences into structured representations to facilitate pattern inference and knowledge transfer. While the hippocampal-entorhinal (HPC-MEC) circuit is known to represent both spatial and conceptual spaces, the mechanisms for concurrently extracting abstract structures from continuous, high-dimensional dynamics remain poorly understood. We propose a brain-inspired hierarchical model that simultaneously infers latent transitions and constructs a predictive visual world model. Our architecture employs an inverse model for structural extraction alongside an HPC-MEC coupling model that dissociates relational structures (MEC) from integrated episodic scenes (HPC). Using primitive transformation dynamics as a benchmark, we demonstrate the model's capacity for structural abstraction. By leveraging velocity-driven path integration, the framework enables robust prediction and structural reuse across diverse contexts, thereby achieving structural generalization. This work provides a novel computational framework for understanding how brain-inspired, self-supervised learning of world models facilitates the acquisition of reusable abstract knowledge.", "date": "2026-05-15", "categories": []}
{"id": "2510.16196", "title": "Seeing Through the Brain: New Insights from Decoding Visual Stimuli with fMRI", "abstract": "Understanding how the brain encodes visual information is a central challenge in neuroscience and machine learning. A promising approach is to reconstruct visual stimuli, essentially images, from functional Magnetic Resonance Imaging (fMRI) signals. This involves two stages: transforming fMRI signals into a latent space and then using a pretrained generative model to reconstruct images. The reconstruction quality depends on how similar the latent space is to the structure of neural activity and how well the generative model produces images from that space. Yet, it remains unclear which type of latent space best supports this transformation and how it should be organized to represent visual stimuli effectively. We present two key findings. First, fMRI signals are more similar to the text space of a language model than to either a vision based space or a joint text image space. Second, text representations and the generative model should be adapted to capture the compositional nature of visual stimuli, including objects, their detailed attributes, and relationships. Building on these insights, we propose PRISM, a model that Projects fMRI sIgnals into a Structured text space as an interMediate representation for visual stimuli reconstruction. It includes an object centric diffusion module that generates images by composing individual objects to reduce object detection errors, and an attribute relationship search module that automatically identifies key attributes and relationships that best align with the neural activity. Extensive experiments on real world datasets demonstrate that our framework outperforms existing methods, achieving up to an 8% reduction in perceptual loss. These results highlight the importance of using structured text as the intermediate space to bridge fMRI signals and image reconstruction.", "date": "2025-10-17", "categories": []}
{"id": "2605.10947", "title": "Interpretable EEG Microstate Discovery via Variational Deep Embedding: A Systematic Architecture Search with Multi-Quadrant Evaluation", "abstract": "EEG microstate analysis segments continuous brain electrical activity into brief, quasi-stable topographic configurations that reflect discrete functional brain states. Conventional approaches such as Modified K-Means operate directly in electrode space with hard assignment, offering no learned latent representation, no generative decoder, and no mechanism to decode latent configurations into verifiable scalp topographies, limiting both model transparency and interpretability. To address this, we present a Convolutional Variational Deep Embedding (Conv-VaDE) model that jointly learns topographic reconstruction and probabilistic soft clustering in a shared latent space. Conv-VaDE enables generative decoding of cluster prototypes into verifiable scalp topographies, replacing opaque hard partitioning with probabilistic soft assignment. A polarity invariance scheme and a four-dimensional grid search over cluster count (K from 3 to 20), latent dimensionality, network depth, and channel width are conducted to systematically reveal how each architectural design choice shapes the quality, stability, and interpretability of learned EEG microstate representations. The model is evaluated on the LEMON resting-state eyes-closed EEG dataset with ten participants using topographic template formation, clustering stability, and global explained variance (GEV). The architecture search reveals that depth L = 4 appears consistently across all 18 best-performing configurations, yielding a best-case GEV of 0.730 and a silhouette of 0.229 at K = 4 across the model sweeps, where moderately deep networks with compact channel widths and small latent dimensionality dominate across the full K range. These results establish that principled architecture search, rather than model scale, is the key to interpretable and stable EEG microstate discovery via variational deep embedding.", "date": "2026-04-29", "categories": []}
{"id": "2605.11410", "title": "What Do EEG Foundation Models Capture from Human Brain Signals?", "abstract": "Clinical electroencephalogram (EEG) analysis rests on a hand-crafted feature catalog refined over decades, \\emph{e.g.,} band power, connectivity, complexity, and more. Modern EEG foundation models bypass this catalog, learn directly from raw signals via self-supervised pretraining, and match or outperform feature-engineered baselines on most clinical benchmarks. Whether the two representations align is an open question, which we decompose into three sub-questions: \\emph{what does the model learn}, \\emph{what does the model use}, and \\emph{how much can be explained}. We answer them with layer-wise ridge probing, LEACE-style cross-covariance subspace erasure, and a transparent classifier benchmarked against a random-feature baseline. The audit covers three foundation models (CSBrain, CBraMod, LaBraM), five clinical tasks (MDD, Stress, ISRUC-Sleep, TUSL, Siena), and a 6-family 63-feature lexicon. Of the $945$ (model, task, feature) units, $648$ ($68.6\\%$) are representation-causal and $199$ ($21.1\\%$) are encoded-only. Across tasks, $50$ features qualify as universal candidates with strong support (all three architectures RC) in two or more tasks. Frequency-domain features dominate, but the other five families each contribute substantial causal mass. Confirmed features recover, on average, $79.3\\%$ of the foundation model's advantage over the random baseline, with a clean task gradient (MDD $\\approx 0.99$ down to Stress $\\approx 0.56$): tasks near ceiling are almost fully recovered by the lexicon, while harder tasks leave a non-trivial residual that pinpoints a concrete target for future concept discovery.", "date": "2026-05-12", "categories": []}
{"id": "2605.13904", "title": "Feature Visualization Recovers Known Cortical Selectivity from TRIBE v2", "abstract": "Brain encoder models predict cortical fMRI responses from the internal activations of pretrained vision and language networks, and are typically evaluated by held-out prediction accuracy. This is a useful signal for training but a poor one for interpretation: it tells us an encoder fits the data without telling us whether it has internalized the functional organization of the brain. We propose feature visualization -- gradient ascent on the encoder's predicted activation for a target region of interest (ROI) -- as a complementary interpretability technique, and apply it to TRIBE v2 composed with V-JEPA 2 (ViT-G, 40 layers), holding both frozen and synthesizing still images for seven regions spanning the ventral and dorsal visual hierarchies. Under identical hyperparameters, the probe recovers a visible progression of increasing spatial scale and feature complexity across V1 to V4, matching the ventral-stream hierarchy. It also produces three distinctive downstream regimes: radial \"frozen-motion\" streaks for the middle temporal area (MT) despite static-only optimization, face-like features for the fusiform face area (FFA), and consistent rectilinear line patterns for the parahippocampal place area (PPA). Optimized FFA stimuli drive the predicted region ~4x as much as a natural face photograph, consistent with feature visualization producing adversarial super-stimuli rather than canonical exemplars. The probe is simple, differentiable, and applicable to any brain encoder with a differentiable backbone, allowing for qualitative evaluation of brain encoders.", "date": "2026-05-13", "categories": []}
{"id": "2605.14025", "title": "Do Language Models Align with Brains? Prediction Scores Are Not Enough", "abstract": "Brain-language model comparisons often interpret neural prediction scores as evidence that model representations capture brain-relevant language computation. We asked whether language models align with brains, and whether prediction scores are enough to support that claim, using L-PACT, a source-audited framework that evaluates predictive, relational, mechanism-stripping, and reliability-bounded evidence. Across primary naturalistic language neural datasets and derived language-model representations, L-PACT compared real model features with nuisance baselines and severe controls, tested whether model-to-brain profiles reproduced brain-to-brain patterns, recomputed held-out scores after mechanism stripping, and normalized evidence against brain-brain ceilings. The locked analysis set contains 414 predictive-control rows, 2304 relational profile rows, 4320 mechanism-stripping rows, 420 brain-brain ceiling rows, and 146 integrated decision rows. Assay-sensitivity checks showed that brain-brain reliability, brain-as-model run-to-run relational profiles, independent low-level neural and WAV-derived acoustic-envelope gates, and a deterministic implanted-signal simulation can produce positive evidence when expected. Nevertheless, no real model row passed the predictive, relational, mechanism-stripping, or operational Turing-bounded reliability gates; all 146 integrated rows were control-explained. Less stringent single-criterion rules would have counted raw positive predictive, relational, stripping-delta, and ceiling-normalized effects, but L-PACT downgraded them because controls explained the apparent evidence. In the analyzed derived artifact set, the tested language-model representations do not satisfy L-PACT alignment gates; apparent positives are converted into an auditable control-explained taxonomy rather than treated as structural alignment.", "date": "2026-05-13", "categories": []}
{"id": "2605.14867", "title": "REALM: Retrospective Encoder Alignment for LFP Modeling", "abstract": "Spike activity has been the dominant neural signal for behavior decoding due to its high spatial and temporal resolution. However, as brain-computer interfaces (BCIs) move toward high channel counts and wireless operation, the high sampling frequency of spike signals becomes a bottleneck due to high power and bandwidth requirements. Local field potentials (LFPs) represent a different spatial-temporal scale of brain activity compared to spikes, offering key advantages including improved long-term stability, reduced energy consumption, and lower bandwidth requirement. Despite these benefits, LFP-based decoding models typically show reduced accuracy and often rely on non-causal architectures that are unsuitable for real-time deployment. To address these challenges, we propose REALM: a retrospective distillation framework that enables causal LFP decoding. Inspired by offline-to-online distillation strategies in speech recognition, REALM transfers representational knowledge from a pretrained multi-session bidirectional LFP model to a causal version for real-time deployment. We first pretrain a bidirectional Mamba-2 teacher model using a masked autoencoding objective. We then distill this teacher model into a compact student model via a combined objective of representation alignment and task supervision. REALM consistently outperforms both causal and non-causal LFP-based SOTA methods for behavior decoding. Notably, our REALM improves decoding performance while achieving a $2\\times$ reduction in parameter count and a $10\\times$ reduction in training time. These results demonstrate that retrospective distillation effectively bridges the gap between offline and real-time neural decoding. REALM shows that LFP-only models can achieve competitive decoding performance without reliance on spike signals, offering a practical and scalable alternative for next-generation wireless implantable BCIs.", "date": "2026-05-14", "categories": []}
{"id": "2605.12536", "title": "Information as Maximum-Caliber Deviation: A bridge between Integrated Information Theory and the Free Energy Principle", "abstract": "The Free Energy Principle (FEP) is a leading framework for mathematically modeling self-organization and learning, while Integrated Information Theory (IIT) is a computational ontology of consciousness oriented around irreducible cause and effect. While conceptual unifications have been proposed and appear to be supported by empirical findings, the absence of a rigorous mathematical mapping places upper bounds on their precision and testability. This work proposes that information can be defined as the deviation $\\psi$ of realized dynamics from a constrained maximum-caliber (MaxCal) path ensemble over a finite time horizon. Under this definition, each of the cause/effect repertoires central to IIT 3.0 emerge directly from MaxCal variational principles, allowing IIT's phenomenological calculus to be re-derived from constrained entropy-maximization (CMEP). This framework supplies a theoretical bridge to active inference, which is mathematically dual to CMEP under Langevin dynamics, and offers a principled route for extending IIT to new dynamical regimes. When the approach is applied under the Central Limit Theorem (CLT) for Markov chains and via large deviations theory (LDT) to Ising models, information $\\psi$ is shown to be equivalent to prediction error under accompanying predictive coding models. This may hold relevance to the ``hill-shaped trajectory'' of $\\Phi$ observed in neuronal cultures adapting to sensory inputs. Together, these results provide a physically and mathematically grounded rationale for studying the convergence of FEP, IIT, and thermodynamic frameworks of cognition such as recent work grounding consciousness in violations of the Fluctuation-Dissipation Theorem (FDT).", "date": "2026-05-03", "categories": []}
{"id": "2605.12992", "title": "SpikeProphecy: A Large-Scale Benchmark for Autoregressive Neural Population Forecasting", "abstract": "Neural population models, which predict the joint firing of many simultaneously recorded neurons forward in time, are typically evaluated by a single aggregate Pearson correlation $r$ between predicted and actual spike counts, a number that masks critical structure. We argue that how we evaluate spike forecasting matters as much as what we build, and introduce SpikeProphecy, the first large-scale benchmark for causal, autoregressive spike-count forecasting on real electrophysiology recordings. Our core contribution is a population metric decomposition that separates aggregate performance into temporal fidelity, spatial pattern accuracy, and magnitude-invariant alignment. The decomposition surfaces aspects of the underlying data that an aggregate scalar collapses together. We apply the protocol to 105 Neuropixels sessions (Steinmetz 2019 + IBL Repeated Site; ~89,800 neurons) with seven architecture baselines spanning four structural families: four SSMs (three diagonal and one non-diagonal), a Transformer, an LSTM, and a spiking network. The decomposition surfaces a brain-region predictability ranking that reproduces across all seven baselines and survives ANCOVA correction for firing-statistics constraints (region $\\Delta R^2 = 0.018$ above the firing-statistics covariates). It also exposes a sub-Poisson evaluation floor where rigorous metrics combine with genuine biophysical constraints on regular spike trains, and yields a negative result on KL-on-output-rates distillation for ANN-to-SNN transfer in this Poisson count domain.", "date": "2026-05-13", "categories": []}
{"id": "2605.12999", "title": "Implicit Behavioral Decoding from Next-Step Spike Forecasts at Population Scale", "abstract": "Closed-loop brain-computer interfaces often require both a forecast of upcoming neural population activity and a readout of the animal's behavioral state. A single Mamba forecaster, trained only on next-step spike counts at Neuropixels scale, can deliver both in one forward pass. A lightweight per-session linear head reading the model's predicted rates decodes behavior better than the same linear classifier reading the raw spike counts, under matched temporal context. We test on the Steinmetz visual-discrimination benchmark, which spans 39 sessions, roughly 27,000 neurons, and 1,994 held-out trials. Across three training seeds, Mamba's predicted rates decode mouse choice at 75.7$\\pm$0.2% trial vote, roughly 2.3 times chance level, and stimulus side at 66.1$\\pm$0.6%, about twice chance. Compared to a matched 500 ms-context linear decoder on the raw spike counts, Mamba wins at trial vote by 4-6 pp on response and 4-6 pp on stimulus side. A session-start calibration block of about 100-150 trials brings the readout within 1-2 pp of asymptote, and the full pipeline fits inside the 50 ms bin budget on workstation-class GPUs typical of tethered chronic Neuropixels recordings.", "date": "2026-05-13", "categories": []}
{"id": "2605.13133", "title": "KAST-BAR: Knowledge-Anchored Semantically-Dynamic Topology Brain Autoregressive Modeling for Universal Neural Interpretation", "abstract": "While EEG foundation models have shown significant potential in universal neural decoding across tasks, their advancement remains constrained by the inadequacy modeling of complex spatiotemporal topology, as well as the inherent modality gap between low-level physiological signals and high-level textual semantics. To address these challenges, we propose a Knowledge-Anchored Semantically-Dynamic Topology Brain Autoregressive Model (KAST-BAR), which dynamically aligns physiological representations derived from multi-level brain topology with an expert-level semantic space. Specifically, we design a Dual-Stream Hierarchical Attention (DSHA) encoder that accurately captures the brain's intrinsic non-Euclidean topology by modeling local temporal dynamics with global spatial contexts. On this basis, a Knowledge-Anchored Semantic Profiler (KASP) is proposed to synthesize physically-grounded and instance-level textual profiles, which subsequently drive a Semantic Text-Aware Refiner (STAR) to dynamically reconstruct EEG representations using Latent Expert Queries. By conducting large-scale pre-training on 21 diverse datasets to build a foundation model, KAST-BAR effectively integrates expert-level medical knowledge into EEG signal representations, consistently achieving superior performance across six downstream tasks. Our code is available at https://github.com/KAST-BAR/KAST-BAR", "date": "2026-05-13", "categories": []}
{"id": "2605.13312", "title": "Supervised Deep Multimodal Matrix Factorization for Interpretable Brain Network Analysis", "abstract": "We present Supervised Deep Multimodal Matrix Factorization (SD3MF), an interpretable framework for integrative brain network analysis that generalizes Symmetric Nonnegative Matrix Tri-Factorization (SNMTF) from unsupervised single-graph clustering to supervised prediction over populations of multimodal graphs. SD3MF learns deep hierarchical factorizations for each modality together with a shared latent representation that aligns subjects across views. An encoder-decoder formulation jointly optimizes graph reconstruction and supervised prediction, while adaptive weights enable data-driven multimodal fusion. By representing each subject through community-level interaction matrices, the model yields interpretable and discriminative features. Experiments on multimodal connectome datasets show that SD3MF consistently outperforms strong deep learning baselines such as CNNs and GNNs, while enabling biologically interpretable insights. Code for reproducibility is available at: https://github.com/amjadseyedi/SD3MF.", "date": "2026-05-13", "categories": []}
{"id": "2605.11718", "title": "Self-organized MT Direction Maps Emerge from Spatiotemporal Contrastive Optimization", "abstract": "The spatial and functional organization of the primate visual cortex is a fundamental problem in neuroscience. While recent computational frameworks like the Topographic Deep Artificial Neural Network (TDANN) have successfully modeled spatial organization in the ventral stream, the computational origins of the dorsal stream's distinct topographies, such as direction-selective maps in the middle temporal (MT) area, remain largely unresolved. In this work, we present a spatiotemporal TDANN to investigate whether MT topography is governed by the same universal principles. By training a 3D ResNet on naturalistic videos via a Momentum Contrast (MoCo) self-supervised paradigm alongside a biologically inspired spatial loss, we demonstrate the spontaneous emergence of brain-like direction maps and topological pinwheel structures. Crucially, we reveal that MT tuning properties, characterized by strong direction selectivity paired with a residual axial component, arise from a strict optimization trade-off between task-driven discriminative pressure and spatial regularization. The model's representations quantitatively match in vivo macaque MT physiological baselines, including direction selectivity index, circular variance, and pinwheel density. These findings unify the computational origins of the ventral and dorsal streams, establishing a general mechanism for cortical self-organization.", "date": "2026-05-12", "categories": []}
{"id": "2605.12049", "title": "Scaling Laws and Tradeoffs in Recurrent Networks of Expressive Neurons", "abstract": "Cortical neurons are complex, multi-timescale processors wired into recurrent circuits, shaped by long evolutionary pressure under stringent biological constraints. Mainstream machine learning, by contrast, predominantly builds models from extremely simple units, a default inherited from early neural-network theory. We treat this as a normative architectural question. How should one split a fixed parameter budget $P$ between the number of units $N$, per-unit effective complexity $k_e$, and per-unit connectivity $k_c$? What controls the optimal allocation? This calls for a model in which per-unit complexity can be tuned independently of width and connectivity. Accordingly, we introduce the ELM Network, whose recurrent layer is built from Expressive Leaky Memory (ELM) neurons, chosen to mirror functional components of cortical neurons. The architecture allows for individually adjusting $N$, $k_e$, and $k_c$ and trains stably across orders of magnitude in scale. We evaluate the model on two qualitatively different sequence benchmarks: the neuromorphic SHD-Adding task and Enwik8 character-level language modeling. Performance improves monotonically along each of the three axes individually. Under a fixed budget, a clear non-trivial optimum emerges in their tradeoff, and larger budgets favor both more and more complex neurons. A closed-form information-theoretic model captures these tradeoffs and attributes the diminishing returns at two ends to: per-neuron signal-to-noise saturation and across-neuron redundancy. A hyperparameter sweep spanning three orders of magnitude in trainable parameters traces a near-Pareto-frontier scaling law consistent with the framework. This suggests that the simple-unit default in ML is not obviously optimal once this tradeoff surface is probed, and offers a normative lens on cortex's reliance on complex spatio-temporal integrators.", "date": "2026-05-12", "categories": []}
{"id": "2509.21671", "title": "Neuroprobe: Evaluating Intracranial Brain Responses to Naturalistic Stimuli", "abstract": "High-resolution neural datasets enable foundation models for the next generation of brain-computer interfaces and neurological treatments. The community requires rigorous benchmarks to discriminate between competing modeling approaches, yet no standardized evaluation frameworks exist for intracranial EEG (iEEG) recordings. To address this gap, we present Neuroprobe: a suite of decoding tasks for studying multi-modal language processing in the brain. Unlike scalp EEG, intracranial EEG requires invasive surgery to implant electrodes that record neural activity directly from the brain with minimal signal distortion. Neuroprobe is built on the BrainTreebank dataset, which consists of over 40 hours of iEEG recordings from 10 human subjects performing a naturalistic movie viewing task. Neuroprobe serves two critical functions. First, it is a source from which neuroscience insights can be drawn. The high temporal and spatial resolution of the labeled iEEG allows researchers to systematically determine when and where computations for each aspect of language processing occur in the brain by measuring the decodability of each feature across time and all electrode locations. Using Neuroprobe, we visualize how information flows from key language and audio processing sites in the superior temporal gyrus to sites in the prefrontal cortex. We also demonstrate the time evolution of processing from simple auditory features (e.g., pitch and volume) to more complex language features (e.g., part of speech) in a purely data-driven manner. Second, as the field moves toward neural foundation models trained on large-scale datasets, Neuroprobe provides a rigorous framework for comparing competing architectures and training protocols. We make the code for Neuroprobe openly available, aiming to enable rapid progress in the field of iEEG foundation models. Public leaderboard: https://neuroprobe.dev/", "date": "2025-09-25", "categories": []}
{"id": "2605.08495", "title": "NeuralBench: A Unifying Framework to Benchmark NeuroAI Models", "abstract": "Deep learning and large public datasets have recently catalyzed the proliferation of AI models for processing brain recordings. However, systematically evaluating these models remains a challenge: not only do the preprocessing pipelines, training and finetuning approaches largely vary across studies, but their downstream evaluation is often limited to small sets of tasks and/or datasets. Here, we present NeuralBench: a unified framework for benchmarking AI models of brain activity. We accompany this framework with NeuralBench-EEG v1.0 -- a large EEG benchmark that includes 36 electroencephalography (EEG) tasks and 14 deep learning architectures, and is evaluated on 94 datasets accessed through a standardized interface. This first EEG-focused release already highlights two main findings. First, current foundation models only marginally outperform task-specific models. Second, a large set of tasks (e.g. cognitive decoding, clinical predictions) remain highly challenging, even for the best models. Critically, NeuralBench is designed for the integration of new tasks, datasets, models, and neuroimaging modalities, as illustrated by preliminary extensions to MEG and fMRI datasets and models. Through this white paper, we invite the community to expand this open-source framework and work together toward a unified benchmarking standard for neuroimaging models.", "date": "2026-05-08", "categories": []}
{"id": "2605.09392", "title": "HyNeuralMap: Hyperbolic Mapping of Visual Semantics to Neural Hierarchies", "abstract": "Understanding the intricate mappings between visual stimuli and neural responses is a fundamental challenge in cognitive neuroscience. While current approaches predominantly align images and functional magnetic resonance imaging (fMRI) responses in Euclidean space, this geometry often struggles to preserve fine-grained semantic relationships and latent hierarchical structures across visual and neural modalities. To overcome this, we propose HyNeuralMap, a framework that employ hyperbolic Lorentz model to map visual semantics into a shared, cross-subject neural hierarchy. By leveraging the negative curvature of hyperbolic space as an inductive bias, the proposed framework better captures hierarchical semantic organization and cross-subject neural similarities. Specifically, visual and neural embeddings are jointly optimized through hyperbolic geometric alignment, where geodesic distances preserve semantic proximity and hierarchical relationships more effectively than Euclidean embeddings. Experiments demonstrate that HyNeuralMap consistently outperforms state-of-the-art Euclidean baselines in both multi-label semantic prediction and cross-modal retrieval tasks. This confirms hyperbolic geometry's superiority for cross-modal semantic alignment and hierarchical modeling, providing a new avenue for vision-neural representation learning.", "date": "2026-05-10", "categories": []}
{"id": "2605.10688", "title": "DANCE: Detect and Classify Events in EEG", "abstract": "Event identification in continuous neural recordings is a critical task in neuroscience. Decoding in EEG is dominated by classifying windows aligned to known event onsets. However, while available in controlled experiments, such onsets are absent in continuous real-world monitoring. Here, we introduce DANCE, a deep learning pipeline that frames neural decoding as a set-prediction problem and jointly detects and classifies events directly from raw, unaligned signals. Evaluated separately on ten datasets curated from the literature with a wide variety of event types (ranging from milliseconds to minutes in duration), our model outperforms existing methods on a broad range of cognitive, clinical and BCI tasks. This single architecture establishes a new state of the art in the competitive task of seizure monitoring and matches the accuracy of onset-informed models for BCI tasks. Overall, our method marks a step towards end-to-end asynchronous neural decoding models", "date": "2026-05-11", "categories": []}
{"id": "2510.18516", "title": "Decoding Dynamic Visual Experience from Calcium Imaging via Cell-Pattern-Aware Pretraining", "abstract": "Neural recordings exhibit a distinctive form of heterogeneity rooted in differences in cell types, intrinsic circuit dynamics, and stochastic stimulus-response variability that goes beyond ordinary dataset variability, mixing statistically regular neurons with highly stochastic, stimulus-contingent ones within the same dataset. This heterogeneity poses a challenge for self-supervised learning (SSL) -- learnable statistical regularity -- thereby destabilizing representation learning and limiting reliable scaling. We introduce POYO-CAP (Cell-pattern Aware Pretraining), a biologically grounded hybrid pretraining strategy that first trains with masked reconstruction plus lightweight auxiliary supervision on statistically regular neurons -- identified via skewness and kurtosis -- and then fine-tunes on more stochastic populations. On the Allen Brain Observatory dataset, this curriculum yields 12--13\\% relative improvements over from-scratch training and enables smooth, monotonic scaling with model size, whereas baselines trained on mixed populations plateau or destabilize. By making statistical predictability an explicit data-selection criterion, POYO-CAP turns neural heterogeneity into a scalable learning advantage for robust neural decoding.", "date": "2025-10-21", "categories": []}
{"id": "2605.03222", "title": "Beyond Activation Alignment: The Geometry of Neural Sensitivity", "abstract": "Activation-alignment measures such as Representational Similarity Analysis (RSA), Canonical Correlation Analysis (CCA), and Centered Kernel Alignment (CKA) are widely used to compare biological and artificial neural representations. Recent theoretical work interprets many of these methods as assessing agreement between optimal linear readouts over broad families of global tasks. However, agreement at the level of global readouts does not determine how a system uses local stimulus evidence. Specifically, representations may align in activation space yet differ in their sensitivity to small perturbations. To address this challenge, we introduce a complementary framework based on local decodable information, which focuses on a representation's ability, under noise, to discriminate small perturbations within a specified stimulus-coordinate subspace. Building on Fisher information and local representation geometry, we summarize each representation using the expected projected pullback/Fisher metric over that subspace. This formulation induces a second-moment family of local discrimination tasks, for which the resulting operator provides a minimal, complete dataset-level summary of expected discriminability. We compare these regularized signatures using a log-spectral distance on the manifold of symmetric positive definite (SPD) matrices, yielding the Spectral Riemannian Alignment Score (S-RAS) and a uniform multiplicative certificate over the corresponding family of lifted task values. Empirically, this framework enables the recovery of corresponding layers across independently trained artificial neural networks, supports transferable class-conditional probes, reveals controlled dissociations between standard and robust training, and uncovers stimulus-coordinate family effects across mouse visual cortex using the Allen Brain Observatory static gratings dataset.", "date": "2026-05-04", "categories": []}
{"id": "2605.02936", "title": "A Universal Space of Brain Dynamics for Unveiling Cognitive Transitions and Individual Differences", "abstract": "Representing dynamical systems through data-driven universal spaces has proven effective; however, achieving this universality for human brain activity remains a significant challenge, further aggravated by diverse cognitive states and individual subjects. Recognizing that spatial properties reflect physical wiring while temporal properties reflect brain function, we develop Universal Brain Dynamics (UBD) to construct a universal space tailored to brain activity and quantify corresponding dynamics using a model-derived Jacobian matrix. Crucially, we validate UBD's universality by accurately predicting functional magnetic resonance imaging (fMRI) signals (Pearson's r > 0.9) across eight states and 963 subjects in the Human Connectome Project (HCP). Through evaluating resting-state fMRI represented within UBD, we gain insight into how infra-slow fluctuation (ISF) underpins brain activity. Furthermore, we reveal a new perspective on structure-function coupling (SFC) by analyzing the temporal sequence of brain dynamics. Extending UBD to task-evoked states, we derive brain dynamics across various cognitive conditions, elucidating the neural mechanisms driving cognitive transitions at a finer granularity. For individual differences, we compare brain dynamics across subjects to identify the neural underpinnings of these variations. Our findings suggest that synergistically integrating spatial and temporal properties of brain activity establishes a universal space for its unfolding, enabling the precise numerical analysis of underlying neural mechanisms across varying conditions.", "date": "2026-05-01", "categories": []}
{"id": "2509.09152", "title": "LITcoder: A General-Purpose Library for Building and Comparing Encoding Models", "abstract": "We introduce LITcoder, an open-source library for building and benchmarking neural encoding models. Designed as a flexible backend, LITcoder provides standardized tools for aligning continuous stimuli (e.g., text and speech) with brain data, transforming stimuli into representational features, mapping those features onto brain data, and evaluating the predictive performance of the resulting model on held-out data. The library implements a modular pipeline covering a wide array of methodological design choices, so researchers can easily compose, compare, and extend encoding models without reinventing core infrastructure. Such choices include brain datasets, brain regions, stimulus feature (both neural-net-based and control, such as word rate), downsampling approaches, and many others. In addition, the library provides built-in logging, plotting, and seamless integration with experiment tracking platforms such as Weights & Biases (W&B). We demonstrate the scalability and versatility of our framework by fitting a range of encoding models to three story listening datasets: LeBel et al. (2023), Narratives, and Little Prince. We also explore the methodological choices critical for building encoding models for continuous fMRI data, illustrating the importance of accounting for all tokens in a TR scan (as opposed to just taking the last one, even when contextualized), incorporating hemodynamic lag effects, using train-test splits that minimize information leakage, and accounting for head motion effects on encoding model predictivity. Overall, LITcoder lowers technical barriers to encoding model implementation, facilitates systematic comparisons across models and datasets, fosters methodological rigor, and accelerates the development of high-quality high-performance predictive models of brain activity.\n  Project page: https://litcoder-brain.github.io", "date": "2025-09-11", "categories": []}
{"id": "2605.01074", "title": "Benchmarking local Hebbian learning rules for memory storage and prototype extraction", "abstract": "Associative memory or content-addressable memory is an important component function in computer science and information processing, and at the same time a key concept in cognitive and computational brain science. Many different neural network architectures and learning rules have been proposed to model the brain's associative memory while investigating key component functions like figure-ground segmentation, perceptual reconstruction and rivalry. A less investigated but equally important capability of associative memory is prototype extraction where the training set comprises distorted prototype instances and the task is to recall the correct generating prototype given a new distorted instance. In this paper we benchmark associative memory function of seven different Hebbian learning rules employed in non-modular and modular recurrent networks with winner-take-all dynamics operating on moderately sparse binary patterns. We measure pattern storage and weight information capacity, prototype extraction capabilities, and sensitivity to correlations in data. The original additive Hebb rule comes out with worst capacity, covariance learning proves to be robust but with moderate capacity, and the Bayesian-Hebbian learning rules show highest capacity in almost all different conditions tested.", "date": "2026-05-01", "categories": []}
{"id": "2507.01946", "title": "Characterizing control between interacting subsystems with deep Jacobian estimation", "abstract": "Biological function arises through the dynamical interactions of multiple subsystems, including those between brain areas, within gene regulatory networks, and more. A common approach to understanding these systems is to model the dynamics of each subsystem and characterize communication between them. An alternative approach is through the lens of control theory: how the subsystems control one another. This approach involves inferring the directionality, strength, and contextual modulation of control between subsystems. However, methods for understanding subsystem control are typically linear and cannot adequately describe the rich contextual effects enabled by nonlinear complex systems. To bridge this gap, we devise a data-driven nonlinear control-theoretic framework to characterize subsystem interactions via the Jacobian of the dynamics. We address the challenge of learning Jacobians from time-series data by proposing the JacobianODE, a deep learning method that leverages properties of the Jacobian to directly estimate it for arbitrary dynamical systems from data alone. We show that JacobianODEs outperform existing Jacobian estimation methods on challenging systems, including high-dimensional chaos. Applying our approach to a multi-area recurrent neural network (RNN) trained on a working memory selection task, we show that the \"sensory\" area gains greater control over the \"cognitive\" area over learning. Furthermore, we leverage the JacobianODE to directly control the trained RNN, enabling precise manipulation of its behavior. Our work lays the foundation for a theoretically grounded and data-driven understanding of interactions among biological subsystems.", "date": "2025-07-02", "categories": []}
{"id": "2512.15891", "title": "Dynamical Mechanisms for Coordinating Long-term Working Memory Based on the Precision of Spike-timing in Cortical Neurons", "abstract": "In the last century, most sensorimotor studies of cortical neurons relied on average firing rates. Rate coding is efficient for fast sensorimotor processing that occurs within a few seconds. Much less is known about the neural mechanisms underlying long-term working memory with a time scale of hours. Cognitive states may not have sensory or motor correlates. For example, you can sit in a quiet room making plans without moving or sensory processing. You can also make plans while out walking. In this perspective, I make the case for a possible second tier of neural activity that coexists with the well-established sensorimotor tier. The prominent physiological feature of the second tier is coordinated spike timing activity. The interplay of data supporting this hypothesis involves three puzzling yet highly intriguing experimental observations, without any obvious indication that they might actually represent different aspects of a single functional organization. First, consider the precision of spiking in individual neurons. The discovery of millisecond-precision spike initiation in cortical neurons was unexpected (Mainen and Sejnowski, 1995). Even more striking was the precision of spiking in vivo, in response to rapidly fluctuating sensory inputs. Second, high temporal resolution can also mediate spike timing-dependent plasticity (STDP) by controlling the relative timing of presynaptic and postsynaptic spikes at the millisecond scale. Third, we observe waves across many frequency bands traveling across the cortex. Strikingly, their timing is highly precise. Gamma waves, for example, which are triggered by attention, can plausibly trigger STDP that lasts for hours in cortical neurons. This temporary cortical network, ostensibly a second tier of functionality, rides astride the long-term sensorimotor network and could support cognitive processing and long-term working memory.", "date": "2025-12-17", "categories": []}
{"id": "2604.16875", "title": "Untrained CNNs Match Backpropagation at V1: A Systematic RSA Comparison of Four Learning Rules Against Human fMRI", "abstract": "A central question in computational neuroscience is whether the learning rule used to train a neural network determines how well its internal representations align with those of the human visual cortex. We present a systematic comparison of four learning rules (backpropagation (BP), feedback alignment (FA), predictive coding (PC), and spike-timing-dependent plasticity (STDP)) applied to identical convolutional architectures and evaluated against human fMRI data from the THINGS-fMRI dataset (720 stimuli, 3 subjects) using Representational Similarity Analysis (RSA). All models process stimuli at 224 x 224 resolution; results are averaged across 5 random seeds. Crucially, we include an untrained random-weights baseline that reveals the dominant role of architecture. At V1/V2, the untrained baseline exceeds backpropagation (rho = 0.076 vs. rho = 0.034; Delta-rho = +0.044, p < 0.001), and STDP achieves the highest V1 alignment among trained rules (rho = 0.064). At LOC, only BP reliably exceeds the random baseline (rho = 0.012 vs. -0.005, p < 0.001). At IT, all five conditions converge (rho = 0.008-0.014) with no significant pairwise differences among trained rules (p > 0.05, FDR-corrected). FA consistently produces the lowest alignment at V1, V2, and LOC (rho = 0.012 at V1, below all other conditions). Partial RSA confirms all effects survive pixel-similarity control. Seed variability is small relative to between-rule differences at V1/V2. These results demonstrate that early visual alignment is architecture-driven, learning rules differentiate only at intermediate areas, and all rules converge at the highest levels of the hierarchy.", "date": "2026-04-18", "categories": []}
{"id": "2604.23865", "title": "Inverting Foundation Models of Brain Function with Simulation-Based Inference", "abstract": "Foundation models of brain activity promise a new frontier for in silico neuroscience by emulating neural responses to complex stimuli across tasks and modalities. A natural next step is to ask whether these models can also be used in reverse. Can we recover a stimulus or its properties from synthetic brain activity? We study this question in a proof-of-concept setting using TRIBEv2. We pair the brain emulator with large language models (LLMs) that generate news headlines from linguistic parameters such as valence, arousal, and dominance. We then use simulation-based inference to learn a probabilistic mapping from brain maps to latent stimulus parameters. Our results show that these parameters can be recovered from predicted brain maps, validating the quality of neural encodings. They also show that LLMs can serve as controllable stimulus generators for simulated experiments. Together, these findings provide a step toward decoding and inverse design with foundation brain models.", "date": "2026-04-26", "categories": []}
{"id": "2604.26218", "title": "ViBE: Visual-to-M/EEG Brain Encoding via Spatio-Temporal VAE and Distribution-Aligned Projection", "abstract": "Brain encoding models not only serve to decipher how visual stimuli are transformed into neural responses, but also represent a critical step toward visual prostheses that restore vision for patients with severe vision disorders. Brain encoding involves two fundamental steps: achieving faithful reconstruction of neural responses and establishing cross-modal alignment between visual stimuli and neural responses. To this end, we propose ViBE, a novel brain encoding framework for generating magnetoencephalography (MEG) and electroencephalography (EEG) signals from visual stimuli. Specifically, we first design a spatio-temporal convolutional variational autoencoder (TSC-VAE) that captures the spatio-temporal characteristics of M/EEG signals for effective neural response reconstruction. To bridge the modality gap between visual features and neural representations, we employ Q-Former to map CLIP image embeddings to the TSC-VAE latent space, producing neural proxy embeddings. For comprehensive cross-modal alignment, we combine mean squared error (MSE) loss for point-wise feature matching with sliced Wasserstein distance (SWD) for probability distribution alignment between the neural proxy embeddings and TSC-VAE latent embeddings. We conduct extensive experiments on the THINGS-EEG2 and THINGS-MEG datasets, demonstrating the effectiveness of our approach in generating high-quality M/EEG signals from visual stimuli.", "date": "2026-04-29", "categories": []}
{"id": "2604.26919", "title": "Causal Learning with Neural Assemblies", "abstract": "Can Neural Assemblies -- groups of neurons that fire together and strengthen through co-activation -- learn the direction of causal influence between variables? While established as a computationally general substrate for classification, parsing, and planning, neural assemblies have not yet been shown to internalize causal directionality. We demonstrate that the inherent operations of neural assemblies -- projection, local plasticity control, and sparse winner selection -- are sufficient for directional learning. We introduce DIRECT (DIRectional Edge Coupling/Training), a mechanism that co-activates source and target assemblies under an adaptive gain schedule to internalize directed relations. Unlike backpropagation-based methods, DIRECT relies solely on local plasticity, making the resulting causal claims auditable at the mechanism level. Our findings are verified through a dual-readout validation strategy: (i) synaptic-strength asymmetry, measuring the emergent weight gap between forward and reverse links, and (ii) functional propagation overlap, quantifying the reliability of directional signal flow. Across multiple domains, the framework achieves perfect structural recovery under a supervised, known-structure setting. These results establish neural assemblies as an auditable bridge between biologically plausible dynamics and formal causal models, offering an \"explainable by design\" framework where causal claims are traceable to specific neural winners and synaptic asymmetries.", "date": "2026-04-29", "categories": []}
{"id": "2604.25481", "title": "Emergent Self-Attention from Astrocyte-Gated Associative Memory Dynamics", "abstract": "We introduce a Hopfield-type associative memory in which effective connectivity is multiplicatively modulated by astrocytic gains evolving under an entropy-regularized replicator equation. The coupled neuron-astrocyte dynamics admit a Lyapunov function, ensuring global convergence. At fixed points, astrocytic gains implement a softmax-normalized allocation over pattern similarity scores, yielding a mechanistic realization of self-attention as emergent routing on the gain simplex. In regimes of high memory load and interference, the model significantly improves retrieval accuracy relative to classical Hopfield dynamics and recent neuron-astrocyte baselines. These results establish a dynamical systems framework linking glial modulation, competitive resource allocation, and attention-like computation.", "date": "2026-04-28", "categories": []}
{"id": "2512.03467", "title": "Bayesian Event-Based Model for Disease Subtype and Stage Inference", "abstract": "Chronic diseases often progress differently across patients. Rather than randomly varying, there are typically a small number of subtypes for how a disease progresses across patients. To capture this structured heterogeneity, the Subtype and Stage Inference Event-Based Model (SuStaIn) estimates the number of subtypes, the order of disease progression for each subtype, and assigns each patient to a subtype from primarily cross-sectional data. It has been widely applied to uncover the subtypes of many diseases and inform our understanding of them. But how robust is its performance? In this paper, we develop a principled Bayesian subtype variant of the event-based model (BEBMS) and compare its performance to SuStaIn in a variety of synthetic data experiments with varied levels of model misspecification. BEBMS substantially outperforms SuStaIn across ordering, staging, and subtype assignment tasks. Further, we apply BEBMS and SuStaIn to a real-world Alzheimer's data set. We find BEBMS has results that are more consistent with the scientific consensus of Alzheimer's disease progression than SuStaIn.", "date": "2025-12-03", "categories": []}
{"id": "2604.23903", "title": "Integrative neurocybernetic modeling in the era of large-scale neuroscience", "abstract": "Large-scale neuroscience is generating rich datasets across animals, brain areas and behavioral contexts, yet our modeling efforts remains fragmented across isolated experiments. We argue that understanding behavior requires integrative neurocybernetic models: understandable dynamical models that capture the closed-loop coupling of brain, body and environment, treat the brain as a controller pursuing latent objectives, represent structured variation across scales, and scale to heterogeneous datasets. Such models shift the goal from predicting neural recordings in isolation to inferring the organizing principles that govern neural and behavioral dynamics. We outline a practical route toward this goal by combining nonlinear state-space models and meta-dynamical extensions with scalable inference, knowledge distillation, mixed open- and closed-loop training, and connectomics-informed architectures. By pooling complementary constraints from recordings, behavior, perturbations and anatomy, integrative neurocybernetic models can provide statistical amplification, few-shot generalization, and mechanistic insight into shared dynamical structure, individual variation, and the control objectives that govern behavior. This agenda offers a model-centric path from fragmented data to a mechanistic science of how brains produce behavior.", "date": "2026-04-26", "categories": []}
{"id": "2304.01844", "title": "Grid-SD2E: A General Grid-Feedback in a System for Cognitive Learning", "abstract": "Comprehending how the brain interacts with the external world through generated neural data is crucial for determining its working mechanism, treating brain diseases, and understanding intelligence. Although many theoretical models have been proposed, they have thus far been difficult to integrate and develop. In this study, we were inspired in part by grid cells in creating a more general and robust grid module and constructing an interactive and self-reinforcing cognitive system together with Bayesian reasoning, an approach called space-division and exploration-exploitation with grid-feedback (Grid-SD2E). Here, a grid module can be used as an interaction medium between the outside world and a system, as well as a self-reinforcement medium within the system. The space-division and exploration-exploitation (SD2E) receives the 0/1 signals of a grid through its space-division (SD) module. The system described in this paper is also a theoretical model derived from experiments conducted by other researchers and our experience on neural decoding. Herein, we analyse the rationality of the system based on the existing theories in both neuroscience and cognitive science, and attempt to propose special and general rules to explain the different interactions between people and between people and the external world. What's more, based on this framework, the smallest computing unit is extracted, which is analogous to a single neuron in the brain.", "date": "2023-04-04", "categories": []}
{"id": "2604.16926", "title": "Test-Time Adaptation for EEG Foundation Models: A Systematic Study under Real-World Distribution Shifts", "abstract": "Electroencephalography (EEG) foundation models have shown strong potential for learning generalizable representations from large-scale neural data, yet their clinical deployment is hindered by distribution shifts across clinical settings, devices, and populations. Test-time adaptation (TTA) offers a promising solution by enabling models to adapt to unlabeled target data during inference without access to source data, a valuable property in healthcare settings constrained by privacy regulations and limited labeled data. However, its effectiveness for EEG remains largely underexplored. In this work, we introduce NeuroAdapt-Bench, a systematic benchmark for evaluating test-time adaptation methods on EEG foundation models under realistic distribution shifts. We evaluate representative TTA approaches from other domains across multiple pretrained foundation models, diverse downstream tasks, and heterogeneous datasets spanning in-distribution, out-of-distribution, and extreme modality shifts (e.g., Ear-EEG). Our results show that standard TTA methods yield inconsistent gains and often degrade performance, with gradient-based approaches particularly prone to heavy degradation. In contrast, optimization-free methods demonstrate greater stability and more reliable improvements. These findings highlight the limitations of existing TTA techniques in EEG, provide guidance for future development, and underscore the need for domain-specific adaptation strategies.", "date": "2026-04-18", "categories": []}
{"id": "2604.15363", "title": "Machine learning approaches to uncover the neural mechanisms of motivated behaviour: from ADHD to individual differences in effort and reward sensitivity", "abstract": "Motivated behaviour relies on the brain's capacity to evaluate effort and reward. Dysregulation within these processes contributes to a spectrum of conditions, from hyperactivity in attention-deficit/hyperactivity disorder (ADHD) to diminished goal-directed behaviour in apathy. This thesis investigates the neural mechanisms underlying ADHD using electroencephalography (EEG) and examines individual differences in effort and reward sensitivity using neuroimaging, applying machine learning approaches through three main studies. In Study 1, task-based and resting-state EEG were employed with machine learning models to classify adult individuals with ADHD and healthy controls. Machine learning classifiers trained on task-based EEG during a stop signal task outperformed those trained on resting-state EEG, with the strongest predictive features arising from gamma-band spectral power over fronto-central and parietal regions. In Study 2, diffusion MRI and whole-brain permutation-based analyses identified associations between white matter integrity and computationally modelled parameters reflecting effort and reward sensitivity, with SMA-connected tracts emerging as a central hub. In Study 3, grey matter volumes from structural T1-weighted MRI were used to examine correlates of effort sensitivity, reward sensitivity, and subclinical apathy, with machine learning confirming robust decoding of reward sensitivity and apathy levels. Across studies, fronto-parietal circuits emerged as central to effort valuation and reward processing. These findings may serve as neural biomarkers for improving diagnostic accuracy in ADHD and motivational impairments, and for guiding personalised neurotechnological interventions.", "date": "2026-04-13", "categories": []}
{"id": "2604.12865", "title": "From edges to meaning: Semantic line sketches as a cognitive scaffold for ancient pictograph invention", "abstract": "Humans readily recognize objects from sparse line drawings, a capacity that appears early in development and persists across cultures, suggesting neural rather than purely learned origins. Yet the computational mechanism by which the brain transforms high-level semantic knowledge into low-level visual symbols remains poorly understood. Here we propose that ancient pictographic writing emerged from the brain's intrinsic tendency to compress visual input into stable, boundary-based abstractions. We construct a biologically inspired digital twin of the visual hierarchy that encodes an image into low-level features, generates a contour sketch, and iteratively refines it through top-down feedback guided by semantic representations, mirroring the feedforward and recurrent architecture of the human visual cortex. The resulting symbols bear striking structural resemblance to early pictographs across culturally distant writing systems, including Egyptian hieroglyphs, Chinese oracle bone characters, and proto-cuneiform, and offer candidate interpretations for undeciphered scripts. Our findings support a neuro-computational origin of pictographic writing and establish a framework in which AI can recapitulate the cognitive processes by which humans first externalized perception into symbols.", "date": "2026-04-14", "categories": []}
{"id": "2503.14333", "title": "Characterizing higher-order representations through generative diffusion models explains human decoded neurofeedback performance", "abstract": "Brains construct not only \"first-order\" representations of the environment but also \"higher-order\" representations about those representations -- including higher-order uncertainty estimates that guide learning and adaptive behavior. Higher-order expectations about representational uncertainty -- i.e., learned through experience -- may play a key role in guiding behavior and learning, but their characterization remains empirically and theoretically challenging. Here, we introduce the Noise Estimation through Reinforcement-based Diffusion (NERD) model, a novel computational framework that trains denoising diffusion models via reinforcement learning to infer distributions of noise in functional MRI data from a decoded neurofeedback task, where healthy human participants learn to achieve target neural states. We hypothesize that participants accomplish this task by learning about and then minimizing their own representational uncertainty. We test this hypothesis with NERD, which mirrors brain-like unsupervised learning. Our results show that NERD outperforms backpropagation-trained control models in capturing human performance with explanatory power enhanced by clustering learned noise distributions. Importantly, our results also reveal individual differences in expected-uncertainty representations that predict task success, demonstrating NERD's utility as a powerful tool for probing higher-order neural representations.", "date": "2025-03-18", "categories": []}
{"id": "2405.11597", "title": "Language Reconstruction with Brain Predictive Coding from fMRI Data", "abstract": "Many recent studies have shown that the perception of speech can be decoded from brain signals and subsequently reconstructed as continuous language. However, there is a lack of neurological basis for how the semantic information embedded within brain signals can be used more effectively to guide language reconstruction. Predictive coding theory suggests the human brain naturally engages in continuously predicting future words that span multiple timescales. This implies that the decoding of brain signals could potentially be associated with a predictable future. To explore the predictive coding theory within the context of language reconstruction, this paper proposes \\textsc{PredFT}~(\\textbf{F}MRI-to-\\textbf{T}ext decoding with \\textbf{Pred}ictive coding). \\textsc{PredFT} consists of a main network and a side network. The side network obtains brain predictive representation from related regions of interest~(ROIs) with a self-attention module. The representation is then fused into the main network for continuous language decoding. Experiments on two naturalistic language comprehension fMRI datasets show that \\textsc{PredFT} outperforms current decoding models on several evaluation metrics.", "date": "2024-05-19", "categories": []}
{"id": "2604.09817", "title": "NeuroFlow: Toward Unified Visual Encoding and Decoding from Neural Activity", "abstract": "Visual encoding and decoding models act as gateways to understanding the neural mechanisms underlying human visual perception. Typically, visual encoding models that predict brain activity from stimuli and decoding models that reproduce stimuli from brain activity are treated as distinct tasks, requiring separate models and training procedures. This separation is inefficient and fails to model the consistency between encoding and decoding processes. To address this limitation, we propose NeuroFlow, the first unified framework that jointly models visual encoding and decoding from neural activity within a single flow model. NeuroFlow introduces two key components: (1) NeuroVAE is designed as a variational backbone to model neural variability and establish a compact, semantically structured latent space for bidirectional modeling across visual and neural modalities. (2) Cross-modal Flow Matching (XFM) bypasses the typical paradigm of noise-to-data diffusion guided by a specific modality condition, instead learning a reversibly consistent flow model between visual and neural latent distributions. For the first time, visual encoding and decoding are reformulated as a time-dependent, reversible process within a shared latent space for unified modeling. Empirical results demonstrate that NeuroFlow achieves superior overall performance in visual encoding and decoding tasks with higher computational efficiency compared to any isolated methods. We further analyze principal factors that steer the model toward encoding-decoding consistency and, through brain functional analyses, demonstrate that NeuroFlow captures consistent activation patterns underlying neural variability. NeuroFlow marks a major step toward unified visual encoding and decoding from neural activity, providing mechanistic insights that inform future bidirectional visual brain-computer interfaces.", "date": "2026-04-10", "categories": []}
{"id": "2604.10627", "title": "Computational Lesions in Multilingual Language Models Separate Shared and Language-specific Brain Alignment", "abstract": "How the brain supports language across different languages is a basic question in neuroscience and a useful test for multilingual artificial intelligence. Neuroimaging has identified language-responsive brain regions across languages, but it cannot by itself show whether the underlying processing is shared or language-specific. Here we use six multilingual large language models (LLMs) as controllable systems and create targeted ``computational lesions'' by zeroing small parameter sets that are important across languages or especially important for one language. We then compare intact and lesioned models in predicting functional magnetic resonance imaging (fMRI) responses during 100 minutes of naturalistic story listening in native English, Chinese and French (112 participants). Lesioning a compact shared core reduces whole-brain encoding correlation by 60.32% relative to intact models, whereas language-specific lesions preserve cross-language separation in embedding space but selectively weaken brain predictivity for the matched native language. These results support a shared backbone with embedded specializations and provide a causal framework for studying multilingual brain-model alignment.", "date": "2026-04-12", "categories": []}
{"id": "2510.21588", "title": "Contribution of task-irrelevant stimuli to drift of neural representations", "abstract": "Biological and artificial learners are inherently exposed to a stream of data and experience throughout their lifetimes and must constantly adapt to, learn from, or selectively ignore the ongoing input. Recent findings reveal that, even when the performance remains stable, the underlying neural representations can change gradually over time, a phenomenon known as representational drift. Studying the different sources of data and noise that may contribute to drift is essential for understanding lifelong learning in neural systems. However, a systematic study of drift across architectures and learning rules, and the connection to task, are missing. Here, in an online learning setup, we characterize drift as a function of data distribution, and specifically show that the learning noise induced by task-irrelevant stimuli, which the agent learns to ignore in a given context, can create long-term drift in the representation of task-relevant stimuli. Using theory and simulations, we demonstrate this phenomenon both in Hebbian-based learning -- Oja's rule and Similarity Matching -- and in stochastic gradient descent applied to autoencoders and a supervised two-layer network. We consistently observe that the drift rate increases with the variance and the dimension of the data in the task-irrelevant subspace. We further show that this yields different qualitative predictions for the geometry and dimension-dependency of drift than those arising from Gaussian synaptic noise. Overall, our study links the structure of stimuli, task, and learning rule to representational drift and could pave the way for using drift as a signal for uncovering underlying computation in the brain.", "date": "2025-10-24", "categories": []}
{"id": "2604.07361", "title": "BLEG: LLM Functions as Powerful fMRI Graph-Enhancer for Brain Network Analysis", "abstract": "Graph Neural Networks (GNNs) have been widely used in diverse brain network analysis tasks based on preprocessed functional magnetic resonance imaging (fMRI) data. However, their performances are constrained due to high feature sparsity and inherent limitations of domain knowledge within uni-modal neurographs. Meanwhile, large language models (LLMs) have demonstrated powerful representation capabilities. Combining LLMs with GNNs presents a promising direction for brain network analysis. While LLMs and MLLMs have emerged in neuroscience, integration of LLMs with graph-based data remains unexplored. In this work, we deal with these issues by incorporating LLM's powerful representation and generalization capabilities. Considering great cost for directly tuning LLMs, we instead function LLM as enhancer to boost GNN's performance on downstream tasks. Our method, namely BLEG, can be divided into three stages. We firstly prompt LLM to get augmented texts for fMRI graph data, then we design a LLM-LM instruction tuning method to get enhanced textual representations at a relatively lower cost. GNN is trained together for coarsened alignment. Finally we finetune an adapter after GNN for given downstream tasks. Alignment loss between LM and GNN logits is designed to further enhance GNN's representation. Extensive experiments on different datasets confirmed BLEG's superiority.Code can be available at https://github.com/KamonRiderDR/BLEG.", "date": "2026-04-01", "categories": []}
{"id": "2604.09229", "title": "The Fast Lane Hypothesis: Von Economo Neurons Implement a Biological Speed-Accuracy Tradeoff", "abstract": "Von Economo neurons (VENs) are large bipolar projection neurons found exclusively in the anterior cingulate cortex (ACC) and frontal insula of species with complex social cognition, including humans, great apes, and cetaceans. Their selective depletion in frontotemporal dementia (FTD) and altered development in autism implicate them in rapid social decision-making, yet no computational model of VEN function has previously existed. We introduce the Fast Lane Hypothesis: VENs implement a biological speed-accuracy tradeoff (SAT) by providing a sparse, fast projection pathway that enables rapid social decisions at the cost of deliberate processing accuracy. We model VENs as fast leaky integrate-and-fire (LIF) neurons with membrane time constant 5 ms and sparse dendritic fan-in of eight afferents, compared to 20 ms and eighty afferents for standard pyramidal neurons, within a spiking cortical circuit of 2,000 neurons trained on a social discrimination task. Networks are evaluated under three clinically motivated conditions across 10 independent random seeds: typical (2% VENs), autism-like (0.4% VENs), and FTD-like (post-training VEN ablation). All configurations achieve equivalent asymptotic classification accuracy (99.4%), consistent with the prediction that VENs modulate decision speed rather than representational capacity. Temporal analysis confirms that VENs produce median first-spike latencies 4 ms earlier than pyramidal neurons. At a fixed decision threshold, the typical condition is significantly faster than FTD-like (t=-23.31, p<0.0001), while autism-like is intermediate (mean RT=26.91+/-9.01 ms vs. typical 20.70+/-2.02 ms; p=0.078). A preliminary evolutionary analysis shows qualitative correspondence between model-optimal VEN fraction and the primate phylogenetic gradient. To our knowledge, this is the first computational model that asks what a Von Economo neuron actually computes.", "date": "2026-04-10", "categories": []}
{"id": "2604.05042", "title": "Energy-Based Dynamical Models for Neurocomputation, Learning, and Optimization", "abstract": "Recent advances at the intersection of control theory, neuroscience, and machine learning have revealed novel mechanisms by which dynamical systems perform computation. These advances encompass a wide range of conceptual, mathematical, and computational ideas, with applications for model learning and training, memory retrieval, data-driven control, and optimization. This tutorial focuses on neuro-inspired approaches to computation that aim to improve scalability, robustness, and energy efficiency across such tasks, bridging the gap between artificial and biological systems. Particular emphasis is placed on energy-based dynamical models that encode information through gradient flows and energy landscapes. We begin by reviewing classical formulations, such as continuous-time Hopfield networks and Boltzmann machines, and then extend the framework to modern developments. These include dense associative memory models for high-capacity storage, oscillator-based networks for large-scale optimization, and proximal-descent dynamics for composite and constrained reconstruction. The tutorial demonstrates how control-theoretic principles can guide the design of next-generation neurocomputing systems, steering the discussion beyond conventional feedforward and backpropagation-based approaches to artificial intelligence.", "date": "2026-04-06", "categories": []}
{"id": "2601.17641", "title": "RPNT: Robust Pre-trained Neural Transformer -- A Pathway for Generalized Motor Decoding", "abstract": "Brain motor decoding aims to interpret and translate neural activity into behaviors. Decoding models should generalize across variations, such as recordings from different brain sites, experimental sessions, behavior types, and subjects, will be critical for real-world applications. Current decoding models only partially address these challenges. In this work, we develop a pretrained neural transformer model, RPNT - Robust Pretrained Neural Transformer, designed to achieve robust generalization through pretraining, which in turn enables effective finetuning for downstream motor decoding tasks. We achieved the proposed RPNT architecture by systematically investigating which transformer building blocks could be suitable for neural spike activity modeling, since components from models developed for other modalities, such as text and images, do not transfer directly to neural data. The final RPNT architecture incorporates three unique enabling components: 1) Multidimensional rotary positional embedding to aggregate experimental metadata such as site coordinates, session ids and behavior types; 2) Context-based attention mechanism via convolution kernels operating on global attention to learn local temporal structures for handling non-stationarity of neural population activity; 3) Robust self-supervised learning objective with stochastic causal masking strategies and contrastive representations. We pretrained two versions of RPNT on distinct datasets that present significant generalization challenges: a) Multi-session, multi-task, and multi-subject microelectrode benchmark; b) Multi-site recordings using high-density Neuropixel 1.0 probes from many cortical locations. After pretraining, we evaluated RPNT generalization on cross-session, cross-type, cross-subject, and cross-site downstream behavior decoding tasks. Our RPNT consistently outperforms the existing decoding models on these tasks.", "date": "2026-01-25", "categories": []}
{"id": "2604.00036", "title": "When and Where: A Model Hippocampal Network Unifies Formation of Time Cells and Place Cells", "abstract": "Hippocampal place and time cells encode spatial and temporal aspects of experience. Both have the same neural substrate, but have been modeled as having different functions and mechanistic origins, place cells as continuous attractors, and time cells as leaky integrators. Here, we show that both types emerge from two dynamical regimes of a single recurrent network (RNN) modeling hippocampal CA3 as a predictive autoencoder. The network receives simulated, partially occluded ``experience vectors\" containing spatial patterns (location-specific activity sampled during environmental traversal) and/or temporal patterns (correlated activity pairs separated by ``void\" intervals), and is trained to reconstruct missing input. During spatial navigation, the network generates stable attractor-like place fields. But trained on temporally structured inputs, the network produces sequentially broadened fields, recapitulating time cells. By varying spatio-temporal input patterning, we observe hidden units transition smoothly between time cell-like and place cell-like representations. These results suggest a shared origin, but task-driven difference, between place and time cells.", "date": "2026-03-22", "categories": []}
{"id": "2603.28931", "title": "Decoding Functional Networks for Visual Categories via GNNs", "abstract": "Understanding how large-scale brain networks represent visual categories is fundamental to linking perception and cortical organization. Using high-resolution 7T fMRI from the Natural Scenes Dataset, we construct parcel-level functional graphs and train a signed Graph Neural Network that models both positive and negative interactions, with a sparse edge mask and class-specific saliency. The model accurately decodes category-specific functional connectivity states (sports, food, vehicles) and reveals reproducible, biologically meaningful subnetworks along the ventral and dorsal visual pathways. This framework bridges machine learning and neuroscience by extending voxel-level category selectivity to a connectivity-based representation of visual processing.", "date": "2026-03-30", "categories": []}
{"id": "2507.02064", "title": "REMI: Reconstructing Episodic Memory During Internally Driven Path Planning", "abstract": "Grid cells in the medial entorhinal cortex (MEC) and place cells in the hippocampus (HC) both form spatial representations. Grid cells fire in triangular grid patterns, while place cells fire at specific locations and respond to contextual cues. How do these interacting systems support not only spatial encoding but also internally driven path planning, such as navigating to locations recalled from cues? Here, we propose a system-level theory of MEC-HC wiring that explains how grid and place cell patterns could be connected to enable cue-triggered goal retrieval, path planning, and reconstruction of sensory experience along planned routes. We suggest that place cells autoassociate sensory inputs with grid cell patterns, allowing sensory cues to trigger recall of goal-location grid patterns. We show analytically that grid-based planning permits shortcuts through unvisited locations and generalizes local transitions to long-range paths. During planning, intermediate grid states trigger place cell pattern completion, reconstructing sensory experiences along the route. Using a single-layer RNN modeling the HC-MEC loop with a planning subnetwork, we demonstrate these effects in both biologically grounded navigation simulations using RatatouGym and visually realistic navigation tasks using Habitat Sim.", "date": "2025-07-02", "categories": []}
{"id": "2603.16885", "title": "DECODE: Dual-Enhanced Conditioned Diffusion for EEG Forecasting", "abstract": "Forecasting Electroncephalography (EEG) signals during cognitive events remains a fundamental challenge in neuroscience and Brain-Computer Interfaces (BCIs), as existing methods struggle to capture both the stochastic nature of neural dynamics and the semantic context of behavioral tasks. We present the Dual-Enhanced COnditioned Diffusion (DECODE) for EEG, a novel framework that unifies semantic guidance from natural language descriptions with temporal dynamics from historical signals to generate event-specific neural responses. DECODE leverages pre-trained language models to condition the diffusion process on rich textual descriptions of cognitive events, while maintaining temporal coherence through history-based Langevin dynamics. Evaluated on a real-world driving task dataset with five distinct behaviors, DECODE achieves sub-microvolt prediction accuracy (MAE = 0.626 microvolt) over 75 timestep horizons while maintaining well-calibrated uncertainty estimates. Our framework demonstrates that natural language can effectively bridge high-level cognitive descriptions and low-level neural dynamics, opening new possibilities for zero-shot generalization to novel behaviors and interpretable BCIs. By generating physiologically plausible, event-specific EEG trajectories conditioned on semantic descriptions, DECODE establishes a new paradigm for understanding and predicting context-dependent neural activity.", "date": "2026-02-25", "categories": []}
{"id": "2603.17676", "title": "Inhibitory normalization of error signals improves learning in neural circuits", "abstract": "Normalization is a critical operation in neural circuits. In the brain, there is evidence that normalization is implemented via inhibitory interneurons and allows neural populations to adjust to changes in the distribution of their inputs. In artificial neural networks (ANNs), normalization is used to improve learning in tasks that involve complex input distributions. However, it is unclear whether inhibition-mediated normalization in biological neural circuits also improves learning. Here, we explore this possibility using ANNs with separate excitatory and inhibitory populations trained on an image recognition task with variable luminosity. We find that inhibition-mediated normalization does not improve learning if normalization is applied only during inference. However, when this normalization is extended to include back-propagated errors, performance improves significantly. These results suggest that if inhibition-mediated normalization improves learning in the brain, it additionally requires the normalization of learning signals.", "date": "2026-03-18", "categories": []}
{"id": "2603.12286", "title": "The DIME Architecture: A Unified Operational Algorithm for Neural Representation, Dynamics, Control and Integration", "abstract": "Modern neuroscience has accumulated extensive evidence on perception, memory, prediction, valuation, and consciousness, yet still lacks an explicit operational architecture capable of integrating these phenomena within a unified computational framework. Existing theories address specific aspects of neural function: predictive coding and active inference emphasize hierarchical inference and prediction error minimization; engram theories explain memory through distributed cell assemblies; neuromodulatory accounts focus on value-dependent regulation of plasticity and behaviour; and global workspace or large-scale network models investigate mechanisms underlying conscious access. Despite their explanatory power, these approaches remain only partially integrated at the architectural level. This work introduces DIME (Detect-Integrate-Mark-Execute), a neural architecture organizing perception, memory, valuation, and conscious access within a common operational cycle. The framework includes four interacting components: engrams, distributed recurrent neural structures supporting multiple activation trajectories; execution threads, spatiotemporal trajectories implementing neural processes; marker systems, neuromodulatory and limbic mechanisms regulating gain, plasticity, and trajectory selection; and hyperengrams, large-scale integrative states associated with operational conscious access. The framework is consistent with empirical evidence from hippocampal indexing, recurrent cortical processing, replay phenomena, large-scale network integration, and neuromodulatory regulation. Formulated at an abstract computational level, DIME may also inform artificial intelligence and robotics by providing an architectural template in which representation, valuation, and temporal sequencing emerge from a unified mechanism. An extended theoretical exposition is available in a companion monograph on Zenodo.", "date": "2026-03-07", "categories": []}
{"id": "2409.05771", "title": "Evidence from fMRI Supports a Two-Phase Abstraction Process in Language Models", "abstract": "Research has repeatedly demonstrated that intermediate hidden states extracted from large language models are able to predict measured brain response to natural language stimuli. Yet, very little is known about the representation properties that enable this high prediction performance. Why is it the intermediate layers, and not the output layers, that are most capable for this unique and highly general transfer task? In this work, we show that evidence from language encoding models in fMRI supports the existence of a two-phase abstraction process within LLMs. We use manifold learning methods to show that this abstraction process naturally arises over the course of training a language model and that the first \"composition\" phase of this abstraction process is compressed into fewer layers as training continues. Finally, we demonstrate a strong correspondence between layerwise encoding performance and the intrinsic dimensionality of representations from LLMs. We give initial evidence that this correspondence primarily derives from the inherent compositionality of LLMs and not their next-word prediction properties.", "date": "2024-09-09", "categories": []}
{"id": "2510.02182", "title": "Uncovering Semantic Selectivity of Latent Groups in Higher Visual Cortex with Mutual Information-Guided Diffusion", "abstract": "Understanding how neural populations in higher visual areas encode object-centered visual information remains a central challenge in computational neuroscience. Prior works have investigated representational alignment between artificial neural networks and the visual cortex. Nevertheless, these findings are indirect and offer limited insights to the structure of neural populations themselves. Similarly, decoding-based methods have quantified semantic features from neural populations but have not uncovered their underlying organizations. This leaves open a scientific question: \"how feature-specific visual information is distributed across neural populations in higher visual areas, and whether it is organized into structured, semantically meaningful subspaces.\" To tackle this problem, we present MIG-Vis, a method that leverages the generative power of diffusion models to visualize and validate the visual-semantic attributes encoded in neural latent subspaces. Our method first uses a variational autoencoder to infer a group-wise disentangled neural latent subspace from neural populations. Subsequently, we propose a mutual information (MI)-guided diffusion synthesis procedure to visualize the specific visual-semantic features encoded by each latent group. We validate MIG-Vis on multi-session neural spiking datasets from the inferior temporal (IT) cortex of two macaques. The synthesized results demonstrate that our method identifies neural latent groups with clear semantic selectivity to diverse visual features, including object pose, inter-category transformations, and intra-class content. These findings provide direct, interpretable evidence of structured semantic representation in the higher visual cortex and advance our understanding of its encoding principles.", "date": "2025-10-02", "categories": []}
{"id": "2603.11000", "title": "Cross-Species Transfer Learning for Electrophysiology-to-Transcriptomics Mapping in Cortical GABAergic Interneurons", "abstract": "Single-cell electrophysiological recordings provide a powerful window into neuronal functional diversity and offer an interpretable route for linking intrinsic physiology to transcriptomic identity. Here, we replicate and extend the electrophysiology-to-transcriptomics framework introduced by Gouwens et al. (2020) using publicly available Allen Institute Patch-seq datasets from both mouse and human cortex. We focus on GABAergic inhibitory interneurons to target a subclass structure (Lamp5, Pvalb, Sst, Vip) that is comparable and conserved across species. After quality control, we analyzed 3,699 mouse visual cortex neurons and 506 human neocortical neurons from neurosurgical resections. Using standardized electrophysiological features and sparse PCA, we reproduced the major class-level separations reported in the original mouse study. For supervised prediction, a class-balanced random forest provided a strong feature-engineered baseline in mouse data and a reduced but still informative baseline in human data. We then developed an attention-based BiLSTM that operates directly on the structured IPFX feature-family representation, avoiding sPCA and providing feature-family-level interpretability via learned attention weights. Finally, we evaluated a cross-species transfer setting in which the sequence model is pretrained on mouse data and fine-tuned on human data for an aligned 4-class task, improving human macro-F1 relative to a human-only training baseline. Together, these results confirm reproducibility of the Gouwens pipeline in mouse data, demonstrate that sequence models can match feature-engineered baselines, and show that mouse-to-human transfer learning can provide measurable gains for human subclass prediction.", "date": "2026-03-11", "categories": []}
{"id": "2603.08742", "title": "Robust Parameter and State Estimation in Multiscale Neuronal Systems Using Physics-Informed Neural Networks", "abstract": "Inferring biophysical parameters and hidden state variables from partial and noisy observations is a fundamental challenge in computational neuroscience. This problem is particularly difficult for fast - slow spiking and bursting models, where strong nonlinearities, multiscale dynamics, and limited observational data often lead to severe sensitivity to initial parameter guesses and convergence failure in the methods replying on the traditional numerical forward solvers. In this work, we developed a physics-informed neural network (PINN) framework for the joint reconstruction of unobserved state variables and the estimation of unknown biophysical parameters in neuronal models. We demonstrate the effectiveness of the method on biophysical neuron models, including the Morris-Lecar model across multiple spiking and bursting regimes and a respiratory model neuron. The method requires only partial voltage observations over short observation windows and remains robust even when initialized with non-informative parameter guesses. These results suggest that PINN can deliver robust and accurate parameter inference and state reconstruction, providing a promising alternative for inverse problems in multiscale neuronal dynamics, where traditional techniques often struggle.", "date": "2026-02-27", "categories": []}
{"id": "2603.07369", "title": "Task learning increases information redundancy of neural responses in macaque visual cortex", "abstract": "How does the brain optimize sensory information for decision-making in new tasks? One hypothesis suggests learning reduces redundancy in neural representations to improve efficiency, while another, based on Bayesian inference, predicts learning increases redundancy by distributing information across neurons. We tested these hypotheses by tracking population responses in macaque cortical area V4 as monkeys learned visual discrimination tasks. We found strong support for the Bayesian predictions: task learning increased redundancy in neural responses over weeks of training and within single trials. This redundancy did not reduce information but instead increased the information carried by individual neurons. These insights suggest sensory processing in the brain reflects a generative rather than discriminative inference process.", "date": "2026-03-07", "categories": []}
{"id": "2509.24222", "title": "Uni-NTFM: A Unified Foundation Model for EEG Signal Representation Learning", "abstract": "Current foundation models for electroencephalography (EEG) rely on architectures adapted from computer vision or natural language processing, typically treating neural signals as pixel grids or token sequences. This approach overlooks that the neural activity is activated by diverse sparse coding across a complex geometric topological cortex. Inspired by biological neural mechanisms, we propose the Unified Neural Topological Foundation Model (Uni-NTFM), an architecture rooted in three core neuroscience principles. In detail, to align with the brain's decoupled coding mechanism, we design the Heterogeneous Feature Projection Module. This module simultaneously encodes both time-domain non-stationary transients and frequency-domain steady-state rhythms, ensuring high quality in both waveform morphology and spectral rhythms. Moreover, we introduce a Topological Embedding mechanism to inject structured spatial priors and align different sensor configurations onto a unified latent functional topography, effectively reconstructing the geometry of brain regions. Furthermore, we achieve functional modularization and sparse coding efficiency of biological networks by constructing the Mixture-of-Experts Transformer network. This dynamic routing mechanism assigns different signal patterns and tasks to specialized neural subnetworks, and effectively preventing task interference while increasing the model capacity to record-breaking 1.9 billion parameters. Uni-NTFM is pre-trained on a diverse corpus comprising 28,000 hours of EEG data, and outperforms existing models across nine distinct downstream tasks under both linear probing and fine-tuning settings, demonstrating that aligning model architecture with neural mechanisms is significant to learn universal representations and achieve generalizable brain decoding. Our code is available at: https://anonymous.4open.science/r/Uni-NTFM-0924.", "date": "2025-09-29", "categories": []}
{"id": "2603.03343", "title": "Neuro-Symbolic Decoding of Neural Activity", "abstract": "We propose NEURONA, a neuro-symbolic framework for fMRI decoding and concept grounding in neural activity. Leveraging image- and video-based fMRI question-answering datasets, NEURONA learns to decode interacting concepts from visual stimuli based on patterns of fMRI responses, integrating symbolic reasoning and compositional execution with fMRI grounding across brain regions. We demonstrate that incorporating structural priors (e.g., compositional predicate-argument dependencies between concepts) into the decoding process significantly improves both decoding accuracy over precise queries, and notably, generalization to unseen queries at test time. With NEURONA, we highlight neuro-symbolic frameworks as promising tools for understanding neural activity.", "date": "2026-02-22", "categories": []}
{"id": "2506.01303", "title": "Dynamic Manifold Hopfield Networks for Context-Dependent Associative Memory", "abstract": "Neural population activity in cortical and hippocampal circuits can be flexibly reorganized by context, suggesting that cognition relies on dynamic manifolds rather than static representations. However, how such dynamic organization can be realized mechanistically within a unified dynamical system remains unclear. Continuous Hopfield networks provide a classical attractor framework in which neural dynamics follow gradient descent on a fixed energy landscape, constraining retrieval within a static attractor manifold geometry. Extending this approach, we introduce Dynamic Manifold Hopfield Networks (DMHN), continuous dynamical models in which contextual modulation dynamically reshapes attractor geometry, transforming a static attractor manifold into a context-dependent family of neural manifolds. In DMHN, network interactions are learned in a data-driven manner, to intrinsically deform the geometry of its attractor manifold across cues without explicit context-specific parameterization. As a result, in associative retrieval, DMHN achieve substantially higher capacity and robustness than classical and modern Hopfield networks: when storing $2N$ patterns in a network of $N$ neurons, DMHN attain reliable retrieval with an average accuracy of 64%, compared with 1% and 13% for classical and modern variants, respectively. Together, these results establish dynamic reorganization of attractor manifold geometry as a principled mechanism for context-dependent remapping in neural associative memory.", "date": "2025-06-02", "categories": []}
{"id": "2508.14492", "title": "Synaptic bundle theory for spike-driven sensor-motor system: More than eight independent synaptic bundles collapse reward-STDP learning", "abstract": "Neuronal spikes directly drive muscles and endow animals with agile movements, but applying the spike-based control signals to actuators in artificial sensor-motor systems inevitably causes a collapse of learning. We developed a system that can vary \\emph{the number of independent synaptic bundles} in sensor-to-motor connections. This paper demonstrates the following four findings: (i) Learning collapses once the number of motor neurons or the number of independent synaptic bundles exceeds a critical limit. (ii) The probability of learning failure is increased by a smaller number of motor neurons, while (iii) if learning succeeds, a smaller number of motor neurons leads to faster learning. (iv) The number of weight updates that move in the opposite direction of the optimal weight can quantitatively explain these results. The functions of spikes remain largely unknown. Identifying the parameter range in which learning systems using spikes can be constructed will make it possible to study the functions of spikes that were previously inaccessible due to the difficulty of learning.", "date": "2025-08-20", "categories": []}
{"id": "2509.23566", "title": "Towards Interpretable Visual Decoding with Attention to Brain Representations", "abstract": "Recent work has demonstrated that complex visual stimuli can be decoded from human brain activity using deep generative models, offering new ways to probe how the brain represents real-world scenes. However, many existing approaches first map brain signals into intermediate image or text feature spaces before guiding the generative process, which obscures the contributions of different brain areas to the final reconstruction output. In this work, we propose NeuroAdapter, a visual decoding framework that directly conditions a latent diffusion model on brain representations, bypassing the need for intermediate feature spaces. Our method demonstrates competitive visual reconstruction quality on public fMRI datasets compared to prior work, while providing greater transparency into how brain signals drive visual reconstruction. To this end, we introduce an Image-Brain BI-directional interpretability framework (IBBI) that analyzes cross-attention patterns across diffusion denoising steps to reveal how different cortical areas influence the unfolding generative trajectory. Our work highlights the potential of end-to-end brain-to-image reconstruction and establishes a path for interpretable neural decoding.", "date": "2025-09-28", "categories": []}
{"id": "2509.26560", "title": "Estimating Dimensionality of Neural Representations from Finite Samples", "abstract": "The global dimensionality of a neural representation manifold provides rich insight into the computational process underlying both artificial and biological neural networks. However, all existing measures of global dimensionality are sensitive to the number of samples, i.e., the number of rows and columns of the sample matrix. We show that, in particular, the participation ratio of eigenvalues, a popular measure of global dimensionality, is highly biased with small sample sizes, and propose a bias-corrected estimator that is more accurate with finite samples and with noise. On synthetic data examples, we demonstrate that our estimator can recover the true known dimensionality. We apply our estimator to neural brain recordings, including calcium imaging, electrophysiological recordings, and fMRI data, and to the neural activations in a large language model and show our estimator is invariant to the sample size. Finally, our estimators can additionally be used to measure the local dimensionalities of curved neural manifolds by weighting the finite samples appropriately.", "date": "2025-09-30", "categories": []}
{"id": "2510.09951", "title": "Emergence of Spatial Representation in an Actor-Critic Agent with Hippocampus-Inspired Sequence Generator", "abstract": "Sequential firing of hippocampal place cells is often attributed to sequential sensory drive along a trajectory, and has also been attributed to planning and other cognitive functions. Here, we propose a mechanistic and parsimonious interpretation to complement these ideas: hippocampal sequences arise from intrinsic recurrent circuitry that propagates transient input over long horizons, acting as a temporal memory buffer that is especially useful when reliable sensory evidence is sparse. We implement this idea with a minimal sequence generator inspired by neurobiology and pair it with an actor-critic learner for egocentric visual navigation. Our agent reliably solves a continuous maze without explicit geometric cues, with performance depending on the length of the recurrent sequence. Crucially, the model outperforms LSTM cores under sparse input conditions (16 channels, $\\sim2.5\\%$ activity), but not under dense input, revealing a strong interaction between representational sparsity and memory architecture. Through learning, units develop localized place fields, distance-dependent spatial kernels, and task-dependent remapping, while inputs to the sequence generator orthogonalize and spatial information increases across layers. These phenomena align with neurobiological data and are causal to performance. Together, our results show that sparse input synergizes with sequence-generating dynamics, providing both a mechanistic account of place cell sequences in the mammalian hippocampus and a simple inductive bias for reinforcement learning based on sparse egocentric inputs in navigation tasks.", "date": "2025-10-11", "categories": []}
{"id": "2510.18808", "title": "Does Feedback Alignment Work at Biological Timescales?", "abstract": "Feedback alignment and related weight-transport-free algorithms are often proposed as biologically plausible alternatives to backpropagation, yet they are typically formulated in discrete phases with implicitly synchronized forward and error signals. We develop a continuous-time model of feedback-alignment-type learning in which neural activities and synaptic weights evolve together under coupled first-order dynamics with distinct propagation, plasticity, and decay time constants. We show that learning is governed by the temporal overlap between presynaptic drive and a locally projected error signal, providing an analytic explanation for robustness to moderate timing mismatch and for failure when mismatch eliminates overlap. Our results show that in order for feedback-alignment-type algorithms to work at biological timescales, they must obey the same temporal overlap principle that applies to other biological processes like eligibility traces.", "date": "2025-10-21", "categories": []}
{"id": "2510.25976", "title": "Brain-IT: Image Reconstruction from fMRI via Brain-Interaction Transformer", "abstract": "Reconstructing images seen by people from their fMRI brain recordings provides a non-invasive window into the human brain. Despite recent progress enabled by diffusion models, current methods often lack faithfulness to the actual seen images. We present \"Brain-IT\", a brain-inspired approach that addresses this challenge through a Brain Interaction Transformer (BIT), allowing effective interactions between clusters of functionally-similar brain-voxels. These functional-clusters are shared by all subjects, serving as building blocks for integrating information both within and across brains. All model components are shared by all clusters & subjects, allowing efficient training with a limited amount of data. To guide the image reconstruction, BIT predicts two complementary localized patch-level image features: (i)high-level semantic features which steer the diffusion model toward the correct semantic content of the image; and (ii)low-level structural features which help to initialize the diffusion process with the correct coarse layout of the image. BIT's design enables direct flow of information from brain-voxel clusters to localized image features. Through these principles, our method achieves image reconstructions from fMRI that faithfully reconstruct the seen images, and surpass current SotA approaches both visually and by standard objective metrics. Moreover, with only 1-hour of fMRI data from a new subject, we achieve results comparable to current methods trained on full 40-hour recordings.", "date": "2025-10-29", "categories": []}
{"id": "2407.06195", "title": "Spectral-Stimulus Information for Self-Supervised Stimulus Encoding", "abstract": "Mammalian spatial navigation relies on specialized neurons, such as place and grid cells, which encode position based on self-motion and environmental cues. While extensive research has explored the computational role of grid cells, the principles underlying efficient place cell coding remain less understood. Existing spatial information rate measures primarily assess single-neuron encoding, limiting insights into population-level representations, while, the role of correlation in neural coding remains a subject of considerable debate. To address this, we introduce novel, correlation-aware information-theoretic measures that quantify the encoding efficiency of multiple neurons, including the joint stimulus information rate for neuron pairs and the spectral-stimulus information for arbitrary sized populations. The spectral-stimulus information, defined as the leading eigenvalue of the stimulus information matrix, is maximized when neurons exhibit localized, non-overlapping firing fields, mirroring place cell and head direction cell activity. We apply these measures to neural data recorded in mice and monkeys, elucidating differences in encoding efficiency across neuronal pairs and populations. Then, we demonstrate that these measures can be used to train recurrent neural networks (RNNs) via self-supervised learning, leading to the emergence of place cells and head direction cells. Our findings highlight how neural populations collectively encode stimuli, offering a more comprehensive framework for understanding stimulus encoding and optimizing artificial navigation systems in novel environments.", "date": "2024-06-10", "categories": []}
{"id": "2507.09513", "title": "Animal behavioral analysis and neural encoding with transformer-based self-supervised pretraining", "abstract": "The brain can only be fully understood through the lens of the behavior it generates -- a guiding principle in modern neuroscience research that nevertheless presents significant technical challenges. Many studies capture behavior with cameras, but video analysis approaches typically rely on specialized models requiring extensive labeled data. We address this limitation with BEAST(BEhavioral Analysis via Self-supervised pretraining of Transformers), a novel and scalable framework that pretrains experiment-specific vision transformers for diverse neuro-behavior analyses. BEAST combines masked autoencoding with temporal contrastive learning to effectively leverage unlabeled video data. Through comprehensive evaluation across multiple species, we demonstrate improved performance in three critical neuro-behavioral tasks: extracting behavioral features that correlate with neural activity, and pose estimation and action segmentation in both the single- and multi-animal settings. Our method establishes a powerful and versatile backbone model that accelerates behavioral analysis in scenarios where labeled data remains scarce.", "date": "2025-07-13", "categories": []}
{"id": "2602.23285", "title": "ODEBrain: Continuous-Time EEG Graph for Modeling Dynamic Brain Networks", "abstract": "Modeling neural population dynamics is crucial for foundational neuroscientific research and various clinical applications. Conventional latent variable methods typically model continuous brain dynamics through discretizing time with recurrent architecture, which necessarily results in compounded cumulative prediction errors and failure of capturing instantaneous, nonlinear characteristics of EEGs. We propose ODEBRAIN, a Neural ODE latent dynamic forecasting framework to overcome these challenges by integrating spatio-temporal-frequency features into spectral graph nodes, followed by a Neural ODE modeling the continuous latent dynamics. Our design ensures that latent representations can capture stochastic variations of complex brain states at any given time point. Extensive experiments verify that ODEBRAIN can improve significantly over existing methods in forecasting EEG dynamics with enhanced robustness and generalization capabilities.", "date": "2026-02-26", "categories": []}
{"id": "2602.23960", "title": "SHINE: Sequential Hierarchical Integration Network for EEG and MEG", "abstract": "How natural speech is represented in the brain constitutes a major challenge for cognitive neuroscience, with cortical envelope-following responses playing a central role in speech decoding. This paper presents our approach to the Speech Detection task in the LibriBrain Competition 2025, utilizing over 50 hours of magnetoencephalography (MEG) signals from a single participant listening to LibriVox audiobooks. We introduce the proposed Sequential Hierarchical Integration Network for EEG and MEG (SHINE) to reconstruct the binary speech-silence sequences from MEG signals. In the Extended Track, we further incorporated auxiliary reconstructions of speech envelopes and Mel spectrograms to enhance training. Ensemble methods combining SHINE with baselines (BrainMagic, AWavNet, ConvConcatNet) achieved F1-macro scores of 0.9155 (Standard Track) and 0.9184 (Extended Track) on the leaderboard test set.", "date": "2026-02-27", "categories": []}
{"id": "2509.24597", "title": "Inducing Dyslexia in Vision Language Models", "abstract": "Dyslexia, a neurodevelopmental disorder characterized by persistent reading difficulties, is often linked to reduced activity of the visual word form area (VWFA) in the ventral occipito-temporal cortex. Traditional approaches to studying dyslexia, such as behavioral and neuroimaging methods, have provided valuable insights but remain limited in their ability to test causal hypotheses about the underlying mechanisms of reading impairments. In this study, we use large-scale vision-language models (VLMs) to simulate dyslexia by functionally identifying and perturbing artificial analogues of word processing. Using stimuli from cognitive neuroscience, we identify visual-word-form-selective units within VLMs and demonstrate that they predict human VWFA neural responses. Ablating model VWF units leads to selective impairments in reading tasks while general visual and language comprehension abilities remain intact. In particular, the resulting model matches dyslexic humans' phonological deficits without a significant change in orthographic processing, and mirrors dyslexic behavior in font sensitivity. Taken together, our modeling results replicate key characteristics of dyslexia and establish a computational framework for investigating brain disorders.", "date": "2025-09-29", "categories": []}
{"id": "2602.20932", "title": "Hierarchic-EEG2Text: Assessing EEG-To-Text Decoding across Hierarchical Abstraction Levels", "abstract": "An electroencephalogram (EEG) records the spatially averaged electrical activity of neurons in the brain, measured from the human scalp. Prior studies have explored EEG-based classification of objects or concepts, often for passive viewing of briefly presented image or video stimuli, with limited classes. Because EEG exhibits a low signal-to-noise ratio, recognizing fine-grained representations across a large number of classes remains challenging; however, abstract-level object representations may exist. In this work, we investigate whether EEG captures object representations across multiple hierarchical levels, and propose episodic analysis, in which a Machine Learning (ML) model is evaluated across various, yet related, classification tasks (episodes). Unlike prior episodic EEG studies that rely on fixed or randomly sampled classes of equal cardinality, we adopt hierarchy-aware episode sampling using WordNet to generate episodes with variable classes of diverse hierarchy. We also present the largest episodic framework in the EEG domain for detecting observed text from EEG signals in the PEERS dataset, comprising $931538$ EEG samples under $1610$ object labels, acquired from $264$ human participants (subjects) performing controlled cognitive tasks, enabling the study of neural dynamics underlying perception, decision-making, and performance monitoring.\n  We examine how the semantic abstraction level affects classification performance across multiple learning techniques and architectures, providing a comprehensive analysis. The models tend to improve performance when the classification categories are drawn from higher levels of the hierarchy, suggesting sensitivity to abstraction. Our work highlights abstraction depth as an underexplored dimension of EEG decoding and motivates future research in this direction.", "date": "2026-02-24", "categories": []}
{"id": "2512.04808", "title": "Setting up for failure: automatic discovery of the neural mechanisms of cognitive errors", "abstract": "Discovering the neural mechanisms underpinning cognition is one of the grand challenges of neuroscience. However, previous approaches for building models of RNN dynamics that explain behaviour required iterative refinement of architectures and/or optimisation objectives, resulting in a piecemeal, and mostly heuristic, human-in-the-loop process. Here, we offer an alternative approach that automates the discovery of viable RNN mechanisms by explicitly training RNNs to reproduce behaviour, including the same characteristic errors and suboptimalities, that humans and animals produce in a cognitive task. Achieving this required two main innovations. First, as the amount of behavioural data that can be collected in experiments is often too limited to train RNNs, we use a non-parametric generative model of behavioural responses to produce surrogate data for training RNNs. Second, to capture all relevant statistical aspects of the data, we developed a novel diffusion model-based approach for training RNNs. To showcase the potential of our approach, we chose a visual working memory task as our test-bed, as behaviour in this task is well known to produce response distributions that are patently multimodal (due to swap errors). The resulting network dynamics correctly qualitative features of macaque neural data. Importantly, these results were not possible to obtain with more traditional approaches, i.e., when only a limited set of behavioural signatures (rather than the full richness of behavioural response distributions) were fitted, or when RNNs were trained for task optimality (instead of reproducing behaviour). Our approach also yields novel predictions about the mechanism of swap errors, which can be readily tested in experiments. These results suggest that fitting RNNs to rich patterns of behaviour provides a powerful way to automatically discover mechanisms of important cognitive functions.", "date": "2025-12-04", "categories": []}
{"id": "2601.08549", "title": "Contrastive and Multi-Task Learning on Noisy Brain Signals with Nonlinear Dynamical Signatures", "abstract": "We introduce a two-stage multitask learning framework for analyzing Electroencephalography (EEG) signals that integrates denoising, dynamical modeling, and representation learning. In the first stage, a denoising autoencoder is trained to suppress artifacts and stabilize temporal dynamics, providing robust signal representations. In the second stage, a multitask architecture processes these denoised signals to achieve three objectives: motor imagery classification, chaotic versus non-chaotic regime discrimination using Lyapunov exponent-based labels, and self-supervised contrastive representation learning with NT-Xent loss. A convolutional backbone combined with a Transformer encoder captures spatial-temporal structure, while the dynamical task encourages sensitivity to nonlinear brain dynamics. This staged design mitigates interference between reconstruction and discriminative goals, improves stability across datasets, and supports reproducible training by clearly separating noise reduction from higher-level feature learning. Empirical studies show that our framework not only enhances robustness and generalization but also surpasses strong baselines and recent state-of-the-art methods in EEG decoding, highlighting the effectiveness of combining denoising, dynamical features, and self-supervised learning.", "date": "2026-01-13", "categories": []}
{"id": "2602.18637", "title": "Online decoding of rat self-paced locomotion speed from EEG using recurrent neural networks", "abstract": "$\\textit{Objective.}$ Accurate neural decoding of locomotion holds promise for advancing rehabilitation, prosthetic control, and understanding neural correlates of action. Recent studies have demonstrated decoding of locomotion kinematics across species on motorized treadmills. However, efforts to decode locomotion speed in more natural contexts$-$where pace is self-selected rather than externally imposed$-$are scarce, generally achieve only modest accuracy, and require intracranial implants. Here, we aim to decode self-paced locomotion speed non-invasively and continuously using cortex-wide EEG recordings from rats. $\\textit{Approach.}$ We introduce an asynchronous brain$-$computer interface (BCI) that processes a stream of 32-electrode skull-surface EEG (0.01$-$45 Hz) to decode instantaneous speed from a non-motorized treadmill during self-paced locomotion in head-fixed rats. Using recurrent neural networks and a dataset of over 133 h of recordings, we trained decoders to map ongoing EEG activity to treadmill speed. $\\textit{Main results.}$ Our decoding achieves a correlation of 0.88 ($R^2$ = 0.78) for speed, primarily driven by visual cortex electrodes and low-frequency ($< 8$ Hz) oscillations. Moreover, pre-training on a single session permitted decoding on other sessions from the same rat, suggesting uniform neural signatures that generalize across sessions but fail to transfer across animals. Finally, we found that cortical states not only carry information about current speed, but also about future and past dynamics, extending up to 1000 ms. $\\textit{Significance.}$ These findings demonstrate that self-paced locomotion speed can be decoded accurately and continuously from non-invasive, cortex-wide EEG. Our approach provides a framework for developing high-performing, non-invasive BCI systems and contributes to understanding distributed neural representations of action dynamics.", "date": "2026-02-20", "categories": []}
{"id": "2602.18401", "title": "Leakage and Second-Order Dynamics Improve Hippocampal RNN Replay", "abstract": "Biological neural networks (like the hippocampus) can internally generate \"replay\" resembling stimulus-driven activity. Recent computational models of replay use noisy recurrent neural networks (RNNs) trained to path-integrate. Replay in these networks has been described as Langevin sampling, but new modifiers of noisy RNN replay have surpassed this description. We re-examine noisy RNN replay as sampling to understand or improve it in three ways: (1) Under simple assumptions, we prove that the gradients replay activity should follow are time-varying and difficult to estimate, but readily motivate the use of hidden state leakage in RNNs for replay. (2) We confirm that hidden state adaptation (negative feedback) encourages exploration in replay, but show that it incurs non-Markov sampling that also slows replay. (3) We propose the first model of temporally compressed replay in noisy path-integrating RNNs through hidden state momentum, connect it to underdamped Langevin sampling, and show that, together with adaptation, it counters slowness while maintaining exploration. We verify our findings via path-integration of 2D triangular and T-maze paths and of high-dimensional paths of synthetic rat place cell activity.", "date": "2026-02-20", "categories": []}
{"id": "2510.15828", "title": "GENESIS: A Generative Model of Episodic-Semantic Interaction", "abstract": "A central challenge in cognitive neuroscience is to explain how semantic and episodic memory, two major forms of declarative memory, typically associated with cortical and hippocampal processing, interact to support learning, recall, and imagination. Despite significant advances, we still lack a unified computational framework that jointly accounts for core empirical phenomena across both semantic and episodic processing domains. Here, we introduce the Generative Episodic-Semantic Integration System (GENESIS), a computational model that formalizes memory as the interaction between two limited-capacity generative systems: a Cortical-VAE, supporting semantic learning and generalization, and a Hippocampal-VAE, supporting episodic encoding and retrieval within a retrieval-augmented generation (RAG) architecture. GENESIS reproduces hallmark behavioral findings, including generalization in semantic memory, recognition, serial recall effects and gist-based distortions in episodic memory, and constructive episodic simulation, while capturing their dynamic interactions. The model elucidates how capacity constraints shape the fidelity and memorability of experiences, how semantic processing introduces systematic distortions in episodic recall, and how episodic replay can recombine previous experiences. Together, these results provide a principled account of memory as an active, constructive, and resource-bounded process. GENESIS thus advances a unified theoretical framework that bridges semantic and episodic memory, offering new insights into the generative foundations of human cognition.", "date": "2025-10-17", "categories": []}
{"id": "2602.08275", "title": "Linguistics and Human Brain: A Perspective of Computational Neuroscience", "abstract": "Elucidating the language-brain relationship requires bridging the methodological gap between the abstract theoretical frameworks of linguistics and the empirical neural data of neuroscience. Serving as an interdisciplinary cornerstone, computational neuroscience formalizes the hierarchical and dynamic structures of language into testable neural models through modeling, simulation, and data analysis. This enables a computational dialogue between linguistic hypotheses and neural mechanisms. Recent advances in deep learning, particularly large language models (LLMs), have powerfully advanced this pursuit. Their high-dimensional representational spaces provide a novel scale for exploring the neural basis of linguistic processing, while the \"model-brain alignment\" framework offers a methodology to evaluate the biological plausibility of language-related theories.", "date": "2026-02-09", "categories": []}
{"id": "2506.18314", "title": "BrainSymphony: A parameter-efficient multimodal foundation model for brain dynamics with limited data", "abstract": "Foundation models are transforming neuroscience but are often prohibitively large, data-hungry, and difficult to deploy. Here, we introduce BrainSymphony, a lightweight and parameter-efficient foundation model with plug-and-play integration of fMRI time series and diffusion-derived structural connectivity, allowing unimodal or multimodal training and deployment without architectural changes while requiring substantially less data compared to the state-of-the-art. The model processes fMRI time series through parallel spatial and temporal transformer streams, distilled into compact embeddings by a Perceiver module, while a novel signed graph transformer encodes anatomical connectivity from diffusion MRI. These complementary representations are then combined through an adaptive fusion mechanism. Despite its compact design, BrainSymphony consistently outperforms larger models on benchmarks spanning prediction, classification, and unsupervised network discovery. Highlighting the model's generalizability and interpretability, attention maps reveal drug-induced context-dependent reorganization of cortical hierarchies in an independent psilocybin neuroimaging dataset. BrainSymphony delivers accessible, interpretable, and clinically meaningful results and demonstrates that architecturally informed, multimodal models can surpass much larger counterparts and advance applications of AI in neuroscience.", "date": "2025-06-23", "categories": []}
{"id": "2602.11558", "title": "Brain4FMs: A Benchmark of Foundation Models for Electrical Brain Signal", "abstract": "Brain Foundation Models (BFMs) are transforming neuroscience by enabling scalable and transferable learning from neural signals, advancing both clinical diagnostics and cutting-edge neuroscience exploration. Their emergence is powered by large-scale clinical recordings, particularly electroencephalography (EEG) and intracranial EEG, which provide rich temporal and spatial representations of brain dynamics. However, despite their rapid proliferation, the field lacks a unified understanding of existing methodologies and a standardized evaluation framework. To fill this gap, we map the benchmark design space along two axes: (i) from the model perspective, we organize BFMs under a self-supervised learning (SSL) taxonomy; and (ii) from the dataset perspective, we summarize common downstream tasks and curate representative public datasets across clinical and human-centric neurotechnology applications. Building on this consolidation, we introduce Brain4FMs, an open evaluation platform with plug-and-play interfaces that integrates 15 representative BFMs and 18 public datasets. It enables standardized comparisons and analysis of how pretraining data, SSL strategies, and architectures affect generalization and downstream performance, guiding more accurate and transferable BFMs. The code is available at https://anonymous.4open.science/r/Brain4FMs-85B8.", "date": "2026-02-12", "categories": []}
{"id": "2602.11956", "title": "TAVAE: A VAE with Adaptable Priors Explains Contextual Modulation in the Visual Cortex", "abstract": "The brain interprets visual information through learned regularities, a computation formalized as probabilistic inference under a prior. The visual cortex establishes priors for this inference, some delivered through established top-down connections that inform low-level cortices about statistics represented at higher levels in the cortical hierarchy. While evidence shows that adaptation leads to priors reflecting the structure of natural images, it remains unclear whether similar priors can be flexibly acquired when learning a specific task. To investigate this, we built a generative model of V1 optimized for a simple discrimination task and analyzed it together with large-scale recordings from mice performing an analogous task. In line with recent approaches, we assumed that neuronal activity in V1 corresponds to latent posteriors in the generative model, enabling investigation of task-related priors in neuronal responses. To obtain a flexible test bed, we extended the VAE formalism so that a task can be acquired efficiently by reusing previously learned representations. Task-specific priors learned by this Task-Amortized VAE were used to investigate biases in mice and model when presenting stimuli that violated trained task statistics. Mismatch between learned task statistics and incoming sensory evidence produced signatures of uncertainty in stimulus category in the TAVAE posterior, reflecting properties of bimodal response profiles in V1 recordings. The task-optimized generative model accounted for key characteristics of V1 population activity, including within-day updates to population responses. Our results confirm that flexible task-specific contextual priors can be learned on demand by the visual system and deployed as early as the entry level of visual cortex.", "date": "2026-02-12", "categories": []}
{"id": "2310.03111", "title": "Multi-modal Gaussian Process Variational Autoencoders for Neural and Behavioral Data", "abstract": "Characterizing the relationship between neural population activity and behavioral data is a central goal of neuroscience. While latent variable models (LVMs) are successful in describing high-dimensional time-series data, they are typically only designed for a single type of data, making it difficult to identify structure shared across different experimental data modalities. Here, we address this shortcoming by proposing an unsupervised LVM which extracts temporally evolving shared and independent latents for distinct, simultaneously recorded experimental modalities. We do this by combining Gaussian Process Factor Analysis (GPFA), an interpretable LVM for neural spiking data with temporally smooth latent space, with Gaussian Process Variational Autoencoders (GP-VAEs), which similarly use a GP prior to characterize correlations in a latent space, but admit rich expressivity due to a deep neural network mapping to observations. We achieve interpretability in our model by partitioning latent variability into components that are either shared between or independent to each modality. We parameterize the latents of our model in the Fourier domain, and show improved latent identification using this approach over standard GP-VAE methods. We validate our model on simulated multi-modal data consisting of Poisson spike counts and MNIST images that scale and rotate smoothly over time. We show that the multi-modal GP-VAE (MM-GPVAE) is able to not only identify the shared and independent latent structure across modalities accurately, but provides good reconstructions of both images and neural rates on held-out trials. Finally, we demonstrate our framework on two real world multi-modal experimental settings: Drosophila whole-brain calcium imaging alongside tracked limb positions, and Manduca sexta spike train measurements from ten wing muscles as the animal tracks a visual stimulus.", "date": "2023-10-04", "categories": []}
{"id": "2506.14957", "title": "POCO: Scalable Neural Forecasting through Population Conditioning", "abstract": "Predicting future neural activity is a core challenge in modeling brain dynamics, with applications ranging from scientific investigation to closed-loop neurotechnology. While recent models of population activity emphasize interpretability and behavioral decoding, neural forecasting-particularly across multi-session, spontaneous recordings-remains underexplored. We introduce POCO, a unified forecasting model that combines a lightweight univariate forecaster with a population-level encoder to capture both neuron-specific and brain-wide dynamics. Trained across five calcium imaging datasets spanning zebrafish, mice, and C. elegans, POCO achieves state-of-the-art accuracy at cellular resolution in spontaneous behaviors. After pre-training, POCO rapidly adapts to new recordings with minimal fine-tuning. Notably, POCO's learned unit embeddings recover biologically meaningful structure-such as brain region clustering-without any anatomical labels. Our comprehensive analysis reveals several key factors influencing performance, including context length, session diversity, and preprocessing. Together, these results position POCO as a scalable and adaptable approach for cross-session neural forecasting and offer actionable insights for future model design. By enabling accurate, generalizable forecasting models of neural dynamics across individuals and species, POCO lays the groundwork for adaptive neurotechnologies and large-scale efforts for neural foundation models. Code is available at https://github.com/yuvenduan/POCO.", "date": "2025-06-17", "categories": []}
{"id": "2412.12112", "title": "Generative Modeling of Neural Dynamics via Latent Stochastic Differential Equations", "abstract": "We propose a probabilistic framework for developing computational models of biological neural systems. In this framework, physiological recordings are viewed as discrete-time partial observations of an underlying continuous-time stochastic dynamical system which implements computations through its state evolution. To model this dynamical system, we employ a system of coupled stochastic differential equations with differentiable drift and diffusion functions and use variational inference to infer its states and parameters. This formulation enables seamless integration of existing mathematical models in the literature, neural networks, or a hybrid of both to learn and compare different models. We demonstrate this in our framework by developing a generative model that combines coupled oscillators with neural networks to capture latent population dynamics from single-cell recordings. Evaluation across three neuroscience datasets spanning different species, brain regions, and behavioral tasks show that these hybrid models achieve competitive performance in predicting stimulus-evoked neural and behavioral responses compared to sophisticated black-box approaches while requiring an order of magnitude fewer parameters, providing uncertainty estimates, and offering a natural language for interpretation.", "date": "2024-12-01", "categories": []}
{"id": "2602.07010", "title": "Learning Alzheimer's Disease Signatures by bridging EEG with Spiking Neural Networks and Biophysical Simulations", "abstract": "As the prevalence of Alzheimer's disease (AD) rises, improving mechanistic insight from non-invasive biomarkers is increasingly critical. Recent work suggests that circuit-level brain alterations manifest as changes in electroencephalography (EEG) spectral features detectable by machine learning. However, conventional deep learning approaches for EEG-based AD detection are computationally intensive and mechanistically opaque. Spiking neural networks (SNNs) offer a biologically plausible and energy-efficient alternative, yet their application to AD diagnosis remains largely unexplored.\n  We propose a neuro-bridge framework that links data-driven learning with minimal, biophysically grounded simulations, enabling bidirectional interpretation between machine learning signatures and circuit-level mechanisms in AD. Using resting-state clinical EEG, we train an SNN classifier that achieves competitive performance (AUC = 0.839) and identifies the aperiodic 1/f slope as a key discriminative marker.\n  The 1/f slope reflects excitation-inhibition balance. To interpret this mechanistically, we construct spiking network simulations in which inhibitory-to-excitatory synaptic ratios are systematically varied to emulate healthy, mild cognitive impairment, and AD-like states. Using both membrane potential-based and synaptic current-based EEG proxies, we reproduce empirical spectral slowing and altered alpha organization.\n  Incorporating empirical functional connectivity priors into multi-subnetwork simulations further enhances spectral differentiation, demonstrating that large-scale network topology constrains EEG signatures more strongly than excitation-inhibition balance alone. Overall, this neuro-bridge approach connects SNN-based classification with interpretable circuit simulations, advancing mechanistic understanding of EEG biomarkers while enabling scalable, explainable AD detection.", "date": "2026-01-30", "categories": []}
{"id": "2602.07261", "title": "Cognitive algorithms and systems of episodic memory, semantic memory and their learnings", "abstract": "Declarative memory, the memory that can be \"declared\" in words or languages, is made up of two dissociated parts: episodic memory and semantic memory. This dissociation has its neuroanatomical basis episodic memory is mostly associated with the hippocampus and semantic memory with the neocortex. The two memories, on the other hand, are closely related. Lesions in the hippocampus often result in various impairments of explicit memory, e.g., anterograde, retrograde and developmental amnesias, and semantic learning deficit. These impairments provide opportunities for us to understand how the two memories may be acquired, stored and organized. This chapter reviews several cognitive systems that are centered to mimic explicit memory, and other systems that are neuroanatomically based and are implemented to simulate those memory impairments mentioned above. This review includes: the structures of the computational systems, their learning rules, and their simulations of memory acquisition and impairments.", "date": "2026-02-06", "categories": []}
{"id": "2602.07495", "title": "Learning Brain Representation with Hierarchical Visual Embeddings", "abstract": "Decoding visual representations from brain signals has attracted significant attention in both neuroscience and artificial intelligence. However, the degree to which brain signals truly encode visual information remains unclear. Current visual decoding approaches explore various brain-image alignment strategies, yet most emphasize high-level semantic features while neglecting pixel-level details, thereby limiting our understanding of the human visual system. In this paper, we propose a brain-image alignment strategy that leverages multiple pre-trained visual encoders with distinct inductive biases to capture hierarchical and multi-scale visual representations, while employing a contrastive learning objective to achieve effective alignment between brain signals and visual embeddings. Furthermore, we introduce a Fusion Prior, which learns a stable mapping on large-scale visual data and subsequently matches brain features to this pre-trained prior, thereby enhancing distributional consistency across modalities. Extensive quantitative and qualitative experiments demonstrate that our method achieves a favorable balance between retrieval accuracy and reconstruction fidelity.", "date": "2026-02-07", "categories": []}
{"id": "2602.07539", "title": "Training-Driven Representational Geometry Modularization Predicts Brain Alignment in Language Models", "abstract": "How large language models (LLMs) align with the neural representation and computation of human language is a central question in cognitive science. Using representational geometry as a mechanistic lens, we addressed this by tracking entropy, curvature, and fMRI encoding scores throughout Pythia (70M-1B) training. We identified a geometric modularization where layers self-organize into stable low- and high-complexity clusters. The low-complexity module, characterized by reduced entropy and curvature, consistently better predicted human language network activity. This alignment followed heterogeneous spatial-temporal trajectories: rapid and stable in temporal regions (AntTemp, PostTemp), but delayed and dynamic in frontal areas (IFG, IFGorb). Crucially, reduced curvature remained a robust predictor of model-brain alignment even after controlling for training progress, an effect that strengthened with model scale. These results links training-driven geometric reorganization to temporal-frontal functional specialization, suggesting that representational smoothing facilitates neural-like linguistic processing.", "date": "2026-02-07", "categories": []}
{"id": "2602.07547", "title": "Linguistic properties and model scale in brain encoding: from small to compressed language models", "abstract": "Recent work has shown that scaling large language models (LLMs) improves their alignment with human brain activity, yet it remains unclear what drives these gains and which representational properties are responsible. Although larger models often yield better task performance and brain alignment, they are increasingly difficult to analyze mechanistically. This raises a fundamental question: what is the minimal model capacity required to capture brain-relevant representations? To address this question, we systematically investigate how constraining model scale and numerical precision affects brain alignment. We compare full-precision LLMs, small language models (SLMs), and compressed variants (quantized and pruned) by predicting fMRI responses during naturalistic language comprehension. Across model families up to 14B parameters, we find that 3B SLMs achieve brain predictivity indistinguishable from larger LLMs, whereas 1B models degrade substantially, particularly in semantic language regions. Brain alignment is remarkably robust to compression: most quantization and pruning methods preserve neural predictivity, with GPTQ as a consistent exception. Linguistic probing reveals a dissociation between task performance and brain predictivity: compression degrades discourse, syntax, and morphology, yet brain predictivity remains largely unchanged. Overall, brain alignment saturates at modest model scales and is resilient to compression, challenging common assumptions about neural scaling and motivating compact models for brain-aligned language modeling.", "date": "2026-02-07", "categories": []}
{"id": "2505.17329", "title": "Transformer brain encoders explain human high-level visual responses", "abstract": "A major goal of neuroscience is to understand brain computations during visual processing in naturalistic settings. A dominant approach is to use image-computable deep neural networks trained with different task objectives as a basis for linear encoding models. However, in addition to requiring estimation of a large number of linear encoding parameters, this approach ignores the structure of the feature maps both in the brain and the models. Recently proposed alternatives factor the linear mapping into separate sets of spatial and feature weights, thus finding static receptive fields for units, which is appropriate only for early visual areas. In this work, we employ the attention mechanism used in the transformer architecture to study how retinotopic visual features can be dynamically routed to category-selective areas in high-level visual processing. We show that this computational motif is significantly more powerful than alternative methods in predicting brain activity during natural scene viewing, across different feature basis models and modalities. We also show that this approach is inherently more interpretable as the attention-routing signals for different high-level categorical areas can be easily visualized for any input image. Given its high performance at predicting brain responses to novel images, the model deserves consideration as a candidate mechanistic model of how visual information from retinotopic maps is routed in the human brain based on the relevance of the input content to different category-selective regions.", "date": "2025-05-22", "categories": []}
{"id": "2509.04661", "title": "Flexible inference for animal learning rules using neural networks", "abstract": "Understanding how animals learn is a central challenge in neuroscience, with growing relevance to the development of animal- or human-aligned artificial intelligence. However, existing approaches tend to assume fixed parametric forms for the learning rule (e.g., Q-learning, policy gradient), which may not accurately describe the complex forms of learning employed by animals in realistic settings. Here we address this gap by developing a framework to infer learning rules directly from behavioral data collected during de novo task learning. We assume that animals follow a decision policy parameterized by a generalized linear model (GLM), and we model their learning rule -- the mapping from task covariates to per-trial weight updates -- using a deep neural network (DNN). This formulation allows flexible, data-driven inference of learning rules while maintaining an interpretable form of the decision policy itself. To capture more complex learning dynamics, we introduce a recurrent neural network (RNN) variant that relaxes the Markovian assumption that learning depends solely on covariates of the current trial, allowing for learning rules that integrate information over multiple trials. Simulations demonstrate that the framework can recover ground-truth learning rules. We applied our DNN and RNN-based methods to a large behavioral dataset from mice learning to perform a sensory decision-making task and found that they outperformed traditional RL learning rules at predicting the learning trajectories of held-out mice. The inferred learning rules exhibited reward-history-dependent learning dynamics, with larger updates following sequences of rewarded trials. Overall, these methods provide a flexible framework for inferring learning rules from behavioral data in de novo learning tasks, setting the stage for improved animal training protocols and the development of behavioral digital twins.", "date": "2025-09-04", "categories": []}
{"id": "2510.04670", "title": "Improving Multimodal Brain Encoding Model with Dynamic Subject-awareness Routing", "abstract": "Naturalistic fMRI encoding must handle multimodal inputs, shifting fusion styles, and pronounced inter-subject variability. We introduce AFIRE (Agnostic Framework for Multimodal fMRI Response Encoding), an agnostic interface that standardizes time-aligned post-fusion tokens from varied encoders, and MIND, a plug-and-play Mixture-of-Experts decoder with a subject-aware dynamic gating. Trained end-to-end for whole-brain prediction, AFIRE decouples the decoder from upstream fusion, while MIND combines token-dependent Top-K sparse routing with a subject prior to personalize expert usage without sacrificing generality. Experiments across multiple multimodal backbones and subjects show consistent improvements over strong baselines, enhanced cross-subject generalization, and interpretable expert patterns that correlate with content type. The framework offers a simple attachment point for new encoders and datasets, enabling robust, plug-and-improve performance for naturalistic neuroimaging studies.", "date": "2025-10-06", "categories": []}
{"id": "2602.04095", "title": "A computational account of dreaming: learning and memory consolidation", "abstract": "A number of studies have concluded that dreaming is mostly caused by randomly arriving internal signals because \"dream contents are random impulses\", and argued that dream sleep is unlikely to play an important part in our intellectual capacity. On the contrary, numerous functional studies have revealed that dream sleep does play an important role in our learning and other intellectual functions. Specifically, recent studies have suggested the importance of dream sleep in memory consolidation, following the findings of neural replaying of recent waking patterns in the hippocampus. The randomness has been the hurdle that divides dream theories into either functional or functionless. This study presents a cognitive and computational model of dream process. This model is simulated to perform the functions of learning and memory consolidation, which are two most popular dream functions that have been proposed. The simulations demonstrate that random signals may result in learning and memory consolidation. Thus, dreaming is proposed as a continuation of brain's waking activities that processes signals activated spontaneously and randomly from the hippocampus. The characteristics of the model are discussed and found in agreement with many characteristics concluded from various empirical studies.", "date": "2026-02-04", "categories": []}
{"id": "2602.04492", "title": "Discovering Mechanistic Models of Neural Activity: System Identification in an in Silico Zebrafish", "abstract": "Constructing mechanistic models of neural circuits is a fundamental goal of neuroscience, yet verifying such models is limited by the lack of ground truth. To rigorously test model discovery, we establish an in silico testbed using neuromechanical simulations of a larval zebrafish as a transparent ground truth. We find that LLM-based tree search autonomously discovers predictive models that significantly outperform established forecasting baselines. Conditioning on sensory drive is necessary but not sufficient for faithful system identification, as models exploit statistical shortcuts. Structural priors prove essential for enabling robust out-of-distribution generalization and recovery of interpretable mechanistic models. Our insights provide guidance for modeling real-world neural recordings and offer a broader template for AI-driven scientific discovery.", "date": "2026-02-04", "categories": []}
{"id": "2602.04512", "title": "BrainVista: Modeling Naturalistic Brain Dynamics as Multimodal Next-Token Prediction", "abstract": "Naturalistic fMRI characterizes the brain as a dynamic predictive engine driven by continuous sensory streams. However, modeling the causal forward evolution in realistic neural simulation is impeded by the timescale mismatch between multimodal inputs and the complex topology of cortical networks. To address these challenges, we introduce BrainVista, a multimodal autoregressive framework designed to model the causal evolution of brain states. BrainVista incorporates Network-wise Tokenizers to disentangle system-specific dynamics and a Spatial Mixer Head that captures inter-network information flow without compromising functional boundaries. Furthermore, we propose a novel Stimulus-to-Brain (S2B) masking mechanism to synchronize high-frequency sensory stimuli with hemodynamically filtered signals, enabling strict, history-only causal conditioning. We validate our framework on Algonauts 2025, CineBrain, and HAD, achieving state-of-the-art fMRI encoding performance. In long-horizon rollout settings, our model yields substantial improvements over baselines, increasing pattern correlation by 36.0\\% and 33.3\\% on relative to the strongest baseline Algonauts 2025 and CineBrain, respectively.", "date": "2026-02-04", "categories": []}
{"id": "2506.04536", "title": "NOBLE -- Neural Operator with Biologically-informed Latent Embeddings to Capture Experimental Variability in Biological Neuron Models", "abstract": "Characterizing the cellular properties of neurons is fundamental to understanding their function in the brain. In this quest, the generation of bio-realistic models is central towards integrating multimodal cellular data sets and establishing causal relationships. However, current modeling approaches remain constrained by the limited availability and intrinsic variability of experimental neuronal data. The deterministic formalism of bio-realistic models currently precludes accounting for the natural variability observed experimentally. While deep learning is becoming increasingly relevant in this space, it fails to capture the full biophysical complexity of neurons, their nonlinear voltage dynamics, and variability. To address these shortcomings, we introduce NOBLE, a neural operator framework that learns a mapping from a continuous frequency-modulated embedding of interpretable neuron features to the somatic voltage response induced by current injection. Trained on synthetic data generated from bio-realistic neuron models, NOBLE predicts distributions of neural dynamics accounting for the intrinsic experimental variability. Unlike conventional bio-realistic neuron models, interpolating within the embedding space offers models whose dynamics are consistent with experimentally observed responses. NOBLE enables the efficient generation of synthetic neurons that closely resemble experimental data and exhibit trial-to-trial variability, offering a $4200\\times$ speedup over the numerical solver. NOBLE is the first scaled-up deep learning framework that validates its generalization with real experimental data. To this end, NOBLE captures fundamental neural properties in a unique and emergent manner that opens the door to a better understanding of cellular composition and computations, neuromorphic architectures, large-scale brain circuits, and general neuroAI applications.", "date": "2025-06-05", "categories": []}
{"id": "2508.15784", "title": "Emergent time-keeping mechanisms in a deep reinforcement learning agent performing an interval timing task", "abstract": "Drawing parallels between Deep Artificial Neural Networks (DNNs) and biological systems can aid in understanding complex biological mechanisms that are difficult to disentangle. Temporal processing, an extensively researched topic, is one such example that lacks a coherent understanding of its underlying mechanisms. In this study, we investigate temporal processing in a Deep Reinforcement Learning (DRL) agent performing an interval timing task and explore potential biological counterparts to its emergent behavior. The agent was successfully trained to perform a duration production task, which involved marking successive occurrences of a target interval while viewing a video sequence. Analysis of the agent's internal states revealed oscillatory neural activations, a ubiquitous pattern in biological systems. Interestingly, the agent's actions were predominantly influenced by neurons exhibiting these oscillations with high amplitudes and frequencies corresponding to the target interval. Parallels are drawn between the agent's time-keeping strategy and the Striatal Beat Frequency (SBF) model, a biologically plausible model of interval timing. Furthermore, the agent maintained its oscillatory representations and task performance when tested on different video sequences (including a blank video). Thus, once learned, the agent internalized its time-keeping mechanism and showed minimal reliance on its environment to perform the timing task. A hypothesis about the resemblance between this emergent behavior and certain aspects of the evolution of biological processes like circadian rhythms, has been discussed. This study aims to contribute to recent research efforts of utilizing DNNs to understand biological systems, with a particular emphasis on temporal processing.", "date": "2025-08-06", "categories": []}
{"id": "2601.13570", "title": "GeoDynamics: A Geometric State-Space Neural Network for Understanding Brain Dynamics on Riemannian Manifolds", "abstract": "State-space models (SSMs) have become a cornerstone for unraveling brain dynamics, revealing how latent neural states evolve over time and give rise to observed signals. By combining the flexibility of deep learning with the principled dynamical structure of SSMs, recent studies have achieved powerful fits to functional neuroimaging data. However, most existing approaches still view the brain as a set of loosely connected regions or impose oversimplified network priors, falling short of a truly holistic and self-organized dynamical system perspective. Brain functional connectivity (FC) at each time point naturally forms a symmetric positive definite (SPD) matrix, which resides on a curved Riemannian manifold rather than in Euclidean space. Capturing the trajectories of these SPD matrices is key to understanding how coordinated networks support cognition and behavior. To this end, we introduce GeoDynamics, a geometric state-space neural network that tracks latent brain-state trajectories directly on the high-dimensional SPD manifold. GeoDynamics embeds each connectivity matrix into a manifold-aware recurrent framework, learning smooth and geometry-respecting transitions that reveal task-driven state changes and early markers of Alzheimer's disease, Parkinson's disease, and autism. Beyond neuroscience, we validate GeoDynamics on human action recognition benchmarks (UTKinect, Florence, HDM05), demonstrating its scalability and robustness in modeling complex spatiotemporal dynamics across diverse domains.", "date": "2026-01-20", "categories": []}
{"id": "2601.20138", "title": "Scaling Next-Brain-Token Prediction for MEG", "abstract": "We present a large autoregressive model for source-space MEG that scales next-token prediction to long context across datasets and scanners: handling a corpus of over 500 hours and thousands of sessions across the three largest MEG datasets. A modified SEANet-style vector-quantizer reduces multichannel MEG into a flattened token stream on which we train a Qwen2.5-VL backbone from scratch to predict the next brain token and to recursively generate minutes of MEG from up to a minute of context. To evaluate long-horizon generation, we introduce task-matched tests: (i) on-manifold stability via generated-only drift compared to the time-resolved distribution of real sliding windows, and (ii) conditional specificity via correct context versus prompt-swap controls using a neurophysiologically grounded metric set. We train on CamCAN and Omega and run all analyses on held-out MOUS, establishing cross-dataset generalization. Across metrics, generations remain relatively stable over long rollouts and are closer to the correct continuation than swapped controls. Code available at: https://github.com/ricsinaruto/brain-gen.", "date": "2026-01-28", "categories": []}
{"id": "2601.21148", "title": "BrainStack: Neuro-MoE with Functionally Guided Expert Routing for EEG-Based Language Decoding", "abstract": "Decoding linguistic information from electroencephalography (EEG) remains challenging due to the brain's distributed and nonlinear organization. We present BrainStack, a functionally guided neuro-mixture-of-experts (Neuro-MoE) framework that models the brain's modular functional architecture through anatomically partitioned expert networks. Each functional region is represented by a specialized expert that learns localized neural dynamics, while a transformer-based global expert captures cross-regional dependencies. A learnable routing gate adaptively aggregates these heterogeneous experts, enabling context-dependent expert coordination and selective fusion. To promote coherent representation across the hierarchy, we introduce cross-regional distillation, where the global expert provides top-down regularization to the regional experts. We further release SilentSpeech-EEG (SS-EEG), a large-scale benchmark comprising over 120 hours of EEG recordings from 12 subjects performing 24 silent words, the largest dataset of its kind. Experiments demonstrate that BrainStack consistently outperforms state-of-the-art models, achieving superior accuracy and generalization across subjects. Our results establish BrainStack as a functionally modular, neuro-inspired MoE paradigm that unifies neuroscientific priors with adaptive expert routing, paving the way for scalable and interpretable brain-language decoding.", "date": "2026-01-29", "categories": []}
{"id": "2601.20447", "title": "Assembling the Mind's Mosaic: Towards EEG Semantic Intent Decoding", "abstract": "Enabling natural communication through brain-computer interfaces (BCIs) remains one of the most profound challenges in neuroscience and neurotechnology. While existing frameworks offer partial solutions, they are constrained by oversimplified semantic representations and a lack of interpretability. To overcome these limitations, we introduce Semantic Intent Decoding (SID), a novel framework that translates neural activity into natural language by modeling meaning as a flexible set of compositional semantic units. SID is built on three core principles: semantic compositionality, continuity and expandability of semantic space, and fidelity in reconstruction. We present BrainMosaic, a deep learning architecture implementing SID. BrainMosaic decodes multiple semantic units from EEG/SEEG signals using set matching and then reconstructs coherent sentences through semantic-guided reconstruction. This approach moves beyond traditional pipelines that rely on fixed-class classification or unconstrained generation, enabling a more interpretable and expressive communication paradigm. Extensive experiments on multilingual EEG and clinical SEEG datasets demonstrate that SID and BrainMosaic offer substantial advantages over existing frameworks, paving the way for natural and effective BCI-mediated communication.", "date": "2026-01-28", "categories": []}
{"id": "2411.07121", "title": "Decoding Visual Experience and Mapping Semantics through Whole-Brain Analysis Using fMRI Foundation Models", "abstract": "Neural decoding, the process of understanding how brain activity corresponds to different stimuli, has been a primary objective in cognitive sciences. Over the past three decades, advances in functional Magnetic Resonance Imaging (fMRI) and machine learning have greatly improved our ability to map visual stimuli to brain activity, especially in the visual cortex. Concurrently, research has expanded to decode more complex processes, such as language and memory across the whole brain, using techniques to handle greater variability and improve signal accuracy. We argue that \"seeing\" involves more than just mapping visual stimuli onto the visual cortex; it engages the entire brain, as various emotions and cognitive states can emerge from observing different scenes. In this paper, we develop algorithms to enhance our understanding of visual processes by incorporating whole-brain activation maps while individuals are exposed to visual stimuli. We utilize transformer-based large-scale fMRI encoders and Image generative models (encoders & decoders) pre-trained on large public datasets, which are then fine-tuned through Image-fMRI contrastive learning. Our models can decode visual experience across the entire cerebral cortex, surpassing the traditional confines of the visual cortex. Using a public dataset (BOLD5000), we first compare our method with state-of-the-art approaches for decoding visual processing and show improved predictive semantic accuracy by 43%. A network ablation analysis suggests that, beyond the visual cortex, the default mode network contributes significantly to stimulus decoding, in line with the proposed role of this network in sense-making and semantic processing.", "date": "2024-11-11", "categories": []}
{"id": "2506.11062", "title": "Decoding Cortical Microcircuits: A Generative Model for Latent Space Exploration and Controlled Synthesis", "abstract": "A central idea in understanding brains and building artificial intelligence is that structure determines function. Yet, how the brain's complex structure arises from a limited set of genetic instructions remains a key question. The ultra high-dimensional detail of neural connections vastly exceeds the information storage capacity of genes, suggesting a compact, low-dimensional blueprint must guide brain development. Our motivation is to uncover this blueprint. We introduce a generative model, to learn this underlying representation from detailed connectivity maps of mouse cortical microcircuits. Our model successfully captures the essential structural information of these circuits in a compressed latent space. We found that specific, interpretable directions within this space directly relate to understandable network properties. Building on this, we demonstrate a novel method to controllably generate new, synthetic microcircuits with desired structural features by navigating this latent space. This work offers a new way to investigate the design principles of neural circuits and explore how structure gives rise to function, potentially informing the development of more advanced artificial neural networks.", "date": "2025-05-29", "categories": []}
{"id": "2601.18792", "title": "MEGnifying Emotion: Sentiment Analysis from Annotated Brain Data", "abstract": "Decoding emotion from brain activity could unlock a deeper understanding of the human experience. While a number of existing datasets align brain data with speech and with speech transcripts, no datasets have annotated brain data with sentiment. To bridge this gap, we explore the use of pre-trained Text-to-Sentiment models to annotate non invasive brain recordings, acquired using magnetoencephalography (MEG), while participants listened to audiobooks. Having annotated the text, we employ force-alignment of the text and audio to align our sentiment labels with the brain recordings. It is straightforward then to train Brainto-Sentiment models on these data. Experimental results show an improvement in balanced accuracy for Brain-to-Sentiment compared to baseline, supporting the proposed approach as a proof-of-concept for leveraging existing MEG datasets and learning to decode sentiment directly from the brain.", "date": "2026-01-26", "categories": []}
{"id": "2407.17914", "title": "Vision-Language Models Align with Human Neural Representations in Concept Processing", "abstract": "Recent studies suggest that transformer-based vision-language models (VLMs) capture the multimodality of concept processing in the human brain. However, a systematic evaluation exploring different types of VLM architectures and the role played by visual and textual context is still lacking. Here, we analyse multiple VLMs employing different strategies to integrate visual and textual modalities, along with language-only counterparts. We measure the alignment between concept representations by models and existing (fMRI) brain responses to concept words presented in two experimental conditions, where either visual (pictures) or textual (sentences) context is provided. Our results reveal that VLMs outperform the language-only counterparts in both experimental conditions. However, controlled ablation studies show that only for some VLMs, such as LXMERT and IDEFICS2, brain alignment stems from genuinely learning more human-like concepts during pretraining, while others are highly sensitive to the context provided at inference. Additionally, we find that vision-language encoders are more brain-aligned than more recent, generative VLMs. Altogether, our study shows that VLMs align with human neural representations in concept processing, while highlighting differences among architectures. We open-source code and materials to reproduce our experiments at: https://github.com/dmg-illc/vl-concept-processing.", "date": "2024-07-25", "categories": []}
{"id": "2506.03088", "title": "Modelling the Effects of Hearing Loss on Neural Coding in the Auditory Midbrain with Variational Conditioning", "abstract": "The mapping from sound to neural activity that underlies hearing is highly non-linear. The first few stages of this mapping in the cochlea have been modelled successfully, with biophysical models built by hand and, more recently, with DNN models trained on datasets simulated by biophysical models. Modelling the auditory brain has been a challenge because central auditory processing is too complex for models to be built by hand, and datasets for training DNN models directly have not been available. Recent work has taken advantage of large-scale high resolution neural recordings from the auditory midbrain to build a DNN model of normal hearing with great success. But this model assumes that auditory processing is the same in all brains, and therefore it cannot capture the widely varying effects of hearing loss.\n  We propose a novel variational-conditional model to learn to encode the space of hearing loss directly from recordings of neural activity in the auditory midbrain of healthy and noise exposed animals. With hearing loss parametrised by only 6 free parameters per animal, our model accurately predicts 62% of the explainable variance in neural responses from normal hearing animals and 68% for hearing impaired animals, within a few percentage points of state of the art animal specific models. We demonstrate that the model can be used to simulate realistic activity from out of sample animals by fitting only the learned conditioning parameters with Bayesian optimisation, achieving crossentropy loss within 2% of the optimum in 15-30 iterations. Including more animals in the training data slightly improved the performance on unseen animals. This model will enable future development of parametrised hearing loss compensation models trained to directly restore normal neural coding in hearing impaired brains, which can be quickly fitted for a new user by human in the loop optimisation.", "date": "2025-06-03", "categories": []}
{"id": "2507.10383", "title": "Dynamical stability for dense patterns in discrete attractor neural networks", "abstract": "Neural networks storing multiple discrete attractors are canonical models of biological memory. Previously, the dynamical stability of such networks could only be guaranteed under highly restrictive conditions. Here, we derive a theory of the local stability of discrete fixed points in a broad class of networks with graded neural activities and in the presence of noise. By directly analyzing the bulk and the outliers of the Jacobian spectrum, we show that all fixed points are stable below a critical load that is distinct from the classical \\textit{critical capacity} and depends on the statistics of neural activities in the fixed points as well as the single-neuron activation function. Our analysis highlights the computational benefits of threshold-linear activation and sparse-like patterns.", "date": "2025-07-14", "categories": []}
{"id": "2601.15336", "title": "Learning Discrete Successor Transitions in Continuous Attractor Networks: Emergence, Limits, and Topological Constraints", "abstract": "Continuous attractor networks (CANs) are a well-established class of models for representing low-dimensional continuous variables such as head direction, spatial position, and phase. In canonical spatial domains, transitions along the attractor manifold are driven by continuous displacement signals, such as angular velocity-provided by sensorimotor systems external to the CAN itself. When such signals are not explicitly provided as dedicated displacement inputs, it remains unclear whether attractor-based circuits can reliably acquire recurrent dynamics that support stable state transitions, or whether alternative predictive strategies dominate.\n  In this work, we present an experimental framework for training CANs to perform successor-like transitions between stable attractor states in the absence of externally provided displacement signals. We compare two recurrent topologies, a circular ring and a folded snake manifold, and systematically vary the temporal regime under which stability is evaluated. We find that, under short evaluation windows, networks consistently converge to impulse-driven associative solutions that achieve high apparent accuracy yet lack persistent attractor dynamics. Only when stability is explicitly enforced over extended free-run periods do genuine attractor-based transition dynamics emerge. This suggests that shortcut solutions are the default outcome of local learning in recurrent networks, while attractor dynamics represent a constrained regime rather than a generic result.\n  Furthermore, we demonstrate that topology strictly limits the capacity for learned transitions. While the continuous ring topology achieves perfect stability over long horizons, the folded snake topology hits a geometric limit characterized by failure at manifold discontinuities, which neither curriculum learning nor basal ganglia-inspired gating can fully overcome.", "date": "2026-01-20", "categories": []}
{"id": "2409.02684", "title": "Neural timescales from a computational perspective", "abstract": "Neural activity fluctuates over a wide range of timescales within and across brain areas. Experimental observations suggest that diverse neural timescales reflect information in dynamic environments. However, how timescales are defined and measured from brain recordings vary across the literature. Moreover, these observations do not specify the mechanisms underlying timescale variations, nor whether specific timescales are necessary for neural computation and brain function. Here, we synthesize three directions where computational approaches can distill the broad set of empirical observations into quantitative and testable theories: We review (i) how different data analysis methods quantify timescales across distinct behavioral states and recording modalities, (ii) how biophysical models provide mechanistic explanations for the emergence of diverse timescales, and (iii) how task-performing networks and machine learning models uncover the functional relevance of neural timescales. This integrative computational perspective thus complements experimental investigations, providing a holistic view on how neural timescales reflect the relationship between brain structure, dynamics, and behavior.", "date": "2024-09-04", "categories": []}
{"id": "2601.12577", "title": "Primate-like perceptual decision making emerges through deep recurrent reinforcement learning", "abstract": "Progress has led to a detailed understanding of the neural mechanisms that underlie decision making in primates. However, less is known about why such mechanisms are present in the first place. Theory suggests that primate decision making mechanisms, and their resultant behavioral abilities, emerged to maximize reward in the face of noisy, temporally evolving information. To test this theory, we trained an end-to-end deep recurrent neural network using reinforcement learning on a noisy perceptual discrimination task. Networks learned several key abilities of primate-like decision making including trading off speed for accuracy, and flexibly changing their mind in the face of new information. Internal dynamics of these networks suggest that these abilities were supported by similar decision mechanisms as those observed in primate neurophysiological studies. These results provide experimental support for key pressures that gave rise to the primate ability to make flexible decisions.", "date": "2026-01-18", "categories": []}
{"id": "2601.11108", "title": "Simple Models, Rich Representations: Visual Decoding from Primate Intracortical Neural Signals", "abstract": "Understanding how neural activity gives rise to perception is a central challenge in neuroscience. We address the problem of decoding visual information from high-density intracortical recordings in primates, using the THINGS Ventral Stream Spiking Dataset. We systematically evaluate the effects of model architecture, training objectives, and data scaling on decoding performance. Results show that decoding accuracy is mainly driven by modeling temporal dynamics in neural signals, rather than architectural complexity. A simple model combining temporal attention with a shallow MLP achieves up to 70% top-1 image retrieval accuracy, outperforming linear baselines as well as recurrent and convolutional approaches. Scaling analyses reveal predictable diminishing returns with increasing input dimensionality and dataset size. Building on these findings, we design a modular generative decoding pipeline that combines low-resolution latent reconstruction with semantically conditioned diffusion, generating plausible images from 200 ms of brain activity. This framework provides principles for brain-computer interfaces and semantic neural decoding.", "date": "2026-01-16", "categories": []}
{"id": "2601.06257", "title": "Gamma2Patterns: Deep Cognitive Attention Region Identification and Gamma-Alpha Pattern Analysis", "abstract": "Deep cognitive attention is characterized by heightened gamma oscillations and coordinated visual behavior. Despite the physiological importance of these mechanisms, computational studies rarely synthesize these modalities or identify the neural regions most responsible for sustained focus. To address this gap, this work introduces Gamma2Patterns, a multimodal framework that characterizes deep cognitive attention by leveraging complementary Gamma and Alpha band EEG activity alongside Eye-tracking measurements. Using the SEED-IV dataset [1], we extract spectral power, burst-based temporal dynamics, and fixation-saccade-pupil signals across 62 channels or electrodes to analyze how neural activation differs between high-focus (Gamma-dominant) and low-focus (Alpha-dominant) states. Our findings reveal that frontopolar, temporal, anterior frontal, and parieto-occipital regions exhibit the strongest Gamma power and burst rates, indicating their dominant role in deep attentional engagement, while Eye-tracking signals confirm complementary contributions from frontal, frontopolar, and frontotemporal regions. Furthermore, we show that Gamma power and burst duration provide more discriminative markers of deep focus than Alpha power alone, demonstrating their value for attention decoding. Collectively, these results establish a multimodal, evidence-based map of cortical regions and oscillatory signatures underlying deep focus, providing a neurophysiological foundation for future brain-inspired attention mechanisms in AI systems.", "date": "2026-01-09", "categories": []}
{"id": "2601.03392", "title": "Better, But Not Sufficient: Testing Video ANNs Against Macaque IT Dynamics", "abstract": "Feedforward artificial neural networks (ANNs) trained on static images remain the dominant models of the the primate ventral visual stream, yet they are intrinsically limited to static computations. The primate world is dynamic, and the macaque ventral visual pathways, specifically the inferior temporal (IT) cortex not only supports object recognition but also encodes object motion velocity during naturalistic video viewing. Does IT's temporal responses reflect nothing more than time-unfolded feedforward transformations, framewise features with shallow temporal pooling, or do they embody richer dynamic computations? We tested this by comparing macaque IT responses during naturalistic videos against static, recurrent, and video-based ANN models. Video models provided modest improvements in neural predictivity, particularly at later response stages, raising the question of what kind of dynamics they capture. To probe this, we applied a stress test: decoders trained on naturalistic videos were evaluated on \"appearance-free\" variants that preserve motion but remove shape and texture. IT population activity generalized across this manipulation, but all ANN classes failed. Thus, current video models better capture appearance-bound dynamics rather than the appearance-invariant temporal computations expressed in IT, underscoring the need for new objectives that encode biological temporal statistics and invariances.", "date": "2026-01-06", "categories": []}
{"id": "2601.02636", "title": "Credit Assignment via Neural Manifold Noise Correlation", "abstract": "Credit assignment--how changes in individual neurons and synapses affect a network's output--is central to learning in brains and machines. Noise correlation, which estimates gradients by correlating perturbations of activity with changes in output, provides a biologically plausible solution to credit assignment but scales poorly as accurately estimating the Jacobian requires that the number of perturbations scale with network size. Moreover, isotropic noise conflicts with neurobiological observations that neural activity lies on a low-dimensional manifold. To address these drawbacks, we propose neural manifold noise correlation (NMNC), which performs credit assignment using perturbations restricted to the neural manifold. We show theoretically and empirically that the Jacobian row space aligns with the neural manifold in trained networks, and that manifold dimensionality scales slowly with network size. NMNC substantially improves performance and sample efficiency over vanilla noise correlation in convolutional networks trained on CIFAR-10, ImageNet-scale models, and recurrent networks. NMNC also yields representations more similar to the primate visual system than vanilla noise correlation. These findings offer a mechanistic hypothesis for how biological circuits could support credit assignment, and suggest that biologically inspired constraints may enable, rather than limit, effective learning at scale.", "date": "2026-01-06", "categories": []}
{"id": "2601.03117", "title": "Transformers self-organize like newborn visual systems when trained in prenatal worlds", "abstract": "Do transformers learn like brains? A key challenge in addressing this question is that transformers and brains are trained on fundamentally different data. Brains are initially \"trained\" on prenatal sensory experiences (e.g., retinal waves), whereas transformers are typically trained on large datasets that are not biologically plausible. We reasoned that if transformers learn like brains, then they should develop the same structure as newborn brains when exposed to the same prenatal data. To test this prediction, we simulated prenatal visual input using a retinal wave generator. Then, using self-supervised temporal learning, we trained transformers to adapt to those retinal waves. During training, the transformers spontaneously developed the same structure as newborn visual systems: (1) early layers became sensitive to edges, (2) later layers became sensitive to shapes, and (3) the models developed larger receptive fields across layers. The organization of newborn visual systems emerges spontaneously when transformers adapt to a prenatal visual world. This developmental convergence suggests that brains and transformers learn in common ways and follow the same general fitting principles.", "date": "2026-01-06", "categories": []}
{"id": "2511.20692", "title": "The Human Brain as a Combinatorial Complex", "abstract": "We propose a framework for constructing combinatorial complexes (CCs) from fMRI time series data that captures both pairwise and higher-order neural interactions through information-theoretic measures, bridging topological deep learning and network neuroscience. Current graph-based representations of brain networks systematically miss the higher-order dependencies that characterize neural complexity, where information processing often involves synergistic interactions that cannot be decomposed into pairwise relationships. Unlike topological lifting approaches that map relational structures into higher-order domains, our method directly constructs CCs from statistical dependencies in the data. Our CCs generalize graphs by incorporating higher-order cells that represent collective dependencies among brain regions, naturally accommodating the multi-scale, hierarchical nature of neural processing. The framework constructs data-driven combinatorial complexes using O-information and S-information measures computed from fMRI signals, preserving both pairwise connections and higher-order cells (e.g., triplets, quadruplets) based on synergistic dependencies. Using NetSim simulations as a controlled proof-of-concept dataset, we demonstrate our CC construction pipeline and show how both pairwise and higher-order dependencies in neural time series can be quantified and represented within a unified structure. This work provides a framework for brain network representation that preserves fundamental higher-order structure invisible to traditional graph methods, and enables the application of topological deep learning (TDL) architectures to neural data.", "date": "2025-11-22", "categories": []}
{"id": "2601.01599", "title": "From Theory of Mind to Theory of Environment: Counterfactual Simulation of Latent Environmental Dynamics", "abstract": "The vertebrate motor system employs dimensionality-reducing strategies to limit the complexity of movement coordination, for efficient motor control. But when environments are dense with hidden action-outcome contingencies, movement complexity can promote behavioral innovation. Humans, perhaps uniquely, may infer the presence of hidden environmental dynamics from social cues, by drawing upon computational mechanisms shared with Theory of Mind. This proposed \"Theory of Environment\" supports behavioral innovation by expanding the dimensionality of motor exploration.", "date": "2026-01-04", "categories": []}
{"id": "2507.11783", "title": "EEG Foundation Models: A Critical Review of Current Progress and Future Directions", "abstract": "Premise. Patterns of electrical brain activity recorded via electroencephalography (EEG) offer immense value for scientific and clinical investigations. The inability of supervised EEG encoders to learn robust EEG patterns and their over-reliance on expensive signal annotations have sparked a transition towards general-purpose self-supervised EEG encoders, i.e., EEG foundation models (EEG-FMs), for robust and scalable EEG feature extraction. However, the real-world readiness of early EEG-FMs and the rubrics for long-term research progress remain unclear. Objective. In this work, we conduct a review of ten early EEG-FMs to capture common trends and identify key directions for future development of EEG-FMs. Methods. We comparatively analyze each EEG-FM using three fundamental pillars of foundation modeling, namely the representation of input data, self-supervised modeling, and the evaluation strategy. Based on this analysis, we present a critical synthesis of EEG-FM methodology, empirical findings, and outstanding research gaps. Results. We find that most EEG-FMs adopt a sequence-based modeling scheme that relies on transformer-based backbones and the reconstruction of masked temporal EEG sequences for self-supervision. However, model evaluations remain heterogeneous and largely limited, making it challenging to assess their practical off-the-shelf utility. In addition to adopting standardized and realistic evaluations, future work should demonstrate more substantial scaling effects and make principled and trustworthy choices throughout the EEG representation learning pipeline. Significance. Our review indicates that the development of benchmarks, software tools, technical methodologies, and applications in collaboration with domain experts may advance the translational utility and real-world adoption of EEG-FMs.", "date": "2025-07-15", "categories": []}
{"id": "2512.20929", "title": "Decoding Predictive Inference in Visual Language Processing via Spatiotemporal Neural Coherence", "abstract": "Human language processing relies on the brain's capacity for predictive inference. We present a machine learning framework for decoding neural (EEG) responses to dynamic visual language stimuli in Deaf signers. Using coherence between neural signals and optical flow-derived motion features, we construct spatiotemporal representations of predictive neural dynamics. Through entropy-based feature selection, we identify frequency-specific neural signatures that differentiate interpretable linguistic input from linguistically disrupted (time-reversed) stimuli. Our results reveal distributed left-hemispheric and frontal low-frequency coherence as key features in language comprehension, with experience-dependent neural signatures correlating with age. This work demonstrates a novel multimodal approach for probing experience-driven generative models of perception in the brain.", "date": "2025-12-24", "categories": []}
{"id": "2502.03198", "title": "SimSort: A Data-Driven Framework for Spike Sorting by Large-Scale Electrophysiology Simulation", "abstract": "Spike sorting is an essential process in neural recording, which identifies and separates electrical signals from individual neurons recorded by electrodes in the brain, enabling researchers to study how specific neurons communicate and process information. Although there exist a number of spike sorting methods which have contributed to significant neuroscientific breakthroughs, many are heuristically designed, making it challenging to verify their correctness due to the difficulty of obtaining ground truth labels from real-world neural recordings. In this work, we explore a data-driven, deep learning-based approach. We begin by creating a large-scale dataset through electrophysiology simulations using biologically realistic computational models. We then present SimSort, a pretraining framework for spike sorting. Trained solely on simulated data, SimSort demonstrates zero-shot generalizability to real-world spike sorting tasks, yielding consistent improvements over existing methods across multiple benchmarks. These results highlight the potential of simulation-driven pretraining to enhance the robustness and scalability of spike sorting in experimental neuroscience.", "date": "2025-02-05", "categories": []}
{"id": "2512.19155", "title": "Can We Test Consciousness Theories on AI? Ablations, Markers, and Robustness", "abstract": "The search for reliable indicators of consciousness has fragmented into competing theoretical camps (Global Workspace Theory (GWT), Integrated Information Theory (IIT), and Higher-Order Theories (HOT)), each proposing distinct neural signatures. We adopt a synthetic neuro-phenomenology approach: constructing artificial agents that embody these mechanisms to test their functional consequences through precise architectural ablations impossible in biological systems. Across three experiments, we report dissociations suggesting these theories describe complementary functional layers rather than competing accounts. In Experiment 1, a no-rewire Self-Model lesion abolishes metacognitive calibration while preserving first-order task performance, yielding a synthetic blindsight analogue consistent with HOT predictions. In Experiment 2, workspace capacity proves causally necessary for information access: a complete workspace lesion produces qualitative collapse in access-related markers, while partial reductions show graded degradation, consistent with GWT's ignition framework. In Experiment 3, we uncover a broadcast-amplification effect: GWT-style broadcasting amplifies internal noise, creating extreme fragility. The B2 agent family is robust to the same latent perturbation; this robustness persists in a Self-Model-off / workspace-read control, cautioning against attributing the effect solely to $z_{\\text{self}}$ compression. We also report an explicit negative result: raw perturbational complexity (PCI-A) decreases under the workspace bottleneck, cautioning against naive transfer of IIT-adjacent proxies to engineered agents. These results suggest a hierarchical design principle: GWT provides broadcast capacity, while HOT provides quality control. We emphasize that our agents are not conscious; they are reference implementations for testing functional predictions of consciousness theories.", "date": "2025-12-22", "categories": []}
{"id": "2503.02303", "title": "Flexible Prefrontal Control over Hippocampal Episodic Memory for Goal-Directed Generalization", "abstract": "Many tasks require flexibly modifying perception and behavior based on current goals. Humans can retrieve episodic memories from days to years ago, using them to contextualize and generalize behaviors across novel but structurally related situations. The brain's ability to control episodic memories based on task demands is often attributed to interactions between the prefrontal cortex (PFC) and hippocampus (HPC). We propose a reinforcement learning model that incorporates a PFC-HPC interaction mechanism for goal-directed generalization. In our model, the PFC learns to generate query-key representations to encode and retrieve goal-relevant episodic memories, modulating HPC memories top-down based on current task demands. Moreover, the PFC adapts its encoding and retrieval strategies dynamically when faced with multiple goals presented in a blocked, rather than interleaved, manner. Our results show that: (1) combining working memory with selectively retrieved episodic memory allows transfer of decisions among similar environments or situations, (2) top-down control from PFC over HPC improves learning of arbitrary structural associations between events for generalization to novel environments compared to a bottom-up sensory-driven approach, and (3) the PFC encodes generalizable representations during both encoding and retrieval of goal-relevant memories, whereas the HPC exhibits event-specific representations. Together, these findings highlight the importance of goal-directed prefrontal control over hippocampal episodic memory for decision-making in novel situations and suggest a computational mechanism by which PFC-HPC interactions enable flexible behavior.", "date": "2025-03-04", "categories": []}
{"id": "2507.03631", "title": "Scientific Machine Learning of Chaotic Systems Discovers Governing Equations for Neural Populations", "abstract": "Discovering governing equations that describe complex chaotic systems remains a fundamental challenge in physics and neuroscience. Here, we introduce the PEM-UDE method, which combines the prediction-error method with universal differential equations to extract interpretable mathematical expressions from chaotic dynamical systems, even with limited or noisy observations. This approach succeeds where traditional techniques fail by smoothing optimization landscapes and removing the chaotic properties during the fitting process without distorting optimal parameters. We demonstrate its efficacy by recovering hidden states in the Rossler system and reconstructing dynamics from noise-corrupted electrical-circuit data, in which the correct functional form of the dynamics is recovered even when one of the observed time series is corrupted by noise 5x the magnitude of the true signal. We demonstrate that this method can recover the correct dynamics, whereas direct symbolic regression methods, such as STLSQ, fail to do so with the available data and noise. Importantly, when applied to neural populations, our method derives novel governing equations that respect biological constraints such as network sparsity - a constraint necessary for cortical information processing yet not captured in next-generation neural mass models - while preserving microscale neuronal parameters. These equations predict an emergent relationship between connection density and both oscillation frequency and synchrony in neural circuits. We validate these predictions using three intracranial electrode recording datasets from the medial entorhinal cortex, prefrontal cortex, and orbitofrontal cortex. Our work provides a pathway to develop mechanistic, multi-scale brain models that generalize across diverse neural architectures, bridging the gap between single-neuron dynamics and macroscale brain activity.", "date": "2025-07-04", "categories": []}
{"id": "2512.12767", "title": "Random matrix theory of sparse neuronal networks with heterogeneous timescales", "abstract": "Training recurrent neuronal networks consisting of excitatory (E) and inhibitory (I) units with additive noise for working memory computation slows and diversifies inhibitory timescales, leading to improved task performance that is attributed to emergent marginally stable equilibria [PNAS 122 (2025) e2316745122]. Yet the link between trained network characteristics and their roles in shaping desirable dynamical landscapes remains unexplored. Here, we investigate the Jacobian matrices describing the dynamics near these equilibria and show that they are sparse, non-Hermitian rectangular-block matrices modified by heterogeneous synaptic decay timescales and activation-function gains. We specify a random matrix ensemble that faithfully captures the spectra of trained Jacobian matrices, arising from the inhibitory core - excitatory periphery network motif (pruned E weights, broadly distributed I weights) observed post-training. An analytic theory of this ensemble is developed using statistical field theory methods: a Hermitized resolvent representation of the spectral density processed with a supersymmetry-based treatment in the style of Fyodorov and Mirlin. In this manner, an analytic description of the spectral edge is obtained, relating statistical parameters of the Jacobians (sparsity, weight variances, E/I ratio, and the distributions of timescales and gains) to near-critical features of the equilibria essential for robust working memory computation.", "date": "2025-12-14", "categories": []}
{"id": "2512.15830", "title": "From Minutes to Days: Scaling Intracranial Speech Decoding with Supervised Pretraining", "abstract": "Decoding speech from brain activity has typically relied on limited neural recordings collected during short and highly controlled experiments. Here, we introduce a framework to leverage week-long intracranial and audio recordings from patients undergoing clinical monitoring, effectively increasing the training dataset size by over two orders of magnitude. With this pretraining, our contrastive learning model substantially outperforms models trained solely on classic experimental data, with gains that scale log-linearly with dataset size. Analysis of the learned representations reveals that, while brain activity represents speech features, its global structure largely drifts across days, highlighting the need for models that explicitly account for cross-day variability. Overall, our approach opens a scalable path toward decoding and modeling brain representations in both real-life and controlled task settings.", "date": "2025-12-17", "categories": []}
{"id": "2406.07151", "title": "CrossPT-EEG: A Benchmark for Cross-Participant and Cross-Time Generalization of EEG-based Visual Decoding", "abstract": "Exploring brain activity in relation to visual perception provides insights into the biological representation of the world. While functional magnetic resonance imaging (fMRI) and magnetoencephalography (MEG) have enabled effective image classification and reconstruction, their high cost and bulk limit practical use. Electroencephalography (EEG), by contrast, offers low cost and excellent temporal resolution, but its potential has been limited by the scarcity of large, high-quality datasets and by block-design experiments that introduce temporal confounds. To fill this gap, we present CrossPT-EEG, a benchmark for cross-participant and cross-time generalization of visual decoding from EEG. We collected EEG data from 16 participants while they viewed 4,000 images sampled from ImageNet, with image stimuli annotated at multiple levels of granularity. Our design includes two stages separated in time to allow cross-time generalization and avoid block-design artifacts. We also introduce benchmarks tailored to non-block design classification, as well as pre-training experiments to assess cross-time and cross-participant generalization. These findings highlight the dataset's potential to enhance EEG-based visual brain-computer interfaces, deepen our understanding of visual perception in biological systems, and suggest promising applications for improving machine vision models.", "date": "2024-06-11", "categories": []}
{"id": "2512.13724", "title": "Graph AI generates neurological hypotheses validated in molecular, organoid, and clinical systems", "abstract": "Neurological diseases are the leading global cause of disability, yet most lack disease-modifying treatments. We present PROTON, a heterogeneous graph transformer that generates testable hypotheses across molecular, organoid, and clinical systems. To evaluate PROTON, we apply it to Parkinson's disease (PD), bipolar disorder (BD), and Alzheimer's disease (AD). In PD, PROTON linked genetic risk loci to genes essential for dopaminergic neuron survival and predicted pesticides toxic to patient-derived neurons, including the insecticide endosulfan, which ranked within the top 1.29% of predictions. In silico screens performed by PROTON reproduced six genome-wide $\\alpha$-synuclein experiments, including a split-ubiquitin yeast two-hybrid system (normalized enrichment score [NES] = 2.30, FDR-adjusted $p < 1 \\times 10^{-4}$), an ascorbate peroxidase proximity labeling assay (NES = 2.16, FDR $< 1 \\times 10^{-4}$), and a high-depth targeted exome sequencing study in 496 synucleinopathy patients (NES = 2.13, FDR $< 1 \\times 10^{-4}$). In BD, PROTON predicted calcitriol as a candidate drug that reversed proteomic alterations observed in cortical organoids derived from BD patients. In AD, we evaluated PROTON predictions in health records from $n = 610,524$ patients at Mass General Brigham, confirming that five PROTON-predicted drugs were associated with reduced seven-year dementia risk (minimum hazard ratio = 0.63, 95% CI: 0.53-0.75, $p < 1 \\times 10^{-7}$). PROTON generated neurological hypotheses that were evaluated across molecular, organoid, and clinical systems, defining a path for AI-driven discovery in neurological disease.", "date": "2025-12-13", "categories": []}
{"id": "2503.06940", "title": "CineBrain: A Large-Scale Multi-Modal Brain Dataset During Naturalistic Audiovisual Narrative Processing", "abstract": "Most research decoding brain signals into images, often using them as priors for generative models, has focused only on visual content. This overlooks the brain's natural ability to integrate auditory and visual information, for instance, sound strongly influences how we perceive visual scenes. To investigate this, we propose a new task of reconstructing continuous video stimuli from multimodal brain signals recorded during audiovisual stimulation. To enable this, we introduce CineBrain, the first large-scale dataset that synchronizes fMRI and EEG during audiovisual viewing, featuring six hours of \\textit{The Big Bang Theory} episodes for cross-modal alignment. We also conduct the first systematic exploration of combining fMRI and EEG for video reconstruction and present CineSync, a framework for reconstructing dynamic video using a Multi-Modal Fusion Encoder and a Neural Latent Decoder. CineSync achieves state-of-the-art performance in dynamic reconstruction, leveraging the complementary strengths of fMRI and EEG to improve visual fidelity. Our analysis shows that auditory cortical activations enhance decoding accuracy, highlighting the role of auditory input in visual perception. Project Page: https://jianxgao.github.io/CineBrain.", "date": "2025-03-10", "categories": []}
{"id": "2512.12135", "title": "BaRISTA: Brain Scale Informed Spatiotemporal Representation of Human Intracranial Neural Activity", "abstract": "Intracranial recordings have opened a unique opportunity to simultaneously measure activity across multiregional networks in the human brain. Recent works have focused on developing transformer-based neurofoundation models of such recordings that can generalize across subjects and datasets. However, these recordings exhibit highly complex spatiotemporal interactions across diverse spatial scales, from the single-channel scale to the scale of brain regions. As such, there remain critical open questions regarding how best to encode spatial information and how to design self-supervision tasks that enable the learning of brain network patterns and enhance downstream decoding performance using such high-dimensional, multiregional recordings. To allow for exploring these questions, we propose a new spatiotemporal transformer model of multiregional neural activity and a corresponding self-supervised masked latent reconstruction task, designed to enable flexibility in the spatial scale used for token encoding and masking. Applying this model on publicly available multiregional intracranial electrophysiology (iEEG) data, we demonstrate that adjusting the spatial scale for both token encoding and masked reconstruction significantly impacts downstream decoding. Further, we find that spatial encoding at larger scales than channel-level encoding, which is commonly used in existing iEEG transformer models, improves downstream decoding performance. Finally, we demonstrate that our method allows for region-level token encoding while also maintaining accurate channel-level neural reconstruction. Taken together, our modeling framework enables exploration of the spatial scales used for token encoding and masking, reveals their importance towards self-supervised pretraining of neurofoundation models of multiregional human brain activity, and enhances downstream decoding performance.", "date": "2025-12-13", "categories": []}
{"id": "2512.12461", "title": "Cross-Modal Representational Knowledge Distillation for Enhanced Spike-Informed LFP Modeling", "abstract": "Local field potentials (LFPs) can be routinely recorded alongside spiking activity in intracortical neural experiments, measure a larger complementary spatiotemporal scale of brain activity for scientific inquiry, and can offer practical advantages over spikes, including greater long-term stability, robustness to electrode degradation, and lower power requirements. Despite these advantages, recent neural modeling frameworks have largely focused on spiking activity since LFP signals pose inherent modeling challenges due to their aggregate, population-level nature, often leading to lower predictive power for downstream task variables such as motor behavior. To address this challenge, we introduce a cross-modal knowledge distillation framework that transfers high-fidelity representational knowledge from pretrained multi-session spike transformer models to LFP transformer models. Specifically, we first train a teacher spike model across multiple recording sessions using a masked autoencoding objective with a session-specific neural tokenization strategy. We then align the latent representations of the student LFP model to those of the teacher spike model. Our results show that the Distilled LFP models consistently outperform single- and multi-session LFP baselines in both fully unsupervised and supervised settings, and can generalize to other sessions without additional distillation while maintaining superior performance. These findings demonstrate that cross-modal knowledge distillation is a powerful and scalable approach for leveraging high-performing spike models to develop more accurate LFP models.", "date": "2025-12-13", "categories": []}
{"id": "2512.12462", "title": "Dynamical modeling of nonlinear latent factors in multiscale neural activity with real-time inference", "abstract": "Real-time decoding of target variables from multiple simultaneously recorded neural time-series modalities, such as discrete spiking activity and continuous field potentials, is important across various neuroscience applications. However, a major challenge for doing so is that different neural modalities can have different timescales (i.e., sampling rates) and different probabilistic distributions, or can even be missing at some time-steps. Existing nonlinear models of multimodal neural activity do not address different timescales or missing samples across modalities. Further, some of these models do not allow for real-time decoding. Here, we develop a learning framework that can enable real-time recursive decoding while nonlinearly aggregating information across multiple modalities with different timescales and distributions and with missing samples. This framework consists of 1) a multiscale encoder that nonlinearly aggregates information after learning within-modality dynamics to handle different timescales and missing samples in real time, 2) a multiscale dynamical backbone that extracts multimodal temporal dynamics and enables real-time recursive decoding, and 3) modality-specific decoders to account for different probabilistic distributions across modalities. In both simulations and three distinct multiscale brain datasets, we show that our model can aggregate information across modalities with different timescales and distributions and missing samples to improve real-time target decoding. Further, our method outperforms various linear and nonlinear multimodal benchmarks in doing so.", "date": "2025-12-13", "categories": []}
{"id": "2512.12881", "title": "Unsupervised learning of multiscale switching dynamical system models from multimodal neural data", "abstract": "Neural population activity often exhibits regime-dependent non-stationarity in the form of switching dynamics. Learning accurate switching dynamical system models can reveal how behavior is encoded in neural activity. Existing switching approaches have primarily focused on learning models from a single neural modality, either continuous Gaussian signals or discrete Poisson signals. However, multiple neural modalities are often recorded simultaneously to measure different spatiotemporal scales of brain activity, and all these modalities can encode behavior. Moreover, regime labels are typically unavailable in training data, posing a significant challenge for learning models of regime-dependent switching dynamics. To address these challenges, we develop a novel unsupervised learning algorithm that learns the parameters of switching multiscale dynamical system models using only multiscale neural observations. We demonstrate our method using both simulations and two distinct experimental datasets with multimodal spike-LFP observations during different motor tasks. We find that our switching multiscale dynamical system models more accurately decode behavior than switching single-scale dynamical models, showing the success of multiscale neural fusion. Further, our models outperform stationary multiscale models, illustrating the importance of tracking regime-dependent non-stationarity in multimodal neural data. The developed unsupervised learning framework enables more accurate modeling of complex multiscale neural dynamics by leveraging information in multimodal recordings while incorporating regime switches. This approach holds promise for improving the performance and robustness of brain-computer interfaces over time and for advancing our understanding of the neural basis of behavior.", "date": "2025-12-14", "categories": []}
{"id": "2503.01220", "title": "Tera-MIND: Tera-scale mouse brain simulation via spatial mRNA-guided diffusion", "abstract": "Holistic 3D modeling of molecularly defined brain structures is crucial for understanding complex brain functions. Using emerging tissue profiling technologies, researchers charted comprehensive atlases of mammalian brain with sub-cellular resolution and spatially resolved transcriptomic data. However, these tera-scale volumetric atlases pose computational challenges for modeling intricate brain structures within the native spatial context. We propose \\textbf{Tera-MIND}, a novel generative framework capable of simulating \\textbf{Tera}-scale \\textbf{M}ouse bra\\textbf{IN}s in 3D using a patch-based and boundary-aware \\textbf{D}iffusion model. Taking spatial gene expression as conditional input, we generate virtual mouse brains with comprehensive cellular morphological detail at teravoxel scale. Through the lens of 3D \\textit{gene}-\\textit{gene} self-attention, we identify spatial molecular interactions for key transcriptomic pathways, including glutamatergic and dopaminergic neuronal systems. Lastly, we showcase the translational applicability of Tera-MIND on previously unseen human brain samples. Tera-MIND offers an efficient generative modeling of whole virtual organisms, paving the way for integrative applications in biomedical research. Project website: https://musikisomorphie.github.io/Tera-MIND.html", "date": "2025-03-03", "categories": []}
{"id": "2512.10988", "title": "Mathematics of natural intelligence", "abstract": "In the process of evolution, the brain has achieved such perfection that artificial intelligence systems do not have and which needs its own mathematics. The concept of cognitome, introduced by the academician K.V. Anokhin, as the cognitive structure of the mind -- a high-order structure of the brain and a neural hypernetwork, is considered as the basis for modeling. Consciousness then is a special form of dynamics in this hypernetwork -- a large-scale integration of its cognitive elements. The cognitome, in turn, consists of interconnected COGs (cognitive groups of neurons) of two types -- functional systems and cellular ensembles. K.V. Anokhin sees the task of the fundamental theory of the brain and mind in describing these structures, their origin, functions and processes in them. The paper presents mathematical models of these structures based on new mathematical results, as well as models of different cognitive processes in terms of these models. In addition, it is shown that these models can be derived based on a fairly general principle of the brain works: \\textit{the brain discovers all possible causal relationships in the external world and draws all possible conclusions from them}. Based on these results, the paper presents models of: ``natural\" classification; theory of functional brain systems by P.K. Anokhin; prototypical theory of categorization by E. Roche; theory of causal models by Bob Rehter; theory of consciousness as integrated information by G. Tononi.", "date": "2025-12-07", "categories": []}
{"id": "2403.15176", "title": "Brain-aligning of semantic vectors improves neural decoding of visual stimuli", "abstract": "The development of algorithms to accurately decode neural information has long been a research focus in the field of neuroscience. Brain decoding typically involves training machine learning models to map neural data onto a preestablished vector representation of stimulus features. These vectors are usually derived from image- and/or text-based feature spaces. Nonetheless, the intrinsic characteristics of these vectors might fundamentally differ from those that are encoded by the brain, limiting the ability of decoders to accurately learn this mapping. To address this issue, we propose a framework, called brain-aligning of semantic vectors, that fine-tunes pretrained feature vectors to better align with the structure of neural representations of visual stimuli in the brain. We trained this model with functional magnetic resonance imaging (fMRI) and then performed zero-shot brain decoding on fMRI, magnetoencephalography (MEG), and electrocorticography (ECoG) data. fMRI-based brain-aligned vectors improved decoding performance across all three neuroimaging datasets when accuracy was determined by calculating the correlation coefficients between true and predicted vectors. Additionally, when decoding accuracy was determined via stimulus identification, this accuracy increased in specific category types; improvements varied depending on the original vector space that was used for brain-alignment, and consistent improvements were observed across all neuroimaging modalities.", "date": "2024-03-22", "categories": []}
{"id": "2505.22563", "title": "Do Large Language Models Think Like the Brain? Sentence-Level Evidences from Layer-Wise Embeddings and fMRI", "abstract": "Understanding whether large language models (LLMs) and the human brain converge on similar computational principles remains a fundamental and important question in cognitive neuroscience and AI. Do the brain-like patterns observed in LLMs emerge simply from scaling, or do they reflect deeper alignment with the architecture of human language processing? This study focuses on the sentence-level neural mechanisms of language models, systematically investigating how layer-wise representations in LLMs align with the dynamic neural responses during human sentence comprehension. By comparing hierarchical embeddings from 14 publicly available LLMs with fMRI data collected from participants, who were exposed to a naturalistic narrative story, we constructed sentence-level neural prediction models to identify the model layers most significantly correlated with brain region activations. Results show that improvements in model performance drive the evolution of representational architectures toward brain-like hierarchies, particularly achieving stronger functional and anatomical correspondence at higher semantic abstraction levels. These findings advance our understanding of the computational parallels between LLMs and the human brain, highlighting the potential of LLMs as models for human language processing.", "date": "2025-05-28", "categories": []}
{"id": "2506.11151", "title": "Self-Calibrating BCIs: Ranking and Recovery of Mental Targets Without Labels", "abstract": "We consider the problem of recovering a mental target (e.g., an image of a face) that a participant has in mind from paired EEG (i.e., brain responses) and image (i.e., perceived faces) data collected during interactive sessions without access to labeled information. The problem has been previously explored with labeled data but not via self-calibration, where labeled data is unavailable. Here, we present the first framework and an algorithm, CURSOR, that learns to recover unknown mental targets without access to labeled data or pre-trained decoders. Our experiments on naturalistic images of faces demonstrate that CURSOR can (1) predict image similarity scores that correlate with human perceptual judgments without any label information, (2) use these scores to rank stimuli against an unknown mental target, and (3) generate new stimuli indistinguishable from the unknown mental target (validated via a user study, N=53).", "date": "2025-06-11", "categories": []}
{"id": "2511.20532", "title": "MIMIC-MJX: Neuromechanical Emulation of Animal Behavior", "abstract": "The primary output of the nervous system is movement and behavior. While recent advances have democratized pose tracking during complex behavior, kinematic trajectories alone provide only indirect access to the underlying control processes. Here we present MIMIC-MJX, a framework for learning biologically-plausible neural control policies from kinematics. MIMIC-MJX models the generative process of motor control by training neural controllers that learn to actuate biomechanically-realistic body models in physics simulation to reproduce real kinematic trajectories. We demonstrate that our implementation is accurate, fast, data-efficient, and generalizable to diverse animal body models. Policies trained with MIMIC-MJX can be utilized to both analyze neural control strategies and simulate behavioral experiments, illustrating its potential as an integrative modeling framework for neuroscience.", "date": "2025-11-25", "categories": []}
{"id": "2512.02419", "title": "The brain-AI convergence: Predictive and generative world models for general-purpose computation", "abstract": "Recent advances in general-purpose AI systems with attention-based transformers offer a potential window into how the neocortex and cerebellum, despite their relatively uniform circuit architectures, give rise to diverse functions and, ultimately, to human intelligence. This Perspective provides a cross-domain comparison between the brain and AI that goes beyond the traditional focus on visual processing, adopting the emerging perspecive of world-model-based computation. Here, we identify shared computational mechanisms in the attention-based neocortex and the non-attentional cerebellum: both predict future world events from past inputs and construct internal world models through prediction-error learning. These predictive world models are repurposed for seemingly distinct functions -- understanding in sensory processing and generation in motor processing -- enabling the brain to achieve multi-domain capabilities and human-like adaptive intelligence. Notably, attention-based AI has independently converged on a similar learning paradigm and world-model-based computation. We conclude that these shared mechanisms in both biological and artificial systems constitute a core computational foundation for realizing diverse functions including high-level intelligence, despite their relatively uniform circuit structures. Our theoretical insights bridge neuroscience and AI, advancing our understanding of the computational essence of intelligence.", "date": "2025-12-02", "categories": []}
{"id": "2310.07908", "title": "Phase codes emerge in recurrent neural networks optimized for modular arithmetic", "abstract": "Recurrent neural networks (RNNs) can implement complex computations by leveraging a range of dynamics, such as oscillations, attractors, and transient trajectories. A growing body of work has highlighted the emergence of phase codes, a type of oscillatory activity where information is encoded in the relative phase of network activity, in RNNs trained for working memory tasks. However, these studies rely on architectural constraints or regularization schemes that explicitly promote oscillatory solutions. Here, we investigate whether phase coding can emerge purely from task optimization by training continuous-time RNNs to perform a simple modular arithmetic task without oscillatory-promoting biases. We find that in the absence of such biases, RNNs can learn phase code solutions. Surprisingly, we also uncover a rich diversity of alternative solutions that solve our modular arithmetic task via qualitatively distinct dynamics and dynamical mechanisms. We map the solution space for our task and show that the phase code solution occupies a distinct region. These results suggest that phase coding can be a natural but not inevitable outcome of training RNNs on modular arithmetic, and highlight the diversity of solutions RNNs can learn to solve simple tasks.", "date": "2023-10-11", "categories": []}
{"id": "2508.04988", "title": "Modeling Rapid Contextual Learning in the Visual Cortex with Fast-Weight Deep Autoencoder Networks", "abstract": "Recent neurophysiological studies have revealed that the early visual cortex can rapidly learn global image context, as evidenced by a sparsification of population responses and a reduction in mean activity when exposed to familiar versus novel image contexts. This phenomenon has been attributed primarily to local recurrent interactions, rather than changes in feedforward or feedback pathways, supported by both empirical findings and circuit-level modeling. Recurrent neural circuits capable of simulating these effects have been shown to reshape the geometry of neural manifolds, enhancing robustness and invariance to irrelevant variations. In this study, we employ a Vision Transformer (ViT)-based autoencoder to investigate, from a functional perspective, how familiarity training can induce sensitivity to global context in the early layers of a deep neural network. We hypothesize that rapid learning operates via fast weights, which encode transient or short-term memory traces, and we explore the use of Low-Rank Adaptation (LoRA) to implement such fast weights within each Transformer layer. Our results show that (1) The proposed ViT-based autoencoder's self-attention circuit performs a manifold transform similar to a neural circuit model of the familiarity effect. (2) Familiarity training aligns latent representations in early layers with those in the top layer that contains global context information. (3) Familiarity training broadens the self-attention scope within the remembered image context. (4) These effects are significantly amplified by LoRA-based fast weights. Together, these findings suggest that familiarity training introduces global sensitivity to earlier layers in a hierarchical network, and that a hybrid fast-and-slow weight architecture may provide a viable computational model for studying rapid global context learning in the brain.", "date": "2025-08-07", "categories": []}
{"id": "2512.01189", "title": "fMRI2GES: Co-speech Gesture Reconstruction from fMRI Signal with Dual Brain Decoding Alignment", "abstract": "Understanding how the brain responds to external stimuli and decoding this process has been a significant challenge in neuroscience. While previous studies typically concentrated on brain-to-image and brain-to-language reconstruction, our work strives to reconstruct gestures associated with speech stimuli perceived by brain. Unfortunately, the lack of paired \\{brain, speech, gesture\\} data hinders the deployment of deep learning models for this purpose. In this paper, we introduce a novel approach, \\textbf{fMRI2GES}, that allows training of fMRI-to-gesture reconstruction networks on unpaired data using \\textbf{Dual Brain Decoding Alignment}. This method relies on two key components: (i) observed texts that elicit brain responses, and (ii) textual descriptions associated with the gestures. Then, instead of training models in a completely supervised manner to find a mapping relationship among the three modalities, we harness an fMRI-to-text model, a text-to-gesture model with paired data and an fMRI-to-gesture model with unpaired data, establishing dual fMRI-to-gesture reconstruction patterns. Afterward, we explicitly align two outputs and train our model in a self-supervision way. We show that our proposed method can reconstruct expressive gestures directly from fMRI recordings. We also investigate fMRI signals from different ROIs in the cortex and how they affect generation results. Overall, we provide new insights into decoding co-speech gestures, thereby advancing our understanding of neuroscience and cognitive science.", "date": "2025-12-01", "categories": []}
{"id": "2512.01591", "title": "Scaling and context steer LLMs along the same computational path as the human brain", "abstract": "Recent studies suggest that the representations learned by large language models (LLMs) are partially aligned to those of the human brain. However, whether and why this alignment score arises from a similar sequence of computations remains elusive. In this study, we explore this question by examining temporally-resolved brain signals of participants listening to 10 hours of an audiobook. We study these neural dynamics jointly with a benchmark encompassing 22 LLMs varying in size and architecture type. Our analyses confirm that LLMs and the brain generate representations in a similar order: specifically, activations in the initial layers of LLMs tend to best align with early brain responses, while the deeper layers of LLMs tend to best align with later brain responses. This brain-LLM alignment is consistent across transformers and recurrent architectures. However, its emergence depends on both model size and context length. Overall, this study sheds light on the sequential nature of computations and the factors underlying the partial convergence between biological and artificial neural networks.", "date": "2025-12-01", "categories": []}
{"id": "2509.10650", "title": "On a Geometry of Interbrain Networks", "abstract": "Effective analysis in neuroscience benefits significantly from robust conceptual frameworks. Traditional metrics of interbrain synchrony in social neuroscience typically depend on fixed, correlation-based approaches, restricting their explanatory capacity to descriptive observations. Inspired by the successful integration of geometric insights in network science, we propose leveraging discrete geometry to examine the dynamic reconfigurations in neural interactions during social exchanges. Unlike conventional synchrony approaches, our method interprets inter-brain connectivity changes through the evolving geometric structures of neural networks. This geometric framework is realized through a pipeline that identifies critical transitions in network connectivity using entropy metrics derived from curvature distributions. By doing so, we significantly enhance the capacity of hyperscanning methodologies to uncover underlying neural mechanisms in interactive social behavior.", "date": "2025-09-12", "categories": []}
{"id": "2511.09947", "title": "EEGAgent: A Unified Framework for Automated EEG Analysis Using Large Language Models", "abstract": "Scalable and generalizable analysis of brain activity is essential for advancing both clinical diagnostics and cognitive research. Electroencephalography (EEG), a non-invasive modality with high temporal resolution, has been widely used for brain states analysis. However, most existing EEG models are usually tailored for individual specific tasks, limiting their utility in realistic scenarios where EEG analysis often involves multi-task and continuous reasoning. In this work, we introduce EEGAgent, a general-purpose framework that leverages large language models (LLMs) to schedule and plan multiple tools to automatically complete EEG-related tasks. EEGAgent is capable of performing the key functions: EEG basic information perception, spatiotemporal EEG exploration, EEG event detection, interaction with users, and EEG report generation. To realize these capabilities, we design a toolbox composed of different tools for EEG preprocessing, feature extraction, event detection, etc. These capabilities were evaluated on public datasets, and our EEGAgent can support flexible and interpretable EEG analysis, highlighting its potential for real-world clinical applications.", "date": "2025-11-13", "categories": []}
{"id": "2511.19757", "title": "What does it mean to understand language?", "abstract": "Language understanding entails not just extracting the surface-level meaning of the linguistic input, but constructing rich mental models of the situation it describes. Here we propose that because processing within the brain's core language system is fundamentally limited, deeply understanding language requires exporting information from the language system to other brain regions that compute perceptual and motor representations, construct mental models, and store our world knowledge and autobiographical memories. We review the existing evidence for this hypothesis, and argue that recent progress in cognitive neuroscience provides both the conceptual foundation and the methods to directly test it, thus opening up a new strategy to reveal what it means, cognitively and neurally, to understand language.", "date": "2025-11-24", "categories": []}
{"id": "2511.11480", "title": "Inferring response times of perceptual decisions with Poisson variational autoencoders", "abstract": "Many properties of perceptual decision making are well-modeled by deep neural networks. However, such architectures typically treat decisions as instantaneous readouts, overlooking the temporal dynamics of the decision process. We present an image-computable model of perceptual decision making in which choices and response times arise from efficient sensory encoding and Bayesian decoding of neural spiking activity. We use a Poisson variational autoencoder to learn unsupervised representations of visual stimuli in a population of rate-coded neurons, modeled as independent homogeneous Poisson processes. A task-optimized decoder then continually infers an approximate posterior over actions conditioned on incoming spiking activity. Combining these components with an entropy-based stopping rule yields a principled and image-computable model of perceptual decisions capable of generating trial-by-trial patterns of choices and response times. Applied to MNIST digit classification, the model reproduces key empirical signatures of perceptual decision making, including stochastic variability, right-skewed response time distributions, logarithmic scaling of response times with the number of alternatives (Hick's law), and speed-accuracy trade-offs.", "date": "2025-11-14", "categories": []}
{"id": "2511.18014", "title": "Modeling Retinal Ganglion Cells with Neural Differential Equations", "abstract": "This work explores Liquid Time-Constant Networks (LTCs) and Closed-form Continuous-time Networks (CfCs) for modeling retinal ganglion cell activity in tiger salamanders across three datasets. Compared to a convolutional baseline and an LSTM, both architectures achieved lower MAE, faster convergence, smaller model sizes, and favorable query times, though with slightly lower Pearson correlation. Their efficiency and adaptability make them well suited for scenarios with limited data and frequent retraining, such as edge deployments in vision prosthetics.", "date": "2025-11-22", "categories": []}
{"id": "2511.18325", "title": "Brain-MGF: Multimodal Graph Fusion Network for EEG-fMRI Brain Connectivity Analysis Under Psilocybin", "abstract": "Psychedelics, such as psilocybin, reorganise large-scale brain connectivity, yet how these changes are reflected across electrophysiological (electroencephalogram, EEG) and haemodynamic (functional magnetic resonance imaging, fMRI) networks remains unclear. We present Brain-MGF, a multimodal graph fusion network for joint EEG-fMRI connectivity analysis. For each modality, we construct graphs with partial-correlation edges and Pearson-profile node features, and learn subject-level embeddings via graph convolution. An adaptive softmax gate then fuses modalities with sample-specific weights to capture context-dependent contributions. Using the world's largest single-site psilocybin dataset, PsiConnect, Brain-MGF distinguishes psilocybin from no-psilocybin conditions in meditation and rest. Fusion improves over unimodal and non-adaptive variants, achieving 74.0% accuracy and 76.5% F1 score on meditation, and 76.0% accuracy with 85.8% ROC-AUC on rest. UMAP visualisations reveal clearer class separation for fused embeddings. These results indicate that adaptive graph fusion effectively integrates complementary EEG-fMRI information, providing an interpretable framework for characterising psilocybin-induced alterations in large-scale neural organisation.", "date": "2025-11-23", "categories": []}
{"id": "2511.16828", "title": "ManifoldFormer: Geometric Deep Learning for Neural Dynamics on Riemannian Manifolds", "abstract": "Existing EEG foundation models mainly treat neural signals as generic time series in Euclidean space, ignoring the intrinsic geometric structure of neural dynamics that constrains brain activity to low-dimensional manifolds. This fundamental mismatch between model assumptions and neural geometry limits representation quality and cross-subject generalization. ManifoldFormer addresses this limitation through a novel geometric deep learning framework that explicitly learns neural manifold representations. The architecture integrates three key innovations: a Riemannian VAE for manifold embedding that preserves geometric structure, a geometric Transformer with geodesic-aware attention mechanisms operating directly on neural manifolds, and a dynamics predictor leveraging neural ODEs for manifold-constrained temporal evolution. Extensive evaluation across four public datasets demonstrates substantial improvements over state-of-the-art methods, with 4.6-4.8% higher accuracy and 6.2-10.2% higher Cohen's Kappa, while maintaining robust cross-subject generalization. The geometric approach reveals meaningful neural patterns consistent with neurophysiological principles, establishing geometric constraints as essential for effective EEG foundation models.", "date": "2025-11-20", "categories": []}
{"id": "2405.15239", "title": "Brain3D: Generating 3D Objects from fMRI", "abstract": "Understanding the hidden mechanisms behind human's visual perception is a fundamental question in neuroscience. To that end, investigating into the neural responses of human mind activities, such as functional Magnetic Resonance Imaging (fMRI), has been a significant research vehicle. However, analyzing fMRI signals is challenging, costly, daunting, and demanding for professional training. Despite remarkable progress in fMRI analysis, existing approaches are limited to generating 2D images and far away from being biologically meaningful and practically useful. Under this insight, we propose to generate visually plausible and functionally more comprehensive 3D outputs decoded from brain signals, enabling more sophisticated modeling of fMRI data. Conceptually, we reformulate this task as a {\\em fMRI conditioned 3D object generation} problem. We design a novel 3D object representation learning method, Brain3D, that takes as input the fMRI data of a subject who was presented with a 2D image, and yields as output the corresponding 3D object images. The key capabilities of this model include tackling the noises with high-level semantic signals and a two-stage architecture design for progressive high-level information integration. Extensive experiments validate the superior capability of our model over previous state-of-the-art 3D object generation methods. Importantly, we show that our model captures the distinct functionalities of each region of human vision system as well as their intricate interplay relationships, aligning remarkably with the established discoveries in neuroscience. Further, preliminary evaluations indicate that Brain3D can successfully identify the disordered brain regions in simulated scenarios, such as V1, V2, V3, V4, and the medial temporal lobe (MTL) within the human visual system. Our data and code will be available at https://brain-3d.github.io/.", "date": "2024-05-24", "categories": []}
{"id": "2410.16136", "title": "Modeling Dynamic Neural Activity by combining Naturalistic Video Stimuli and Stimulus-independent Latent Factors", "abstract": "The neural activity in the visual processing is influenced by both external stimuli and internal brain states. Ideally, a neural predictive model should account for both of them. Currently, there are no dynamic encoding models that explicitly model a latent state and the entire neuronal response distribution. We address this gap by proposing a probabilistic model that predicts the joint distribution of the neuronal responses from video stimuli and stimulus-independent latent factors. After training and testing our model on mouse V1 neuronal responses, we find that it outperforms video-only models in terms of log-likelihood and achieves improvements in likelihood and correlation when conditioned on responses from other neurons. Furthermore, we find that the learned latent factors strongly correlate with mouse behavior and that they exhibit patterns related to the neurons' position on the visual cortex, although the model was trained without behavior and cortical coordinates. Our findings demonstrate that unsupervised learning of latent factors from population responses can reveal biologically meaningful structure that bridges sensory processing and behavior, without requiring explicit behavioral annotations during training.", "date": "2024-10-21", "categories": []}
{"id": "2505.15813", "title": "Meta-Learning an In-Context Transformer Model of Human Higher Visual Cortex", "abstract": "Understanding functional representations within higher visual cortex is a fundamental question in computational neuroscience. While artificial neural networks pretrained on large-scale datasets exhibit striking representational alignment with human neural responses, learning image-computable models of visual cortex relies on individual-level, large-scale fMRI datasets. The necessity for expensive, time-intensive, and often impractical data acquisition limits the generalizability of encoders to new subjects and stimuli. BraInCoRL uses in-context learning to predict voxelwise neural responses from few-shot examples without any additional finetuning for novel subjects and stimuli. We leverage a transformer architecture that can flexibly condition on a variable number of in-context image stimuli, learning an inductive bias over multiple subjects. During training, we explicitly optimize the model for in-context learning. By jointly conditioning on image features and voxel activations, our model learns to directly generate better performing voxelwise models of higher visual cortex. We demonstrate that BraInCoRL consistently outperforms existing voxelwise encoder designs in a low-data regime when evaluated on entirely novel images, while also exhibiting strong test-time scaling behavior. The model also generalizes to an entirely new visual fMRI dataset, which uses different subjects and fMRI data acquisition parameters. Further, BraInCoRL facilitates better interpretability of neural signals in higher visual cortex by attending to semantically relevant stimuli. Finally, we show that our framework enables interpretable mappings from natural language queries to voxel selectivity.", "date": "2025-05-21", "categories": []}
{"id": "2510.03156", "title": "Neural Correlates of Language Models Are Specific to Human Language", "abstract": "Previous work has shown correlations between the hidden states of large language models and fMRI brain responses, on language tasks. These correlations have been taken as evidence of the representational similarity of these models and brain states. This study tests whether these previous results are robust to several possible concerns. Specifically this study shows: (i) that the previous results are still found after dimensionality reduction, and thus are not attributable to the curse of dimensionality; (ii) that previous results are confirmed when using new measures of similarity; (iii) that correlations between brain representations and those from models are specific to models trained on human language; and (iv) that the results are dependent on the presence of positional encoding in the models. These results confirm and strengthen the results of previous research and contribute to the debate on the biological plausibility and interpretability of state-of-the-art large language models.", "date": "2025-10-03", "categories": []}
{"id": "2511.07912", "title": "Neurophysiological Characteristics of Adaptive Reasoning for Creative Problem-Solving Strategy", "abstract": "Adaptive reasoning enables humans to flexibly adjust inference strategies when environmental rules or contexts change, yet its underlying neural dynamics remain unclear. This study investigated the neurophysiological mechanisms of adaptive reasoning using a card-sorting paradigm combined with electroencephalography and compared human performance with that of a multimodal large language model. Stimulus- and feedback-locked analyses revealed coordinated delta-theta-alpha dynamics: early delta-theta activity reflected exploratory monitoring and rule inference, whereas occipital alpha engagement indicated confirmatory stabilization of attention after successful rule identification. In contrast, the multimodal large language model exhibited only short-term feedback-driven adjustments without hierarchical rule abstraction or genuine adaptive reasoning. These findings identify the neural signatures of human adaptive reasoning and highlight the need for brain-inspired artificial intelligence that incorporates oscillatory feedback coordination for true context-sensitive adaptation.", "date": "2025-11-11", "categories": []}
{"id": "2511.07918", "title": "Distinct Theta Synchrony across Speech Modes: Perceived, Spoken, Whispered, and Imagined", "abstract": "Human speech production encompasses multiple modes such as perceived, overt, whispered, and imagined, each reflecting distinct neural mechanisms. Among these, theta-band synchrony has been closely associated with language processing, attentional control, and inner speech. However, previous studies have largely focused on a single mode, such as overt speech, and have rarely conducted an integrated comparison of theta synchrony across different speech modes. In this study, we analyzed differences in theta-band neural synchrony across speech modes based on connectivity metrics, focusing on region-wise variations. The results revealed that overt and whispered speech exhibited broader and stronger frontotemporal synchrony, reflecting active motor-phonological coupling during overt articulation, whereas perceived speech showed dominant posterior and temporal synchrony patterns, consistent with auditory perception and comprehension processes. In contrast, imagined speech demonstrated a more spatially confined but internally coherent synchronization pattern, primarily involving frontal and supplementary motor regions. These findings indicate that the extent and spatial distribution of theta synchrony differ substantially across modes, with overt articulation engaging widespread cortical interactions, whispered speech showing intermediate engagement, and perception relying predominantly on temporoparietal networks. Therefore, this study aims to elucidate the differences in theta-band neural synchrony across various speech modes, thereby uncovering both the shared and distinct neural dynamics underlying language perception and imagined speech.", "date": "2025-11-11", "categories": []}
{"id": "2511.07988", "title": "The One Where They Brain-Tune for Social Cognition: Multi-Modal Brain-Tuning on Friends", "abstract": "Recent studies on audio models show brain-tuning - fine-tuning models to better predict corresponding fMRI activity - improves brain alignment and increases performance on downstream semantic and audio tasks. We extend this approach to a multimodal audio-video model to enhance social cognition, targeting the Superior Temporal Sulcus (STS), a key region for social processing, while subjects watch Friends. We find significant increases in brain alignment to the STS and an adjacent ROI, as well as improvements to a social cognition task related to the training data - sarcasm detection in sitcoms. In summary, our study extends brain-tuning to the multi-modal domain, demonstrating improvements to a downstream task after tuning to a relevant functional region.", "date": "2025-11-11", "categories": []}
{"id": "2511.08436", "title": "Understanding Electro-communication and Electro-sensing in Weakly Electric Fish using Multi-Agent Deep Reinforcement Learning", "abstract": "Weakly electric fish, like Gnathonemus petersii, use a remarkable electrical modality for active sensing and communication, but studying their rich electrosensing and electrocommunication behavior and associated neural activity in naturalistic settings remains experimentally challenging. Here, we present a novel biologically-inspired computational framework to study these behaviors, where recurrent neural network (RNN) based artificial agents trained via multi-agent reinforcement learning (MARL) learn to modulate their electric organ discharges (EODs) and movement patterns to collectively forage in virtual environments. Trained agents demonstrate several emergent features consistent with real fish collectives, including heavy tailed EOD interval distributions, environmental context dependent shifts in EOD interval distributions, and social interaction patterns like freeloading, where agents reduce their EOD rates while benefiting from neighboring agents' active sensing. A minimal two-fish assay further isolates the role of electro-communication, showing that access to conspecific EODs and relative dominance jointly shape foraging success. Notably, these behaviors emerge through evolution-inspired rewards for individual fitness and emergent inter-agent interactions, rather than through rewarding agents explicitly for social interactions. Our work has broad implications for the neuroethology of weakly electric fish, as well as other social, communicating animals in which extensive recordings from multiple individuals, and thus traditional data-driven modeling, are infeasible.", "date": "2025-11-11", "categories": []}
{"id": "2511.00065", "title": "Aligning Brain Signals with Multimodal Speech and Vision Embeddings", "abstract": "When we hear the word \"house\", we don't just process sound, we imagine walls, doors, memories. The brain builds meaning through layers, moving from raw acoustics to rich, multimodal associations. Inspired by this, we build on recent work from Meta that aligned EEG signals with averaged wav2vec2 speech embeddings, and ask a deeper question: which layers of pre-trained models best reflect this layered processing in the brain? We compare embeddings from two models: wav2vec2, which encodes sound into language, and CLIP, which maps words to images. Using EEG recorded during natural speech perception, we evaluate how these embeddings align with brain activity using ridge regression and contrastive decoding. We test three strategies: individual layers, progressive concatenation, and progressive summation. The findings suggest that combining multimodal, layer-aware representations may bring us closer to decoding how the brain understands language, not just as sound, but as experience.", "date": "2025-10-29", "categories": []}
{"id": "2506.05320", "title": "Generalizable, real-time neural decoding with hybrid state-space models", "abstract": "Real-time decoding of neural activity is central to neuroscience and neurotechnology applications, from closed-loop experiments to brain-computer interfaces, where models are subject to strict latency constraints. Traditional methods, including simple recurrent neural networks, are fast and lightweight but often struggle to generalize to unseen data. In contrast, recent Transformer-based approaches leverage large-scale pretraining for strong generalization performance, but typically have much larger computational requirements and are not always suitable for low-resource or real-time settings. To address these shortcomings, we present POSSM, a novel hybrid architecture that combines individual spike tokenization via a cross-attention module with a recurrent state-space model (SSM) backbone to enable (1) fast and causal online prediction on neural activity and (2) efficient generalization to new sessions, individuals, and tasks through multi-dataset pretraining. We evaluate POSSM's decoding performance and inference speed on intracortical decoding of monkey motor tasks, and show that it extends to clinical applications, namely handwriting and speech decoding in human subjects. Notably, we demonstrate that pretraining on monkey motor-cortical recordings improves decoding performance on the human handwriting task, highlighting the exciting potential for cross-species transfer. In all of these tasks, we find that POSSM achieves decoding accuracy comparable to state-of-the-art Transformers, at a fraction of the inference cost (up to 9x faster on GPU). These results suggest that hybrid SSMs are a promising approach to bridging the gap between accuracy, inference speed, and generalization when training neural decoders for real-time, closed-loop applications.", "date": "2025-06-05", "categories": []}
{"id": "2511.04691", "title": "A Penny for Your Thoughts: Decoding Speech from Inexpensive Brain Signals", "abstract": "We explore whether neural networks can decode brain activity into speech by mapping EEG recordings to audio representations. Using EEG data recorded as subjects listened to natural speech, we train a model with a contrastive CLIP loss to align EEG-derived embeddings with embeddings from a pre-trained transformer-based speech model. Building on the state-of-the-art EEG decoder from Meta, we introduce three architectural modifications: (i) subject-specific attention layers (+0.15% WER improvement), (ii) personalized spatial attention (+0.45%), and (iii) a dual-path RNN with attention (-1.87%). Two of the three modifications improved performance, highlighting the promise of personalized architectures for brain-to-speech decoding and applications in brain-computer interfaces.", "date": "2025-10-28", "categories": []}
{"id": "2511.02722", "title": "Association-sensory spatiotemporal hierarchy and functional gradient-regularised recurrent neural network with implications for schizophrenia", "abstract": "The human neocortex is functionally organised at its highest level along a continuous sensory-to-association (AS) hierarchy. This study characterises the AS hierarchy of patients with schizophrenia in a comparison with controls. Using a large fMRI dataset (N=355), we extracted individual AS gradients via spectral analysis of brain connectivity, quantified hierarchical specialisation by gradient spread, and related this spread with connectivity geometry. We found that schizophrenia compresses the AS hierarchy indicating reduced functional differentiation. By modelling neural timescale with the Ornstein-Uhlenbeck process, we observed that the most specialised, locally cohesive regions at the gradient extremes exhibit dynamics with a longer time constant, an effect that is attenuated in schizophrenia. To study computation, we used the gradients to regularise subject-specific recurrent neural networks (RNNs) trained on working memory tasks. Networks endowed with greater gradient spread learned more efficiently, plateaued at lower task loss, and maintained stronger alignment to the prescribed AS hierarchical geometry. Fixed point linearisation showed that high-range networks settled into more stable neural states during memory delay, evidenced by lower energy and smaller maximal Jacobian eigenvalues. This gradient-regularised RNN framework therefore links large-scale cortical architecture with fixed point stability, providing a mechanistic account of how gradient de-differentiation could destabilise neural computations in schizophrenia, convergently supported by empirical timescale flattening and model-based evidence of less stable fixed points.", "date": "2025-11-04", "categories": []}
{"id": "2509.04633", "title": "The Physical Basis of Prediction: World Model Formation in Neural Organoids via an LLM-Generated Curriculum", "abstract": "The capacity of an embodied agent to understand, predict, and interact with its environment is fundamentally contingent on an internal world model. This paper introduces a novel framework for investigating the formation and adaptation of such world models within a biological substrate: human neural organoids. We present a curriculum of three scalable, closed-loop virtual environments designed to train these biological agents and probe the underlying synaptic mechanisms of learning, such as long-term potentiation (LTP) and long-term depression (LTD). We detail the design of three distinct task environments that demand progressively more sophisticated world models for successful decision-making: (1) a conditional avoidance task for learning static state-action contingencies, (2) a one-dimensional predator-prey scenario for goal-directed interaction, and (3) a replication of the classic Pong game for modeling dynamic, continuous-time systems. For each environment, we formalize the state and action spaces, the sensory encoding and motor decoding mechanisms, and the feedback protocols based on predictable (reward) and unpredictable (punishment) stimulation, which serve to drive model refinement. In a significant methodological advance, we propose a meta-learning approach where a Large Language Model automates the generative design and optimization of experimental protocols, thereby scaling the process of environment and curriculum design. Finally, we outline a multi-modal evaluation strategy that moves beyond task performance to directly measure the physical correlates of the learned world model by quantifying synaptic plasticity at electrophysiological, cellular, and molecular levels. This work bridges the gap between model-based reinforcement learning and computational neuroscience, offering a unique platform for studying embodiment, decision-making, and the physical basis of intelligence.", "date": "2025-09-04", "categories": []}
{"id": "2511.01868", "title": "Condition-Invariant fMRI Decoding of Speech Intelligibility with Deep State Space Model", "abstract": "Clarifying the neural basis of speech intelligibility is critical for computational neuroscience and digital speech processing. Recent neuroimaging studies have shown that intelligibility modulates cortical activity beyond simple acoustics, primarily in the superior temporal and inferior frontal gyri. However, previous studies have been largely confined to clean speech, leaving it unclear whether the brain employs condition-invariant neural codes across diverse listening environments. To address this gap, we propose a novel architecture built upon a deep state space model for decoding intelligibility from fMRI signals, specifically tailored to their high-dimensional temporal structure. We present the first attempt to decode intelligibility across acoustically distinct conditions, showing our method significantly outperforms classical approaches. Furthermore, region-wise analysis highlights contributions from auditory, frontal, and parietal regions, and cross-condition transfer indicates the presence of condition-invariant neural codes, thereby advancing understanding of abstract linguistic representations in the brain.", "date": "2025-10-21", "categories": []}
{"id": "2510.25943", "title": "InputDSA: Demixing then Comparing Recurrent and Externally Driven Dynamics", "abstract": "In control problems and basic scientific modeling, it is important to compare observations with dynamical simulations. For example, comparing two neural systems can shed light on the nature of emergent computations in the brain and deep neural networks. Recently, Ostrow et al. (2023) introduced Dynamical Similarity Analysis (DSA), a method to measure the similarity of two systems based on their recurrent dynamics rather than geometry or topology. However, DSA does not consider how inputs affect the dynamics, meaning that two similar systems, if driven differently, may be classified as different. Because real-world dynamical systems are rarely autonomous, it is important to account for the effects of input drive. To this end, we introduce a novel metric for comparing both intrinsic (recurrent) and input-driven dynamics, called InputDSA (iDSA). InputDSA extends the DSA framework by estimating and comparing both input and intrinsic dynamic operators using a variant of Dynamic Mode Decomposition with control (DMDc) based on subspace identification. We demonstrate that InputDSA can successfully compare partially observed, input-driven systems from noisy data. We show that when the true inputs are unknown, surrogate inputs can be substituted without a major deterioration in similarity estimates. We apply InputDSA on Recurrent Neural Networks (RNNs) trained with Deep Reinforcement Learning, identifying that high-performing networks are dynamically similar to one another, while low-performing networks are more diverse. Lastly, we apply InputDSA to neural data recorded from rats performing a cognitive task, demonstrating that it identifies a transition from input-driven evidence accumulation to intrinsically-driven decision-making. Our work demonstrates that InputDSA is a robust and efficient method for comparing intrinsic dynamics and the effect of external input on dynamical systems.", "date": "2025-10-29", "categories": []}
{"id": "2507.17897", "title": "Multimodal Recurrent Ensembles for Predicting Brain Responses to Naturalistic Movies (Algonauts 2025)", "abstract": "Accurately predicting distributed cortical responses to naturalistic stimuli requires models that integrate visual, auditory and semantic information over time. We present a hierarchical multimodal recurrent ensemble that maps pretrained video, audio, and language embeddings to fMRI time series recorded while four subjects watched almost 80 hours of movies provided by the Algonauts 2025 challenge. Modality-specific bidirectional RNNs encode temporal dynamics; their hidden states are fused and passed to a second recurrent layer, and lightweight subject-specific heads output responses for 1000 cortical parcels. Training relies on a composite MSE-correlation loss and a curriculum that gradually shifts emphasis from early sensory to late association regions. Averaging 100 model variants further boosts robustness. The resulting system ranked third on the competition leaderboard, achieving an overall Pearson r = 0.2094 and the highest single-parcel peak score (mean r = 0.63) among all participants, with particularly strong gains for the most challenging subject (Subject 5). The approach establishes a simple, extensible baseline for future multimodal brain-encoding benchmarks.", "date": "2025-07-23", "categories": []}
{"id": "2510.25023", "title": "Disentangling Shared and Private Neural Dynamics with SPIRE: A Latent Modeling Framework for Deep Brain Stimulation", "abstract": "Disentangling shared network-level dynamics from region-specific activity is a central challenge in modeling multi-region neural data. We introduce SPIRE (Shared-Private Inter-Regional Encoder), a deep multi-encoder autoencoder that factorizes recordings into shared and private latent subspaces with novel alignment and disentanglement losses. Trained solely on baseline data, SPIRE robustly recovers cross-regional structure and reveals how external perturbations reorganize it. On synthetic benchmarks with ground-truth latents, SPIRE outperforms classical probabilistic models under nonlinear distortions and temporal misalignments. Applied to intracranial deep brain stimulation (DBS) recordings, SPIRE shows that shared latents reliably encode stimulation-specific signatures that generalize across sites and frequencies. These results establish SPIRE as a practical, reproducible tool for analyzing multi-region neural dynamics under stimulation.", "date": "2025-10-28", "categories": []}
{"id": "2510.23940", "title": "Modeling Biological Multifunctionality with Echo State Networks", "abstract": "In this work, a three-dimensional multicomponent reaction-diffusion model has been developed, combining excitable-system dynamics with diffusion processes and sharing conceptual features with the FitzHugh-Nagumo model. Designed to capture the spatiotemporal behavior of biological systems, particularly electrophysiological processes, the model was solved numerically to generate time-series data. These data were subsequently used to train and evaluate an Echo State Network (ESN), which successfully reproduced the system's dynamic behavior. The results demonstrate that simulating biological dynamics using data-driven, multifunctional ESN models is both feasible and effective.", "date": "2025-10-27", "categories": []}
{"id": "2502.01299", "title": "Probabilistic adaptation of language comprehension for individual speakers: evidence from neural oscillations", "abstract": "Listeners adapt language comprehension based on their mental representations of speakers, but how these representations are updated remains unclear. We investigated whether listeners probabilistically adapt comprehension based on the frequency of speakers making stereotype-incongruent statements. In two EEG experiments, participants heard speakers make stereotype-congruent or incongruent statements, with incongruency base rate manipulated. In Experiment 1, stereotype-incongruent statements decreased high-beta (21-30 Hz) and theta (4-6 Hz) oscillatory power in the low base rate condition but increased it in the high base rate condition. The theta effect varied with listeners' openness trait: less open-minded participants tended to show theta increases to stereotype incongruencies, while more open-minded participants tended to show theta decreases. In Experiment 2, we dissociated incongruency base rate from the target speaker by manipulating it using a non-target speaker and found that only the high-beta effect persisted. Our findings reveal two potential mechanisms: a speaker-general mechanism (indicated by high-beta oscillations) that adjusts overall expectations about hearing statements that violate social stereotypes, and a speaker-specific mechanism (indicated by theta oscillations) that updates a more detailed mental model specifically about an individual speaker. These findings provide evidence for how language processing interacts with social cognition.", "date": "2025-02-03", "categories": []}
{"id": "2505.14806", "title": "Place Cells as Multi-Scale Position Embeddings: Random Walk Transition Kernels for Path Planning", "abstract": "The hippocampus supports spatial navigation by encoding cognitive maps through collective place cell activity. We model the place cell population as non-negative spatial embeddings derived from the spectral decomposition of multi-step random walk transition kernels. In this framework, inner product or equivalently Euclidean distance between embeddings encode similarity between locations in terms of their transition probability across multiple scales, forming a cognitive map of adjacency. The combination of non-negativity and inner-product structure naturally induces sparsity, providing a principled explanation for the localized firing fields of place cells without imposing explicit constraints. The temporal parameter that defines the diffusion scale also determines field size, aligning with the hippocampal dorsoventral hierarchy. Our approach constructs global representations efficiently through recursive composition of local transitions, enabling smooth, trap-free navigation and preplay-like trajectory generation. Moreover, theta phase arises intrinsically as the angular relation between embeddings, linking spatial and temporal coding within a single representational geometry.", "date": "2025-05-20", "categories": []}
{"id": "2506.01659", "title": "Engram Memory Encoding and Retrieval: A Neurocomputational Perspective", "abstract": "Despite substantial research into the biological basis of memory, the precise mechanisms by which experiences are encoded, stored, and retrieved in the brain remain incompletely understood. A growing body of evidence supports the engram theory, which posits that sparse populations of neurons undergo lasting physical and biochemical changes to support long-term memory. Yet, a comprehensive computational framework that integrates biological findings with mechanistic models remains elusive. This work synthesizes insights from cellular neuroscience and computational modeling to address key challenges in engram research: how engram neurons are identified and manipulated; how synaptic plasticity mechanisms contribute to stable memory traces; and how sparsity promotes efficient, interference-resistant representations. Relevant computational approaches -- such as sparse regularization, engram gating, and biologically inspired architectures like Sparse Distributed Memory and spiking neural networks -- are also examined. Together, these findings suggest that memory efficiency, capacity, and stability emerge from the interaction of plasticity and sparsity constraints. By integrating neurobiological and computational perspectives, this paper provides a comprehensive theoretical foundation for engram research and proposes a roadmap for future inquiry into the mechanisms underlying memory, with implications for the diagnosis and treatment of memory-related disorders.", "date": "2025-06-02", "categories": []}
{"id": "2507.22313", "title": "Time-Resolved EEG Decoding of Semantic Processing Reveals Altered Neural Dynamics in Depression and Suicidality", "abstract": "Depression and suicidality affect cognitive and emotional processes, yet objective, task-evoked neural readouts of mental health remain limited. We investigated the spatiotemporal dynamics of affective semantic processing using multivariate decoding of time-resolved, 64-channel electroencephalography (EEG). Participants (N=137) performed a sentence-evaluation task with emotionally salient, self-referential statements. We identified robust neural signatures of semantic processing, with peak decoding accuracy between 300-600 ms -- a window associated with rapid, stimulus-driven semantic evaluation and conflict monitoring. Relative to healthy controls, individuals with depression and suicidal ideation showed earlier onset, longer duration, and greater amplitude decoding responses, along with broader cross-temporal generalization and enhanced contributions from frontocentral and parietotemporal components. These findings suggest altered sensitivity and impaired disengagement from emotionally salient content in the clinical groups, advancing our understanding of the neurocognitive basis of mental health and establishing a compact and interpretable EEG-based index of semantic-evaluation dynamics with potential diagnostic relevance.", "date": "2025-07-30", "categories": []}
{"id": "2509.17174", "title": "Self-Supervised Discovery of Neural Circuits in Spatially Patterned Neural Responses with Graph Neural Networks", "abstract": "Inferring synaptic connectivity from neural population activity is a fundamental challenge in computational neuroscience, complicated by partial observability and mismatches between inference models and true circuit dynamics. In this study, we propose a graph-based neural inference model that simultaneously predicts neural activity and infers latent connectivity by modeling neurons as interacting nodes in a graph. The architecture features two distinct modules: one for learning structural connectivity and another for predicting future spiking activity via a graph neural network (GNN). Our model accommodates unobserved neurons through auxiliary nodes, allowing for inference in partially observed circuits. We evaluate this approach using synthetic data generated from ring attractor network models and real spike recordings from head direction cells in mice. Across a wide range of conditions, including varying recurrent connectivity, external inputs, and incomplete observations, our model reliably resolves spurious correlations and recovers accurate weight profiles. When applied to real data, the inferred connectivity aligns with theoretical predictions of continuous attractor models. These results highlight the potential of GNN-based models to infer latent neural circuitry through self-supervised structure learning, while leveraging the spike prediction task to flexibly link connectivity and dynamics across both simulated and biological neural systems.", "date": "2025-09-21", "categories": []}
{"id": "2410.19315", "title": "Brain-like Variational Inference", "abstract": "Inference in both brains and machines can be formalized by optimizing a shared objective: maximizing the evidence lower bound (ELBO) in machine learning, or minimizing variational free energy (F) in neuroscience (ELBO = -F). While this equivalence suggests a unifying framework, it leaves open how inference is implemented in neural systems. Here, we introduce FOND (Free energy Online Natural-gradient Dynamics), a framework that derives neural inference dynamics from three principles: (1) natural gradients on F, (2) online belief updating, and (3) iterative refinement. We apply FOND to derive iP-VAE (iterative Poisson variational autoencoder), a recurrent spiking neural network that performs variational inference through membrane potential dynamics, replacing amortized encoders with iterative inference updates. Theoretically, iP-VAE yields several desirable features such as emergent normalization via lateral competition, and hardware-efficient integer spike count representations. Empirically, iP-VAE outperforms both standard VAEs and Gaussian-based predictive coding models in sparsity, reconstruction, and biological plausibility, and scales to complex color image datasets such as CelebA. iP-VAE also exhibits strong generalization to out-of-distribution inputs, exceeding hybrid iterative-amortized VAEs. These results demonstrate how deriving inference algorithms from first principles can yield concrete architectures that are simultaneously biologically plausible and empirically effective.", "date": "2024-10-25", "categories": []}
{"id": "2506.00138", "title": "Intrinsic Goals for Autonomous Agents: Model-Based Exploration in Virtual Zebrafish Predicts Ethological Behavior and Whole-Brain Dynamics", "abstract": "Autonomy is a hallmark of animal intelligence, enabling adaptive and intelligent behavior in complex environments without relying on external reward or task structure. Existing reinforcement learning approaches to exploration in reward-free environments, including a class of methods known as model-based intrinsic motivation, exhibit inconsistent exploration patterns and do not converge to an exploratory policy, thus failing to capture robust autonomous behaviors observed in animals. Moreover, systems neuroscience has largely overlooked the neural basis of autonomy, focusing instead on experimental paradigms where animals are motivated by external reward rather than engaging in ethological, naturalistic and task-independent behavior. To bridge these gaps, we introduce a novel model-based intrinsic drive explicitly designed after the principles of autonomous exploration in animals. Our method (3M-Progress) achieves animal-like exploration by tracking divergence between an online world model and a fixed prior learned from an ecological niche. To the best of our knowledge, we introduce the first autonomous embodied agent that predicts brain data entirely from self-supervised optimization of an intrinsic goal -- without any behavioral or neural training data -- demonstrating that 3M-Progress agents capture the explainable variance in behavioral patterns and whole-brain neural-glial dynamics recorded from autonomously behaving larval zebrafish, thereby providing the first goal-driven, population-level model of neural-glial computation. Our findings establish a computational framework connecting model-based intrinsic motivation to naturalistic behavior, providing a foundation for building artificial agents with animal-like autonomy.", "date": "2025-05-30", "categories": []}
{"id": "2510.20831", "title": "BACE: Behavior-Adaptive Connectivity Estimation for Interpretable Graphs of Neural Dynamics", "abstract": "Understanding how distributed brain regions coordinate to produce behavior requires models that are both predictive and interpretable. We introduce Behavior-Adaptive Connectivity Estimation (BACE), an end-to-end framework that learns phase-specific, directed inter-regional connectivity directly from multi-region intracranial local field potentials (LFP). BACE aggregates many micro-contacts within each anatomical region via per-region temporal encoders, applies a learnable adjacency specific to each behavioral phase, and is trained on a forecasting objective. On synthetic multivariate time series with known graphs, BACE accurately recovers ground-truth directed interactions while achieving forecasting performance comparable to state-of-the-art baselines. Applied to human subcortical LFP recorded simultaneously from eight regions during a cued reaching task, BACE yields an explicit connectivity matrix for each within-trial behavioral phase. The resulting behavioral phase-specific graphs reveal behavior-aligned reconfiguration of inter-regional influence and provide compact, interpretable adjacency matrices for comparing network organization across behavioral phases. By linking predictive success to explicit connectivity estimates, BACE offers a practical tool for generating data-driven hypotheses about the dynamic coordination of subcortical regions during behavior.", "date": "2025-10-11", "categories": []}
{"id": "2510.21315", "title": "Seemingly Redundant Modules Enhance Robust Odor Learning in Fruit Flies", "abstract": "Biological circuits have evolved to incorporate multiple modules that perform similar functions. In the fly olfactory circuit, both lateral inhibition (LI) and neuronal spike frequency adaptation (SFA) are thought to enhance pattern separation for odor learning. However, it remains unclear whether these mechanisms play redundant or distinct roles in this process. In this study, we present a computational model of the fly olfactory circuit to investigate odor discrimination under varying noise conditions that simulate complex environments. Our results show that LI primarily enhances odor discrimination in low- and medium-noise scenarios, but this benefit diminishes and may reverse under higher-noise conditions. In contrast, SFA consistently improves discrimination across all noise levels. LI is preferentially engaged in low- and medium-noise environments, whereas SFA dominates in high-noise settings. When combined, these two sparsification mechanisms enable optimal discrimination performance. This work demonstrates that seemingly redundant modules in biological circuits can, in fact, be essential for achieving optimal learning in complex contexts.", "date": "2025-10-24", "categories": []}
{"id": "2510.21585", "title": "REVE: A Foundation Model for EEG -- Adapting to Any Setup with Large-Scale Pretraining on 25,000 Subjects", "abstract": "Foundation models have transformed AI by reducing reliance on task-specific data through large-scale pretraining. While successful in language and vision, their adoption in EEG has lagged due to the heterogeneity of public datasets, which are collected under varying protocols, devices, and electrode configurations. Existing EEG foundation models struggle to generalize across these variations, often restricting pretraining to a single setup, resulting in suboptimal performance, in particular under linear probing. We present REVE (Representation for EEG with Versatile Embeddings), a pretrained model explicitly designed to generalize across diverse EEG signals. REVE introduces a novel 4D positional encoding scheme that enables it to process signals of arbitrary length and electrode arrangement. Using a masked autoencoding objective, we pretrain REVE on over 60,000 hours of EEG data from 92 datasets spanning 25,000 subjects, representing the largest EEG pretraining effort to date. REVE achieves state-of-the-art results on 10 downstream EEG tasks, including motor imagery classification, seizure detection, sleep staging, cognitive load estimation, and emotion recognition. With little to no fine-tuning, it demonstrates strong generalization, and nuanced spatio-temporal modeling. We release code, pretrained weights, and tutorials to support standardized EEG research and accelerate progress in clinical neuroscience.", "date": "2025-10-24", "categories": []}
{"id": "2510.20068", "title": "Coupled Transformer Autoencoder for Disentangling Multi-Region Neural Latent Dynamics", "abstract": "Simultaneous recordings from thousands of neurons across multiple brain areas reveal rich mixtures of activity that are shared between regions and dynamics that are unique to each region. Existing alignment or multi-view methods neglect temporal structure, whereas dynamical latent variable models capture temporal dependencies but are usually restricted to a single area, assume linear read-outs, or conflate shared and private signals. We introduce the Coupled Transformer Autoencoder (CTAE) - a sequence model that addresses both (i) non-stationary, non-linear dynamics and (ii) separation of shared versus region-specific structure in a single framework. CTAE employs transformer encoders and decoders to capture long-range neural dynamics and explicitly partitions each region's latent space into orthogonal shared and private subspaces. We demonstrate the effectiveness of CTAE on two high-density electrophysiology datasets with simultaneous recordings from multiple regions, one from motor cortical areas and the other from sensory areas. CTAE extracts meaningful representations that better decode behavioral variables compared to existing approaches.", "date": "2025-10-22", "categories": []}
{"id": "2510.20683", "title": "A Scalable, Causal, and Energy Efficient Framework for Neural Decoding with Spiking Neural Networks", "abstract": "Brain-computer interfaces (BCIs) promise to enable vital functions, such as speech and prosthetic control, for individuals with neuromotor impairments. Central to their success are neural decoders, models that map neural activity to intended behavior. Current learning-based decoding approaches fall into two classes: simple, causal models that lack generalization, or complex, non-causal models that generalize and scale offline but struggle in real-time settings. Both face a common challenge, their reliance on power-hungry artificial neural network backbones, which makes integration into real-world, resource-limited systems difficult. Spiking neural networks (SNNs) offer a promising alternative. Because they operate causally these models are suitable for real-time use, and their low energy demands make them ideal for battery-constrained environments. To this end, we introduce Spikachu: a scalable, causal, and energy-efficient neural decoding framework based on SNNs. Our approach processes binned spikes directly by projecting them into a shared latent space, where spiking modules, adapted to the timing of the input, extract relevant features; these latent representations are then integrated and decoded to generate behavioral predictions. We evaluate our approach on 113 recording sessions from 6 non-human primates, totaling 43 hours of recordings. Our method outperforms causal baselines when trained on single sessions using between 2.26 and 418.81 times less energy. Furthermore, we demonstrate that scaling up training to multiple sessions and subjects improves performance and enables few-shot transfer to unseen sessions, subjects, and tasks. Overall, Spikachu introduces a scalable, online-compatible neural decoding framework based on SNNs, whose performance is competitive relative to state-of-the-art models while consuming orders of magnitude less energy.", "date": "2025-10-23", "categories": []}
{"id": "2510.20762", "title": "MEIcoder: Decoding Visual Stimuli from Neural Activity by Leveraging Most Exciting Inputs", "abstract": "Decoding visual stimuli from neural population activity is crucial for understanding the brain and for applications in brain-machine interfaces. However, such biological data is often scarce, particularly in primates or humans, where high-throughput recording techniques, such as two-photon imaging, remain challenging or impossible to apply. This, in turn, poses a challenge for deep learning decoding techniques. To overcome this, we introduce MEIcoder, a biologically informed decoding method that leverages neuron-specific most exciting inputs (MEIs), a structural similarity index measure loss, and adversarial training. MEIcoder achieves state-of-the-art performance in reconstructing visual stimuli from single-cell activity in primary visual cortex (V1), especially excelling on small datasets with fewer recorded neurons. Using ablation studies, we demonstrate that MEIs are the main drivers of the performance, and in scaling experiments, we show that MEIcoder can reconstruct high-fidelity natural-looking images from as few as 1,000-2,500 neurons and less than 1,000 training data points. We also propose a unified benchmark with over 160,000 samples to foster future research. Our results demonstrate the feasibility of reliable decoding in early visual system and provide practical insights for neuroscience and neuroengineering applications.", "date": "2025-10-23", "categories": []}
{"id": "2507.09024", "title": "CNeuroMod-THINGS, a densely-sampled fMRI dataset for visual neuroscience", "abstract": "Data-hungry neuro-AI modelling requires ever larger neuroimaging datasets. CNeuroMod-THINGS meets this need by capturing neural representations for a wide set of semantic concepts using well-characterized images in a new densely-sampled, large-scale fMRI dataset. Importantly, CNeuroMod-THINGS exploits synergies between two existing projects: the THINGS initiative (THINGS) and the Courtois Project on Neural Modelling (CNeuroMod). THINGS has developed a common set of thoroughly annotated images broadly sampling natural and man-made objects which is used to acquire a growing collection of large-scale multimodal neural responses. Meanwhile, CNeuroMod is acquiring hundreds of hours of fMRI data from a core set of participants during controlled and naturalistic tasks, including visual tasks like movie watching and videogame playing. For CNeuroMod-THINGS, four CNeuroMod participants each completed 33-36 sessions of a continuous recognition paradigm using approximately 4000 images from the THINGS stimulus set spanning 720 categories. We report behavioural and neuroimaging metrics that showcase the quality of the data. By bridging together large existing resources, CNeuroMod-THINGS expands our capacity to model broad slices of the human visual experience.", "date": "2025-07-11", "categories": []}
{"id": "2510.17833", "title": "Brain-Language Model Alignment: Insights into the Platonic Hypothesis and Intermediate-Layer Advantage", "abstract": "Do brains and language models converge toward the same internal representations of the world? Recent years have seen a rise in studies of neural activations and model alignment. In this work, we review 25 fMRI-based studies published between 2023 and 2025 and explicitly confront their findings with two key hypotheses: (i) the Platonic Representation Hypothesis -- that as models scale and improve, they converge to a representation of the real world, and (ii) the Intermediate-Layer Advantage -- that intermediate (mid-depth) layers often encode richer, more generalizable features. Our findings provide converging evidence that models and brains may share abstract representational structures, supporting both hypotheses and motivating further research on brain-model alignment.", "date": "2025-10-03", "categories": []}
{"id": "2510.18080", "title": "MEG-GPT: A transformer-based foundation model for magnetoencephalography data", "abstract": "Modelling the complex spatiotemporal patterns of large-scale brain dynamics is crucial for neuroscience, but traditional methods fail to capture the rich structure in modalities such as magnetoencephalography (MEG). Recent advances in deep learning have enabled significant progress in other domains, such as language and vision, by using foundation models at scale. Here, we introduce MEG-GPT, a transformer based foundation model that uses time-attention and next time-point prediction. To facilitate this, we also introduce a novel data-driven tokeniser for continuous MEG data, which preserves the high temporal resolution of continuous MEG signals without lossy transformations. We trained MEG-GPT on tokenised brain region time-courses extracted from a large-scale MEG dataset (N=612, eyes-closed rest, Cam-CAN data), and show that the learnt model can generate data with realistic spatio-spectral properties, including transient events and population variability. Critically, it performs well in downstream decoding tasks, improving downstream supervised prediction task, showing improved zero-shot generalisation across sessions (improving accuracy from 0.54 to 0.59) and subjects (improving accuracy from 0.41 to 0.49) compared to a baseline methods. Furthermore, we show the model can be efficiently fine-tuned on a smaller labelled dataset to boost performance in cross-subject decoding scenarios. This work establishes a powerful foundation model for electrophysiological data, paving the way for applications in computational neuroscience and neural decoding.", "date": "2025-10-20", "categories": []}
{"id": "2303.04203", "title": "What is Memory? A Homological Perspective", "abstract": "We introduce the delta-homology model of memory, a unified framework in which recall, learning, and prediction emerge from cycle closure, the completion of topologically constrained trajectories within the brain's latent manifold. A Dirac-like memory trace corresponds to a nontrivial homology generator, representing a sparse, irreducible attractor that reactivates only when inference trajectories close upon themselves. In this view, memory is not a static attractor landscape but a topological process of recurrence, where structure arises through the stabilization of closed loops. Building on this principle, we represent spike-timing dynamics as spatiotemporal complexes, in which temporally consistent transitions among neurons form chain complexes supporting persistent activation cycles. These cycles are organized into cell posets, compact causal representations that encode overlapping and compositional memory traces. Within this construction, learning and recall correspond to cycle closure under contextual modulation: inference trajectories stabilize into nontrivial homology classes when both local synchrony (context) and global recurrence (content) are satisfied. We formalize this mechanism through the Context-Content Uncertainty Principle (CCUP), which states that cognition minimizes joint uncertainty between a high-entropy context variable and a low-entropy content variable. Synchronization acts as a context filter selecting coherent subnetworks, while recurrence acts as a content filter validating nontrivial cycles.", "date": "2023-03-07", "categories": []}
{"id": "2311.06928", "title": "Attention for Causal Relationship Discovery from Biological Neural Dynamics", "abstract": "This paper explores the potential of the transformer models for learning Granger causality in networks with complex nonlinear dynamics at every node, as in neurobiological and biophysical networks. Our study primarily focuses on a proof-of-concept investigation based on simulated neural dynamics, for which the ground-truth causality is known through the underlying connectivity matrix. For transformer models trained to forecast neuronal population dynamics, we show that the cross attention module effectively captures the causal relationship among neurons, with an accuracy equal or superior to that for the most popular Granger causality analysis method. While we acknowledge that real-world neurobiology data will bring further challenges, including dynamic connectivity and unobserved variability, this research offers an encouraging preliminary glimpse into the utility of the transformer model for causal representation learning in neuroscience.", "date": "2023-11-12", "categories": []}
{"id": "2510.17318", "title": "CausalMamba: Scalable Conditional State Space Models for Neural Causal Inference", "abstract": "We introduce CausalMamba, a scalable framework that addresses fundamental limitations in fMRI-based causal inference: the ill-posed nature of inferring neural causality from hemodynamically distorted BOLD signals and the computational intractability of existing methods like Dynamic Causal Modeling (DCM). Our approach decomposes this complex inverse problem into two tractable stages: BOLD deconvolution to recover latent neural activity, followed by causal graph inference using a novel Conditional Mamba architecture. On simulated data, CausalMamba achieves 37% higher accuracy than DCM. Critically, when applied to real task fMRI data, our method recovers well-established neural pathways with 88% fidelity, whereas conventional approaches fail to identify these canonical circuits in over 99% of subjects. Furthermore, our network analysis of working memory data reveals that the brain strategically shifts its primary causal hub-recruiting executive or salience networks depending on the stimulus-a sophisticated reconfiguration that remains undetected by traditional methods. This work provides neuroscientists with a practical tool for large-scale causal inference that captures both fundamental circuit motifs and flexible network dynamics underlying cognitive function.", "date": "2025-10-20", "categories": []}
{"id": "2510.14486", "title": "Semantic representations emerge in biologically inspired ensembles of cross-supervising neural networks", "abstract": "Brains learn to represent information from a large set of stimuli, typically by weak supervision. Unsupervised learning is therefore a natural approach for exploring the design of biological neural networks and their computations. Accordingly, redundancy reduction has been suggested as a prominent design principle of neural encoding, but its ``mechanistic'' biological implementation is unclear. Analogously, unsupervised training of artificial neural networks yields internal representations that allow for accurate stimulus classification or decoding, but typically rely on biologically-implausible implementations. We suggest that interactions between parallel subnetworks in the brain may underlie such learning: we present a model of representation learning by ensembles of neural networks, where each network learns to encode stimuli into an abstract representation space by cross-supervising interactions with other networks, for inputs they receive simultaneously or in close temporal proximity. Aiming for biological plausibility, each network has a small ``receptive field'', thus receiving a fixed part of the external input, and the networks do not share weights. We find that for different types of network architectures, and for both visual or neuronal stimuli, these cross-supervising networks learn semantic representations that are easily decodable and that decoding accuracy is comparable to supervised networks -- both at the level of single networks and the ensemble. We further show that performance is optimal for small receptive fields, and that sparse connectivity between networks is nearly as accurate as all-to-all interactions, with far fewer computations. We thus suggest a sparsely interacting collective of cross-supervising networks as an algorithmic framework for representational learning and collective computation in the brain.", "date": "2025-10-16", "categories": []}
{"id": "2505.18185", "title": "BrainOmni: A Brain Foundation Model for Unified EEG and MEG Signals", "abstract": "Electroencephalography (EEG) and magnetoencephalography (MEG) measure neural activity non-invasively by capturing electromagnetic fields generated by dendritic currents. Although rooted in the same biophysics, EEG and MEG exhibit distinct signal patterns, further complicated by variations in sensor configurations across modalities and recording devices. Existing approaches typically rely on separate, modality- and dataset-specific models, which limits the performance and cross-domain scalability. This paper proposes BrainOmni, the first brain foundation model that generalises across heterogeneous EEG and MEG recordings. To unify diverse data sources, we introduce BrainTokenizer,the first tokenizer that quantises spatiotemporal brain activity into discrete representations. Central to BrainTokenizer is a novel Sensor Encoder that encodes sensor properties such as spatial layout, orientation, and type, enabling compatibility across devices and modalities. Building upon the discrete representations, BrainOmni learns unified semantic embeddings of brain signals by self-supervised pretraining. To the best of our knowledge, it is the first foundation model to support both EEG and MEG signals, as well as the first to incorporate large-scale MEG pretraining. A total of 1,997 hours of EEG and 656 hours of MEG data are curated and standardised from publicly available sources for pretraining. Experiments show that BrainOmni outperforms both existing foundation models and state-of-the-art task-specific models on a range of downstream tasks. It also demonstrates strong generalisation to unseen EEG and MEG devices. Further analysis reveals that joint EEG-MEG (EMEG) training yields consistent improvements across both modalities. Code and checkpoints are publicly available at https://github.com/OpenTSLab/BrainOmni.", "date": "2025-05-18", "categories": []}
{"id": "2510.13433", "title": "Beyond Pixels: A Differentiable Pipeline for Probing Neuronal Selectivity in 3D", "abstract": "Visual perception relies on inference of 3D scene properties such as shape, pose, and lighting. To understand how visual sensory neurons enable robust perception, it is crucial to characterize their selectivity to such physically interpretable factors. However, current approaches mainly operate on 2D pixels, making it difficult to isolate selectivity for physical scene properties. To address this limitation, we introduce a differentiable rendering pipeline that optimizes deformable meshes to obtain MEIs directly in 3D. The method parameterizes mesh deformations with radial basis functions and learns offsets and scales that maximize neuronal responses while enforcing geometric regularity. Applied to models of monkey area V4, our approach enables probing neuronal selectivity to interpretable 3D factors such as pose and lighting. This approach bridges inverse graphics with systems neuroscience, offering a way to probe neural selectivity with physically grounded, 3D stimuli beyond conventional pixel-based methods.", "date": "2025-10-15", "categories": []}
{"id": "2510.11984", "title": "Learning by Steering the Neural Dynamics: A Statistical Mechanics Perspective", "abstract": "Despite the striking successes of deep neural networks trained with gradient-based optimization, these methods differ fundamentally from their biological counterparts. This gap raises key questions about how nature achieves robust, sample-efficient learning at minimal energy costs and solves the credit-assignment problem without backpropagation. We take a step toward bridging contemporary AI and computational neuroscience by studying how neural dynamics can support fully local, distributed learning that scales to simple machine-learning benchmarks. Using tools from statistical mechanics, we identify conditions for the emergence of robust dynamical attractors in random asymmetric recurrent networks. We derive a closed-form expression for the number of fixed points as a function of self-coupling strength, and we reveal a phase transition in their structure: below a critical self-coupling, isolated fixed points coexist with exponentially many narrow clusters showing the overlap-gap property; above it, subdominant yet dense and extensive clusters appear. These fixed points become accessible, including to a simple asynchronous dynamical rule, after an algorithm-dependent self-coupling threshold. Building on this analysis, we propose a biologically plausible algorithm for supervised learning with any binary recurrent network. Inputs are mapped to fixed points of the dynamics, by relaxing under transient external stimuli and stabilizing the resulting configurations via local plasticity. We show that our algorithm can learn an entangled version of MNIST, leverages depth to develop hierarchical representations and increase hetero-association capacity, and is applicable to several architectures. Finally, we highlight the strong connection between algorithm performance and the unveiled phase transition, and we suggest a cortex-inspired alternative to self-couplings for its emergence.", "date": "2025-10-13", "categories": []}
{"id": "2510.12355", "title": "Fine-grained Analysis of Brain-LLM Alignment through Input Attribution", "abstract": "Understanding the alignment between large language models (LLMs) and human brain activity can reveal computational principles underlying language processing. We introduce a fine-grained input attribution method to identify the specific words most important for brain-LLM alignment, and leverage it to study a contentious research question about brain-LLM alignment: the relationship between brain alignment (BA) and next-word prediction (NWP). Our findings reveal that BA and NWP rely on largely distinct word subsets: NWP exhibits recency and primacy biases with a focus on syntax, while BA prioritizes semantic and discourse-level information with a more targeted recency effect. This work advances our understanding of how LLMs relate to human language processing and highlights differences in feature reliance between BA and NWP. Beyond this study, our attribution method can be broadly applied to explore the cognitive relevance of model predictions in diverse language processing tasks.", "date": "2025-10-14", "categories": []}
{"id": "2505.18361", "title": "Task-Optimized Convolutional Recurrent Networks Align with Tactile Processing in the Rodent Brain", "abstract": "Tactile sensing remains far less understood in neuroscience and less effective in artificial systems compared to more mature modalities such as vision and language. We bridge these gaps by introducing a novel Encoder-Attender-Decoder (EAD) framework to systematically explore the space of task-optimized temporal neural networks trained on realistic tactile input sequences from a customized rodent whisker-array simulator. We identify convolutional recurrent neural networks (ConvRNNs) as superior encoders to purely feedforward and state-space architectures for tactile categorization. Crucially, these ConvRNN-encoder-based EAD models achieve neural representations closely matching rodent somatosensory cortex, saturating the explainable neural variability and revealing a clear linear relationship between supervised categorization performance and neural alignment. Furthermore, contrastive self-supervised ConvRNN-encoder-based EADs, trained with tactile-specific augmentations, match supervised neural fits, serving as an ethologically-relevant, label-free proxy.\n  For neuroscience, our findings highlight nonlinear recurrent processing as important for general-purpose tactile representations in somatosensory cortex, providing the first quantitative characterization of the underlying inductive biases in this system. For embodied AI, our results emphasize the importance of recurrent EAD architectures to handle realistic tactile inputs, along with tailored self-supervised learning methods for achieving robust tactile perception with the same type of sensors animals use to sense in unstructured environments.", "date": "2025-05-23", "categories": []}
{"id": "2502.16456", "title": "Language learning shapes visual category-selectivity in deep neural networks", "abstract": "Category-selective regions in the human brain-such as the fusiform face area (FFA), extrastriate body area (EBA), parahippocampal place area (PPA), and visual word form area (VWFA)-support high-level visual recognition. Here, we investigate whether artificial neural networks (ANNs) exhibit analogous category-selective neurons and how these representations are shaped by language experience. Using an fMRI-inspired functional localizer approach, we identified face-, body-, place-, and word-selective neurons in deep networks presented with category images and scrambled controls. Both the purely visual ResNet and a linguistically supervised Lang-Learned ResNet contained category-selective neurons that increased in proportion across layers. However, compared to the vision-only model, the Lang-Learned ResNet showed a greater number but lower specificity of category-selective neurons, along with reduced spatial localization and attenuated activation strength-indicating a shift toward more distributed, semantically aligned coding. These effects were replicated in the large-scale vision-language model CLIP. Together, our findings reveal that language experience systematically reorganizes visual category representations in ANNs, providing a computational parallel to how linguistic context may shape categorical organization in the human brain.", "date": "2025-02-23", "categories": []}
{"id": "2505.01670", "title": "Efficient Multi Subject Visual Reconstruction from fMRI Using Aligned Representations", "abstract": "This work introduces a novel approach to fMRI-based visual image reconstruction using a subject-agnostic common representation space. We show that the brain signals of the subjects can be aligned in this common space during training to form a semantically aligned common brain. This is leveraged to demonstrate that aligning subject-specific lightweight modules to a reference subject is significantly more efficient than traditional end-to-end training methods. Our approach excels in low-data scenarios. We evaluate our methods on different datasets, demonstrating that the common space is subject and dataset-agnostic.", "date": "2025-05-03", "categories": []}
{"id": "2510.05637", "title": "From Neural Activity to Computation: Biological Reservoirs for Pattern Recognition in Digit Classification", "abstract": "In this paper, we present a biologically grounded approach to reservoir computing (RC), in which a network of cultured biological neurons serves as the reservoir substrate. This system, referred to as biological reservoir computing (BRC), replaces artificial recurrent units with the spontaneous and evoked activity of living neurons. A multi-electrode array (MEA) enables simultaneous stimulation and readout across multiple sites: inputs are delivered through a subset of electrodes, while the remaining ones capture the resulting neural responses, mapping input patterns into a high-dimensional biological feature space. We evaluate the system through a case study on digit classification using a custom dataset. Input images are encoded and delivered to the biological reservoir via electrical stimulation, and the corresponding neural activity is used to train a simple linear classifier. To contextualize the performance of the biological system, we also include a comparison with a standard artificial reservoir trained on the same task. The results indicate that the biological reservoir can effectively support classification, highlighting its potential as a viable and interpretable computational substrate. We believe this work contributes to the broader effort of integrating biological principles into machine learning and aligns with the goals of human-inspired vision by exploring how living neural systems can inform the design of efficient and biologically plausible models.", "date": "2025-10-07", "categories": []}
{"id": "2510.03286", "title": "A Biologically Interpretable Cognitive Architecture for Online Structuring of Episodic Memories into Cognitive Maps", "abstract": "Cognitive maps provide a powerful framework for understanding spatial and abstract reasoning in biological and artificial agents. While recent computational models link cognitive maps to hippocampal-entorhinal mechanisms, they often rely on global optimization rules (e.g., backpropagation) that lack biological plausibility. In this work, we propose a novel cognitive architecture for structuring episodic memories into cognitive maps using local, Hebbian-like learning rules, compatible with neural substrate constraints. Our model integrates the Successor Features framework with episodic memories, enabling incremental, online learning through agent-environment interaction. We demonstrate its efficacy in a partially observable grid-world, where the architecture autonomously organizes memories into structured representations without centralized optimization. This work bridges computational neuroscience and AI, offering a biologically grounded approach to cognitive map formation in artificial adaptive agents.", "date": "2025-09-29", "categories": []}
{"id": "2510.03684", "title": "Model-Guided Microstimulation Steers Primate Visual Behavior", "abstract": "Brain stimulation is a powerful tool for understanding cortical function and holds promise for therapeutic interventions in neuropsychiatric disorders. Initial visual prosthetics apply electric microstimulation to early visual cortex which can evoke percepts of simple symbols such as letters. However, these approaches are fundamentally limited by hardware constraints and the low-level representational properties of this cortical region. In contrast, higher-level visual areas encode more complex object representations and therefore constitute a promising target for stimulation - but determining representational targets that reliably evoke object-level percepts constitutes a major challenge. We here introduce a computational framework to causally model and guide stimulation of high-level cortex, comprising three key components: (1) a perturbation module that translates microstimulation parameters into spatial changes to neural activity, (2) topographic models that capture the spatial organization of cortical neurons and thus enable prototyping of stimulation experiments, and (3) a mapping procedure that links model-optimized stimulation sites back to primate cortex. Applying this framework in two macaque monkeys performing a visual recognition task, model-predicted stimulation experiments produced significant in-vivo changes in perceptual choices. Per-site model predictions and monkey behavior were strongly correlated, underscoring the promise of model-guided stimulation. Image generation further revealed a qualitative similarity between in-silico stimulation of face-selective sites and a patient's report of facephenes. This proof-of-principle establishes a foundation for model-guided microstimulation and points toward next-generation visual prosthetics capable of inducing more complex visual experiences.", "date": "2025-10-04", "categories": []}
{"id": "2510.02354", "title": "Modeling the language cortex with form-independent and enriched representations of sentence meaning reveals remarkable semantic abstractness", "abstract": "The human language system represents both linguistic forms and meanings, but the abstractness of the meaning representations remains debated. Here, we searched for abstract representations of meaning in the language cortex by modeling neural responses to sentences using representations from vision and language models. When we generate images corresponding to sentences and extract vision model embeddings, we find that aggregating across multiple generated images yields increasingly accurate predictions of language cortex responses, sometimes rivaling large language models. Similarly, averaging embeddings across multiple paraphrases of a sentence improves prediction accuracy compared to any single paraphrase. Enriching paraphrases with contextual details that may be implicit (e.g., augmenting \"I had a pancake\" to include details like \"maple syrup\") further increases prediction accuracy, even surpassing predictions based on the embedding of the original sentence, suggesting that the language system maintains richer and broader semantic representations than language models. Together, these results demonstrate the existence of highly abstract, form-independent meaning representations within the language cortex.", "date": "2025-09-27", "categories": []}
{"id": "2510.03155", "title": "Stimulus-Voltage-Based Prediction of Action Potential Onset Timing: Classical vs. Quantum-Inspired Approaches", "abstract": "Accurate modeling of neuronal action potential (AP) onset timing is crucial for understanding neural coding of danger signals. Traditional leaky integrate-and-fire (LIF) models, while widely used, exhibit high relative error in predicting AP onset latency, especially under strong or rapidly changing stimuli. Inspired by recent experimental findings and quantum theory, we present a quantum-inspired leaky integrate-and-fire (QI-LIF) model that treats AP onset as a probabilistic event, represented by a Gaussian wave packet in time. This approach captures the biological variability and uncertainty inherent in neuronal firing. We systematically compare the relative error of AP onset predictions between the classical LIF and QI-LIF models using synthetic data from hippocampal and sensory neurons subjected to varying stimulus amplitudes. Our results demonstrate that the QI-LIF model significantly reduces prediction error, particularly for high-intensity stimuli, aligning closely with observed biological responses. This work highlights the potential of quantum-inspired computational frameworks in advancing the accuracy of neural modeling and has implications for quantum engineering approaches to brain-inspired computing.", "date": "2025-10-03", "categories": []}
{"id": "2501.14615", "title": "Integration of Calcium Imaging Traces via Deep Generative Modeling", "abstract": "Calcium imaging allows for the parallel measurement of large neuronal populations in a spatially resolved and minimally invasive manner, and has become a gold-standard for neuronal functionality. While deep generative models have been successfully applied to study the activity of neuronal ensembles, their potential for learning single-neuron representations from calcium imaging fluorescence traces remains largely unexplored, and batch effects remain an important hurdle. To address this, we explore supervised variational autoencoder architectures that learn compact representations of individual neurons from fluorescent traces without relying on spike inference algorithms. We find that this approach outperforms state-of-the-art models, preserving biological variability while mitigating batch effects. Across simulated and experimental datasets, this framework enables robust visualization, clustering, and interpretation of single-neuron dynamics.", "date": "2025-01-24", "categories": []}
{"id": "2509.24039", "title": "End-to-end Topographic Auditory Models Replicate Signatures of Human Auditory Cortex", "abstract": "The human auditory cortex is topographically organized. Neurons with similar response properties are spatially clustered, forming smooth maps for acoustic features such as frequency in early auditory areas, and modular regions selective for music and speech in higher-order cortex. Yet, evaluations for current computational models of auditory perception do not measure whether such topographic structure is present in a candidate model. Here, we show that cortical topography is not present in the previous best-performing models at predicting human auditory fMRI responses. To encourage the emergence of topographic organization, we adapt a cortical wiring-constraint loss originally designed for visual perception. The new class of topographic auditory models, TopoAudio, are trained to classify speech, and environmental sounds from cochleagram inputs, with an added constraint that nearby units on a 2D cortical sheet develop similar tuning. Despite these additional constraints, TopoAudio achieves high accuracy on benchmark tasks comparable to the unconstrained non-topographic baseline models. Further, TopoAudio predicts the fMRI responses in the brain as well as standard models, but unlike standard models, TopoAudio develops smooth, topographic maps for tonotopy and amplitude modulation (common properties of early auditory representation, as well as clustered response modules for music and speech (higher-order selectivity observed in the human auditory cortex). TopoAudio is the first end-to-end biologically grounded auditory model to exhibit emergent topography, and our results emphasize that a wiring-length constraint can serve as a general-purpose regularization tool to achieve biologically aligned representations.", "date": "2025-09-28", "categories": []}
{"id": "2509.24919", "title": "Meta-Learning Theory-Informed Inductive Biases using Deep Kernel Gaussian Processes", "abstract": "Normative and task-driven theories offer powerful top-down explanations for biological systems, yet the goals of quantitatively arbitrating between competing theories, and utilizing them as inductive biases to improve data-driven fits of real biological datasets are prohibitively laborious, and often impossible. To this end, we introduce a Bayesian meta-learning framework designed to automatically convert raw functional predictions from normative theories into tractable probabilistic models. We employ adaptive deep kernel Gaussian processes, meta-learning a kernel on synthetic data generated from a normative theory. This Theory-Informed Kernel specifies a probabilistic model representing the theory predictions -- usable for both fitting data and rigorously validating the theory. As a demonstration, we apply our framework to the early visual system, using efficient coding as our normative theory. We show improved response prediction accuracy in ex vivo recordings of mouse retinal ganglion cells stimulated by natural scenes compared to conventional data-driven baselines, while providing well-calibrated uncertainty estimates and interpretable representations. Using exact Bayesian model selection, we also show that our informed kernel can accurately infer the degree of theory-match from data, confirming faithful encapsulation of theory structure. This work provides a more general, scalable, and automated approach for integrating theoretical knowledge into data-driven scientific inquiry in neuroscience and beyond.", "date": "2025-09-29", "categories": []}
{"id": "2509.21340", "title": "Cycle is All You Need: More Is Different", "abstract": "We propose an information-topological framework in which cycle closure is the fundamental mechanism of memory and consciousness. Memory is not a static store but the ability to re-enter latent cycles in neural state space, with invariant cycles serving as carriers of meaning by filtering order-specific noise and preserving what persists across contexts. The dot-cycle dichotomy captures this: transient dots scaffold exploration, while nontrivial cycles encode low-entropy content invariants that stabilize memory. Biologically, polychronous neural groups realize 1-cycles through delay-locked spiking reinforced by STDP, nested within theta-gamma rhythms that enforce boundary cancellation. These micro-cycles compose hierarchically, extending navigation loops into general memory and cognition. The perception-action cycle introduces high-order invariance: closure holds even across sense-act alternations, generalizing ancestral homing behavior. Sheaf-cosheaf duality formalizes this process: sheaves glue perceptual fragments into global sections, cosheaves decompose global plans into actions and closure aligns top-down predictions with bottom-up cycles. Consciousness then arises as the persistence of high-order invariants that integrate (unity) yet differentiate (richness) across contexts. We conclude that cycle is all you need: persistent invariants enable generalization in non-ergodic environments with long-term coherence at minimal energetic cost.", "date": "2025-09-15", "categories": []}
{"id": "2509.21381", "title": "Toward a Realistic Encoding Model of Auditory Affective Understanding in the Brain", "abstract": "In affective neuroscience and emotion-aware AI, understanding how complex auditory stimuli drive emotion arousal dynamics remains unresolved. This study introduces a computational framework to model the brain's encoding of naturalistic auditory inputs into dynamic behavioral/neural responses across three datasets (SEED, LIRIS, self-collected BAVE). Guided by neurobiological principles of parallel auditory hierarchy, we decompose audio into multilevel auditory features (through classical algorithms and wav2vec 2.0/Hubert) from the original and isolated human voice/background soundtrack elements, mapping them to emotion-related responses via cross-dataset analyses. Our analysis reveals that high-level semantic representations (derived from the final layer of wav2vec 2.0/Hubert) exert a dominant role in emotion encoding, outperforming low-level acoustic features with significantly stronger mappings to behavioral annotations and dynamic neural synchrony across most brain regions ($p < 0.05$). Notably, middle layers of wav2vec 2.0/hubert (balancing acoustic-semantic information) surpass the final layers in emotion induction across datasets. Moreover, human voices and soundtracks show dataset-dependent emotion-evoking biases aligned with stimulus energy distribution (e.g., LIRIS favors soundtracks due to higher background energy), with neural analyses indicating voices dominate prefrontal/temporal activity while soundtracks excel in limbic regions. By integrating affective computing and neuroscience, this work uncovers hierarchical mechanisms of auditory-emotion encoding, providing a foundation for adaptive emotion-aware systems and cross-disciplinary explorations of audio-affective interactions.", "date": "2025-09-23", "categories": []}
{"id": "2503.13212", "title": "MAME: Multidimensional Adaptive Metamer Exploration with Human Perceptual Feedback", "abstract": "Alignment between human brain networks and artificial models has become an active research area in vision science and machine learning. A widely adopted approach is identifying \"metamers,\" stimuli physically different yet perceptually equivalent within a system. However, conventional methods lack a direct approach to searching for the human metameric space. Instead, researchers first develop biologically inspired models and then infer about human metamers indirectly by testing whether model metamers also appear as metamers to humans. Here, we propose the Multidimensional Adaptive Metamer Exploration (MAME) framework, enabling direct, high-dimensional exploration of human metameric spaces through online image generation guided by human perceptual feedback. MAME modulates reference images across multiple dimensions based on hierarchical neural network responses, adaptively updating generation parameters according to participants' perceptual discriminability. Using MAME, we successfully measured multidimensional human metameric spaces within a single psychophysical experiment. Experimental results using a biologically plausible CNN model showed that human discrimination sensitivity was lower for metameric images based on low-level features compared to high-level features, which image contrast metrics could not explain. The finding suggests a relatively worse alignment between the metameric spaces of humans and the CNN model for low-level processing compared to high-level processing. Counterintuitively, given recent discussions on alignment at higher representational levels, our results highlight the importance of early visual computations in shaping biologically plausible models. Our MAME framework can serve as a future scientific tool for directly investigating the functional organization of human vision.", "date": "2025-03-17", "categories": []}
{"id": "2509.19351", "title": "The Impact of Structural Changes on Learning Capacity in the Fly Olfactory Neural Circuit", "abstract": "The Drosophila mushroom body (MB) is known to be involved in olfactory learning and memory; the synaptic plasticity of the Kenyon cell (KC) to mushroom body output neuron (MBON) synapses plays a key role in the learning process. Previous research has focused on projection neuron (PN) to Kenyon cell (KC) connectivity within the MB; we examine how perturbations to the mushroom body circuit structure and changes in connectivity, specifically within the KC to mushroom body output neuron (MBON) neural circuit, affect the MBONs' ability to distinguish between odor classes. We constructed a neural network that incorporates the connectivity between PNs, KCs, and MBONs. To train our model, we generated ten artificial input classes, which represent the projection neuron activity in response to different odors. We collected data on the number of KC-to-MBON connections, MBON error rates, and KC-to-MBON synaptic weights, among other metrics. We observed that MBONs with very few presynaptic KCs consistently performed worse than others in the odor classification task. The developmental types of KCs also played a significant role in each MBON's output. We performed random and targeted KC ablation and observed that ablating developmentally mature KCs had a greater negative impact on MBONs' learning capacity than ablating immature KCs. Random and targeted pruning of KC-MBON synaptic connections yielded results largely consistent with the ablation experiments. To further explore the various types of KCs, we also performed rewiring experiments in the PN to KC circuit. Our study furthers our understanding of olfactory neuroplasticity and provides important clues to understanding learning and memory in general. Understanding how the olfactory circuits process and learn can also have potential applications in artificial intelligence and treatments for neurodegenerative diseases.", "date": "2025-09-18", "categories": []}
{"id": "2509.18507", "title": "Dynamical Modeling of Behaviorally Relevant Spatiotemporal Patterns in Neural Imaging Data", "abstract": "High-dimensional imaging of neural activity, such as widefield calcium and functional ultrasound imaging, provide a rich source of information for understanding the relationship between brain activity and behavior. Accurately modeling neural dynamics in these modalities is crucial for understanding this relationship but is hindered by the high-dimensionality, complex spatiotemporal dependencies, and prevalent behaviorally irrelevant dynamics in these modalities. Existing dynamical models often employ preprocessing steps to obtain low-dimensional representations from neural image modalities. However, this process can discard behaviorally relevant information and miss spatiotemporal structure. We propose SBIND, a novel data-driven deep learning framework to model spatiotemporal dependencies in neural images and disentangle their behaviorally relevant dynamics from other neural dynamics. We validate SBIND on widefield imaging datasets, and show its extension to functional ultrasound imaging, a recent modality whose dynamical modeling has largely remained unexplored. We find that our model effectively identifies both local and long-range spatial dependencies across the brain while also dissociating behaviorally relevant neural dynamics. Doing so, SBIND outperforms existing models in neural-behavioral prediction. Overall, SBIND provides a versatile tool for investigating the neural mechanisms underlying behavior using imaging modalities.", "date": "2025-09-23", "categories": []}
{"id": "2509.18758", "title": "Complexity of Activity Patterns in a Bio-Inspired Hopfield-Type Network in Different Topologies", "abstract": "Neural network models capable of storing memory have been extensively studied in computer science and computational neuroscience. The Hopfield network is a prototypical example of a model designed for associative, or content-addressable, memory and has been analyzed in many forms. Further, ideas and methods from complex network theory have been incorporated into artificial neural networks and learning, emphasizing their structural properties. Nevertheless, the temporal dynamics also play a vital role in biological neural networks, whose temporal structure is a crucial feature to examine. Biological neural networks display complex intermittency and, thus, can be studied through the lens of the temporal complexity (TC) theory. The TC approach look at the metastability of self-organized states, characterized by a power-law decay in the inter-event time distribution and in the total activity distribution or a scaling behavior in the corresponding event-driven diffusion processes. In this study, we present a temporal complexity (TC) analysis of a biologically-inspired Hopfield-type neural network model. We conducted a comparative assessment between scale-free and random network topologies, with particular emphasis on their global activation patterns. Our parametric analysis revealed comparable dynamical behaviors across both neural network architectures. Furthermore, our investigation into temporal complexity characteristics uncovered that seemingly distinct dynamical patterns exhibit similar temporal complexity behaviors. In particular, similar power-law decay in the activity distribution and similar complexity levels are observed in both topologies, but with a much reduced noise in the scale-free topology. Notably, most of the complex dynamical profiles were consistently observed in scale-free network configurations, thus confirming the crucial role of hubs in neural network dynamics.", "date": "2025-09-23", "categories": []}
{"id": "2408.07637", "title": "Synaptic Theory of Chunking in Working Memory", "abstract": "Working memory often appears to exceed its basic span by organizing items into compact representations called chunks. Chunking can be learned over time for familiar inputs; however, it can also arise spontaneously for novel stimuli. Such on-the-fly structuring is crucial for cognition, yet the underlying neural mechanism remains unclear. Here we introduce a synaptic theory of chunking, in which short-term synaptic plasticity enables the formation of chunk representations in working memory. We show that a specialized population of ``chunking neurons'' selectively controls groups of stimulus-responsive neurons, akin to gating. As a result, the network maintains and retrieves the stimuli in chunks, thereby exceeding the basic capacity. Moreover, we show that our model can dynamically construct hierarchical representations within working memory through hierarchical chunking. A consequence of this proposed mechanism is a new limit on the number of items that can be stored and subsequently retrieved from working memory, depending only on the basic working memory capacity when chunking is not invoked. Predictions from our model were confirmed by analyzing single-unit responses in epileptic patients and memory experiments with verbal material. Our work provides a novel conceptual and analytical framework for understanding how the brain organizes information in real time.", "date": "2024-08-14", "categories": []}
{"id": "2509.14360", "title": "Embodied sensorimotor control: computational modeling of the neural control of movement", "abstract": "We review how sensorimotor control is dictated by interacting neural populations, optimal feedback mechanisms, and the biomechanics of bodies. First, we outline the distributed anatomical loops that shuttle sensorimotor signals between cortex, subcortical regions, and spinal cord. We then summarize evidence that neural population activity occupies low-dimensional, dynamically evolving manifolds during planning and execution of movements. Next, we summarize literature explaining motor behavior through the lens of optimal control theory, which clarifies the role of internal models and feedback during motor control. Finally, recent studies on embodied sensorimotor control address gaps within each framework by aiming to elucidate neural population activity through the explicit control of musculoskeletal dynamics. We close by discussing open problems and opportunities: multi-tasking and cognitively rich behavior, multi-regional circuit models, and the level of anatomical detail needed in body and network models. Together, this review and recent advances point towards reaching an integrative account of the neural control of movement.", "date": "2025-09-17", "categories": []}
{"id": "2509.12497", "title": "Prediction and Causality of functional MRI and synthetic signal using a Zero-Shot Time-Series Foundation Model", "abstract": "Time-series forecasting and causal discovery are central in neuroscience, as predicting brain activity and identifying causal relationships between neural populations and circuits can shed light on the mechanisms underlying cognition and disease. With the rise of foundation models, an open question is how they compare to traditional methods for brain signal forecasting and causality analysis, and whether they can be applied in a zero-shot setting. In this work, we evaluate a foundation model against classical methods for inferring directional interactions from spontaneous brain activity measured with functional magnetic resonance imaging (fMRI) in humans. Traditional approaches often rely on Wiener-Granger causality. We tested the forecasting ability of the foundation model in both zero-shot and fine-tuned settings, and assessed causality by comparing Granger-like estimates from the model with standard Granger causality. We validated the approach using synthetic time series generated from ground-truth causal models, including logistic map coupling and Ornstein-Uhlenbeck processes. The foundation model achieved competitive zero-shot forecasting fMRI time series (mean absolute percentage error of 0.55 in controls and 0.27 in patients). Although standard Granger causality did not show clear quantitative differences between models, the foundation model provided a more precise detection of causal interactions.\n  Overall, these findings suggest that foundation models offer versatility, strong zero-shot performance, and potential utility for forecasting and causal discovery in time-series data.", "date": "2025-09-15", "categories": []}
{"id": "2509.13612", "title": "Rest2Visual: Predicting Visually Evoked fMRI from Resting-State Scans", "abstract": "Understanding how spontaneous brain activity relates to stimulus-driven neural responses is a fundamental challenge in cognitive neuroscience. While task-based functional magnetic resonance imaging (fMRI) captures localized stimulus-evoked brain activation, its acquisition is costly, time-consuming, and difficult to scale across populations. In contrast, resting-state fMRI (rs-fMRI) is task-free and abundant, but lacks direct interpretability. We introduce Rest2Visual, a conditional generative model that predicts visually evoked fMRI (ve-fMRI) from resting-state input and 2D visual stimuli. It follows a volumetric encoder--decoder design, where multiscale 3D features from rs-fMRI are modulated by image embeddings via adaptive normalization, enabling spatially accurate, stimulus-specific activation synthesis. To enable model training, we construct a large-scale triplet dataset from the Natural Scenes Dataset (NSD), aligning each rs-fMRI volume with stimulus images and their corresponding ve-fMRI activation maps. Quantitative evaluation shows that the predicted activations closely match ground truth across standard similarity and representational metrics, and support successful image reconstruction in downstream decoding. Notably, the predicted maps preserve subject-specific structure, demonstrating the model's capacity to generate individualized functional surrogates. Our results provide compelling evidence that individualized spontaneous neural activity can be transformed into stimulus-aligned representations, opening new avenues for scalable, task-free functional brain modeling.", "date": "2025-09-17", "categories": []}
{"id": "2208.09677", "title": "Net2Brain: A Toolbox to compare artificial vision models with human brain responses", "abstract": "We introduce Net2Brain, a graphical and command-line user interface toolbox for comparing the representational spaces of artificial deep neural networks (DNNs) and human brain recordings. While different toolboxes facilitate only single functionalities or only focus on a small subset of supervised image classification models, Net2Brain allows the extraction of activations of more than 600 DNNs trained to perform a diverse range of vision-related tasks (e.g semantic segmentation, depth estimation, action recognition, etc.), over both image and video datasets. The toolbox computes the representational dissimilarity matrices (RDMs) over those activations and compares them to brain recordings using representational similarity analysis (RSA), weighted RSA, both in specific ROIs and with searchlight search. In addition, it is possible to add a new data set of stimuli and brain recordings to the toolbox for evaluation. We demonstrate the functionality and advantages of Net2Brain with an example showcasing how it can be used to test hypotheses of cognitive computational neuroscience.", "date": "2022-08-20", "categories": []}
{"id": "2509.03521", "title": "BiND: A Neural Discriminator-Decoder for Accurate Bimanual Trajectory Prediction in Brain-Computer Interfaces", "abstract": "Decoding bimanual hand movements from intracortical recordings remains a critical challenge for brain-computer interfaces (BCIs), due to overlapping neural representations and nonlinear interlimb interactions. We introduce BiND (Bimanual Neural Discriminator-Decoder), a two-stage model that first classifies motion type (unimanual left, unimanual right, or bimanual) and then uses specialized GRU-based decoders, augmented with a trial-relative time index, to predict continuous 2D hand velocities. We benchmark BiND against six state-of-the-art models (SVR, XGBoost, FNN, CNN, Transformer, GRU) on a publicly available 13-session intracortical dataset from a tetraplegic patient. BiND achieves a mean $R^2$ of 0.76 ($\\pm$0.01) for unimanual and 0.69 ($\\pm$0.03) for bimanual trajectory prediction, surpassing the next-best model (GRU) by 2% in both tasks. It also demonstrates greater robustness to session variability than all other benchmarked models, with accuracy improvements of up to 4% compared to GRU in cross-session analyses. This highlights the effectiveness of task-aware discrimination and temporal modeling in enhancing bimanual decoding.", "date": "2025-08-19", "categories": []}
{"id": "2402.10251", "title": "BrainWave: A Brain Signal Foundation Model for Clinical Applications", "abstract": "Neural electrical activity is fundamental to brain function, underlying a range of cognitive and behavioral processes, including movement, perception, decision-making, and consciousness. Abnormal patterns of neural signaling often indicate the presence of underlying brain diseases. The variability among individuals, the diverse array of clinical symptoms from various brain disorders, and the limited availability of diagnostic classifications, have posed significant barriers to formulating reliable model of neural signals for diverse application contexts. Here, we present BrainWave, the first foundation model for both invasive and non-invasive neural recordings, pretrained on more than 40,000 hours of electrical brain recordings (13.79 TB of data) from approximately 16,000 individuals. Our analysis show that BrainWave outperforms all other competing models and consistently achieves state-of-the-art performance in the diagnosis and identification of neurological disorders. We also demonstrate robust capabilities of BrainWave in enabling zero-shot transfer learning across varying recording conditions and brain diseases, as well as few-shot classification without fine-tuning, suggesting that BrainWave learns highly generalizable representations of neural signals. We hence believe that open-sourcing BrainWave will facilitate a wide range of clinical applications in medicine, paving the way for AI-driven approaches to investigate brain disorders and advance neuroscience research.", "date": "2024-02-15", "categories": []}
{"id": "2405.18808", "title": "BRACTIVE: A Brain Activation Approach to Human Visual Brain Learning", "abstract": "The human brain is a highly efficient processing unit, and understanding how it works can inspire new algorithms and architectures in machine learning. In this work, we introduce a novel framework named Brain Activation Network (BRACTIVE), a transformer-based approach to studying the human visual brain. The primary objective of BRACTIVE is to align the visual features of subjects with their corresponding brain representations using functional Magnetic Resonance Imaging (fMRI) signals. It enables us to identify the brain's Regions of Interest (ROIs) in the subjects. Unlike previous brain research methods, which can only identify ROIs for one subject at a time and are limited by the number of subjects, BRACTIVE automatically extends this identification to multiple subjects and ROIs. Our experiments demonstrate that BRACTIVE effectively identifies person-specific regions of interest, such as face and body-selective areas, aligning with neuroscience findings and indicating potential applicability to various object categories. More importantly, we found that leveraging human visual brain activity to guide deep neural networks enhances performance across various benchmarks. It encourages the potential of BRACTIVE in both neuroscience and machine intelligence studies.", "date": "2024-05-29", "categories": []}
{"id": "2411.07388", "title": "Firing Rate Models as Associative Memory: Excitatory-Inhibitory Balance for Robust Retrieval", "abstract": "Firing rate models are dynamical systems widely used in applied and theoretical neuroscience to describe local cortical dynamics in neuronal populations. By providing a macroscopic perspective of neuronal activity, these models are essential for investigating oscillatory phenomena, chaotic behavior, and associative memory processes. Despite their widespread use, the application of firing rate models to associative memory networks has received limited mathematical exploration, and most existing studies are focused on specific models. Conversely, well-established associative memory designs, such as Hopfield networks, lack key biologically-relevant features intrinsic to firing rate models, including positivity and interpretable synaptic matrices that reflect excitatory and inhibitory interactions. To address this gap, we propose a general framework that ensures the emergence of re-scaled memory patterns as stable equilibria in the firing rate dynamics. Furthermore, we analyze the conditions under which the memories are locally and globally asymptotically stable, providing insights into constructing biologically-plausible and robust systems for associative memory retrieval.", "date": "2024-11-11", "categories": []}
{"id": "2505.12408", "title": "ViEEG: Hierarchical Visual Neural Representation for EEG Brain Decoding", "abstract": "Understanding and decoding brain activity into visual representations is a fundamental challenge at the intersection of neuroscience and artificial intelligence. While EEG visual decoding has shown promise due to its non-invasive, and low-cost nature, existing methods suffer from Hierarchical Neural Encoding Neglect (HNEN)-a critical limitation where flat neural representations fail to model the brain's hierarchical visual processing hierarchy. Inspired by the hierarchical organization of visual cortex, we propose ViEEG, a neuro-We further adopt hierarchical contrastive learning for EEG-CLIP representation alignment, enabling zero-shot object recognition. Extensive experiments on the THINGS-EEG dataset demonstrate that ViEEG significantly outperforms previous methods by a large margin in both subject-dependent and subject-independent settings. Results on the THINGS-MEG dataset further confirm ViEEG's generalization to different neural modalities. Our framework not only advances the performance frontier but also sets a new paradigm for EEG brain decoding. inspired framework that addresses HNEN. ViEEG decomposes each visual stimulus into three biologically aligned components-contour, foreground object, and contextual scene-serving as anchors for a three-stream EEG encoder. These EEG features are progressively integrated via cross-attention routing, simulating cortical information flow from low-level to high-level vision.", "date": "2025-05-18", "categories": []}
{"id": "2509.00602", "title": "TranCIT: Transient Causal Interaction Toolbox", "abstract": "Quantifying transient causal interactions from non-stationary neural signals is a fundamental challenge in neuroscience. Traditional methods are often inadequate for brief neural events, and advanced, event-specific techniques have lacked accessible implementations within the Python ecosystem. Here, we introduce trancit (Transient Causal Interaction Toolbox), an open-source Python package designed to bridge this gap. TranCIT implements a comprehensive analysis pipeline, including Granger Causality, Transfer Entropy, and the more robust Structural Causal Model-based Dynamic Causal Strength (DCS) and relative Dynamic Causal Strength (rDCS) for accurately detecting event-driven causal effects. We demonstrate TranCIT's utility by successfully capturing causality in high-synchrony regimes where traditional methods fail and by identifying the known transient information flow from hippocampal CA3 to CA1 during sharp-wave ripple events in real-world data. The package offers a user-friendly, validated solution for investigating the transient causal dynamics that govern complex systems.", "date": "2025-08-30", "categories": []}
{"id": "2509.01987", "title": "Semantic and episodic memories in a predictive coding model of the neocortex", "abstract": "Complementary Learning Systems theory holds that intelligent agents need two learning systems. Semantic memory is encoded in the neocortex with dense, overlapping representations and acquires structured knowledge. Episodic memory is encoded in the hippocampus with sparse, pattern-separated representations and quickly learns the specifics of individual experiences. Recently, this duality between semantic and episodic memories has been challenged by predictive coding, a biologically plausible neural network model of the neocortex which was shown to have hippocampus-like abilities on auto-associative memory tasks. These results raise the question of the episodic capabilities of the neocortex and their relation to semantic memory. In this paper, we present such a predictive coding model of the neocortex and explore its episodic capabilities. We show that this kind of model can indeed recall the specifics of individual examples but only if it is trained on a small number of examples. The model is overfitted to these exemples and does not generalize well, suggesting that episodic memory can arise from semantic learning. Indeed, a model trained with many more examples loses its recall capabilities. This work suggests that individual examples can be encoded gradually in the neocortex using dense, overlapping representations but only in a limited number, motivating the need for sparse, pattern-separated representations as found in the hippocampus.", "date": "2025-09-02", "categories": []}
{"id": "2410.19779", "title": "BrainGPT: Unleashing the Potential of EEG Generalist Foundation Model by Autoregressive Pre-training", "abstract": "Electroencephalogram (EEG) signals are pivotal in providing insights into spontaneous brain activity, highlighting their significant importance in neuroscience research. However, the exploration of versatile EEG models is constrained by diverse data formats, outdated pre-training paradigms, and limited transfer learning methods, only leading to specialist models on single dataset. In this paper, we introduce EEGPT, the first generalist EEG foundation model designed to address these challenges. First, we propose an electrode-wise modeling strategy that treats each electrode as a fundamental unit, enabling the integration of diverse EEG datasets collected from up to 138 electrodes, amassing 37.5M pre-training samples. Second, we develop the first autoregressive EEG pre-trained model, moving away from traditional masked autoencoder approaches to a next signal prediction task that better captures the sequential and temporal dependencies of EEG data. We also explore scaling laws with model up to 1.1B parameters: the largest in EEG research to date. Third, we introduce a multi-task transfer learning paradigm using a learnable electrode graph network shared across tasks, which for the first time confirms multi-task compatibility and synergy. As the first generalist EEG foundation model, EEGPT shows broad compatibility with various signal acquisition devices, subjects, and tasks. It supports up to 138 electrodes and any combination thereof as input. Furthermore, we simultaneously evaluate it on 5 distinct tasks across 12 benchmarks. EEGPT consistently outperforms existing specialist models across all downstream tasks, with its effectiveness further validated through extensive ablation studies. This work sets a new direction for generalist EEG modeling, offering improved scalability, transferability, and adaptability for a wide range of EEG applications. The code and models will be released.", "date": "2024-10-14", "categories": []}
{"id": "2503.16340", "title": "Human locomotor control timescales depend on the environmental context and sensory input modality", "abstract": "Everyday locomotion is a complex sensorimotor process that can unfold over multiple timescales, from long-term path planning to rapid, reactive adjustments. However, we lack an understanding of how factors such as environmental demands, or the available sensory information simultaneously influence these control timescales. To address this, we present a unified data-driven framework to quantify the control timescales by identifying how early we can predict future actions from past inputs. We apply this framework across tasks including walking and running, environmental contexts including treadmill, overground, and varied terrains, and sensory input modalities including gaze fixations and body states. We find that deep neural network architectures that effectively handle long-range dependencies, specifically Gated Recurrent Units and Transformers, outperform other architectures and widely used linear models when predicting future actions. Our framework reveals the factors that influence locomotor foot placement control timescales. Across environmental contexts, we discover that humans rely more on fast timescale control in more complex terrain. Across input modalities, we find a hierarchy of control timescales where gaze predicts foot placement before full-body states, which predict before center-of-mass states. Our model also identifies mid-swing as a critical phase when the swing foot's state predicts its future placement, with this timescale adapting across environments. Overall, this work offers data-driven insights into locomotor control in everyday settings, offering models that can be integrated with rehabilitation technologies and movement simulations to improve their applicability in everyday settings.", "date": "2025-03-20", "categories": []}
{"id": "2508.19048", "title": "Time Series Analysis of Spiking Neural Systems via Transfer Entropy and Directed Persistent Homology", "abstract": "We present a topological framework for analysing neural time series that integrates Transfer Entropy (TE) with directed Persistent Homology (PH) to characterize information flow in spiking neural systems. TE quantifies directional influence between neurons, producing weighted, directed graphs that reflect dynamic interactions. These graphs are then analyzed using PH, enabling assessment of topological complexity across multiple structural scales and dimensions.\n  We apply this TE+PH pipeline to synthetic spiking networks trained on logic gate tasks, image-classification networks exposed to structured and perturbed inputs, and mouse cortical recordings annotated with behavioral events. Across all settings, the resulting topological signatures reveal distinctions in task complexity, stimulus structure, and behavioral regime. Higher-dimensional features become more prominent in complex or noisy conditions, reflecting interaction patterns that extend beyond pairwise connectivity. Our findings offer a principled approach to mapping directed information flow onto global organizational patterns in both artificial and biological neural systems. The framework is generalizable and interpretable, making it well suited for neural systems with time-resolved and binary spiking data.", "date": "2025-08-26", "categories": []}
{"id": "2405.14425", "title": "When predict can also explain: few-shot prediction to select better neural latents", "abstract": "Latent variable models serve as powerful tools to infer underlying dynamics from observed neural activity. Ideally, the inferred dynamics should align with true ones. However, due to the absence of ground truth data, prediction benchmarks are often employed as proxies. One widely-used method, $\\textit{co-smoothing}$, involves jointly estimating latent variables and predicting observations along held-out channels to assess model performance. In this study, we reveal the limitations of the co-smoothing prediction framework and propose a remedy. Using a student-teacher setup, we demonstrate that models with high co-smoothing can have arbitrary extraneous dynamics in their latent representations. To address this, we introduce a secondary metric -- $\\textit{few-shot co-smoothing}$, performing regression from the latent variables to held-out neurons in the data using fewer trials. Our results indicate that among models with near-optimal co-smoothing, those with extraneous dynamics underperform in the few-shot co-smoothing compared to `minimal' models that are devoid of such dynamics. We provide analytical insights into the origin of this phenomenon and further validate our findings on four standard neural datasets using a state-of-the-art method: STNDT. In the absence of ground truth, we suggest a novel measure to validate our approach. By cross-decoding the latent variables of all model pairs with high co-smoothing, we identify models with minimal extraneous dynamics. We find a correlation between few-shot co-smoothing performance and this new measure. In summary, we present a novel prediction metric designed to yield latent variables that more accurately reflect the ground truth, offering a significant improvement for latent dynamics inference.", "date": "2024-05-23", "categories": []}
{"id": "2508.17561", "title": "Consciousness as a Functor", "abstract": "We propose a novel theory of consciousness as a functor (CF) that receives and transmits contents from unconscious memory into conscious memory. Our CF framework can be seen as a categorial formulation of the Global Workspace Theory proposed by Baars. CF models the ensemble of unconscious processes as a topos category of coalgebras. The internal language of thought in CF is defined as a Multi-modal Universal Mitchell-Benabou Language Embedding (MUMBLE). We model the transmission of information from conscious short-term working memory to long-term unconscious memory using our recently proposed Universal Reinforcement Learning (URL) framework. To model the transmission of information from unconscious long-term memory into resource-constrained short-term memory, we propose a network economic model.", "date": "2025-08-25", "categories": []}
{"id": "2508.18226", "title": "Disentangling the Factors of Convergence between Brains and Computer Vision Models", "abstract": "Many AI models trained on natural images develop representations that resemble those of the human brain. However, the factors that drive this brain-model similarity remain poorly understood. To disentangle how the model, training and data independently lead a neural network to develop brain-like representations, we trained a family of self-supervised vision transformers (DINOv3) that systematically varied these different factors. We compare their representations of images to those of the human brain recorded with both fMRI and MEG, providing high resolution in spatial and temporal analyses. We assess the brain-model similarity with three complementary metrics focusing on overall representational similarity, topographical organization, and temporal dynamics. We show that all three factors - model size, training amount, and image type - independently and interactively impact each of these brain similarity metrics. In particular, the largest DINOv3 models trained with the most human-centric images reach the highest brain-similarity. This emergence of brain-like representations in AI models follows a specific chronology during training: models first align with the early representations of the sensory cortices, and only align with the late and prefrontal representations of the brain with considerably more training. Finally, this developmental trajectory is indexed by both structural and functional properties of the human cortex: the representations that are acquired last by the models specifically align with the cortical areas with the largest developmental expansion, thickness, least myelination, and slowest timescales. Overall, these findings disentangle the interplay between architecture and experience in shaping how artificial neural networks come to see the world as humans do, thus offering a promising framework to understand how the human brain comes to represent its visual world.", "date": "2025-08-25", "categories": []}
{"id": "2411.00983", "title": "Testing Components of the Attention Schema Theory in Artificial Neural Networks", "abstract": "Growing evidence suggests that the brain uses an attention schema, or a simplified model of attention, to help control what it attends to. One proposed benefit of this model is to allow agents to model the attention states of other agents, and thus predict and interact with other agents. The effects of an attention schema may be examined in artificial agents. Although attention mechanisms in artificial agents are different from in biological brains, there may be some principles in common. In both cases, select features or representations are emphasized for better performance. Here, using neural networks with transformer attention mechanisms, we asked whether the addition of an attention schema affected the ability of agents to make judgements about and cooperate with each other. First, we found that an agent with an attention schema is better at categorizing the attention states of other agents (higher accuracy). Second, an agent with an attention schema develops a pattern of attention that is easier for other agents to categorize. Third, in a joint task where two agents must predict each other to paint a scene together, adding an attention schema improves performance. Finally, the performance improvements are not caused by a general increase in network complexity. Instead, improvement is specific to tasks involving judging, categorizing, or predicting the attention of other agents. These results support the hypothesis that an attention schema has computational properties beneficial to mutual interpretability and interactive behavior. We speculate that the same principles might pertain to biological attention and attention schemas in people.", "date": "2024-11-01", "categories": []}
{"id": "2508.11672", "title": "Revealing Neurocognitive and Behavioral Patterns by Unsupervised Manifold Learning from Dynamic Brain Data", "abstract": "Dynamic brain data, teeming with biological and functional insights, are becoming increasingly accessible through advanced measurements, providing a gateway to understanding the inner workings of the brain in living subjects. However, the vast size and intricate complexity of the data also pose a daunting challenge in reliably extracting meaningful information across various data sources. This paper introduces a generalizable unsupervised deep manifold learning for exploration of neurocognitive and behavioral patterns. Unlike existing methods that extract patterns directly from the input data as in the existing methods, the proposed Brain-dynamic Convolutional-Network-based Embedding (BCNE) seeks to capture the brain-state trajectories by deciphering the temporospatial correlations within the data and subsequently applying manifold learning to this correlative representation. The performance of BCNE is showcased through the analysis of several important dynamic brain datasets. The results, both visual and quantitative, reveal a diverse array of intriguing and interpretable patterns. BCNE effectively delineates scene transitions, underscores the involvement of different brain regions in memory and narrative processing, distinguishes various stages of dynamic learning processes, and identifies differences between active and passive behaviors. BCNE provides an effective tool for exploring general neuroscience inquiries or individual-specific patterns.", "date": "2025-08-07", "categories": []}
{"id": "2508.12702", "title": "A Unified Cortical Circuit Model with Divisive Normalization and Self-Excitation for Robust Representation and Memory Maintenance", "abstract": "Robust information representation and its persistent maintenance are fundamental for higher cognitive functions. Existing models employ distinct neural mechanisms to separately address noise-resistant processing or information maintenance, yet a unified framework integrating both operations remains elusive -- a critical gap in understanding cortical computation. Here, we introduce a recurrent neural circuit that combines divisive normalization with self-excitation to achieve both robust encoding and stable retention of normalized inputs. Mathematical analysis shows that, for suitable parameter regimes, the system forms a continuous attractor with two key properties: (1) input-proportional stabilization during stimulus presentation; and (2) self-sustained memory states persisting after stimulus offset. We demonstrate the model's versatility in two canonical tasks: (a) noise-robust encoding in a random-dot kinematogram (RDK) paradigm; and (b) approximate Bayesian belief updating in a probabilistic Wisconsin Card Sorting Test (pWCST). This work establishes a unified mathematical framework that bridges noise suppression, working memory, and approximate Bayesian inference within a single cortical microcircuit, offering fresh insights into the brain's canonical computation and guiding the design of biologically plausible artificial neural architectures.", "date": "2025-08-18", "categories": []}
{"id": "2508.11214", "title": "How Causal Abstraction Underpins Computational Explanation", "abstract": "Explanations of cognitive behavior often appeal to computations over representations. What does it take for a system to implement a given computation over suitable representational vehicles within that system? We argue that the language of causality -- and specifically the theory of causal abstraction -- provides a fruitful lens on this topic. Drawing on current discussions in deep learning with artificial neural networks, we illustrate how classical themes in the philosophy of computation and cognition resurface in contemporary machine learning. We offer an account of computational implementation grounded in causal abstraction, and examine the role for representation in the resulting picture. We argue that these issues are most profitably explored in connection with generalization and prediction.", "date": "2025-08-15", "categories": []}
{"id": "2508.11536", "title": "Language models align with brain regions that represent concepts across modalities", "abstract": "Cognitive science and neuroscience have long faced the challenge of disentangling representations of language from representations of conceptual meaning. As the same problem arises in today's language models (LMs), we investigate the relationship between LM--brain alignment and two neural metrics: (1) the level of brain activation during processing of sentences, targeting linguistic processing, and (2) a novel measure of meaning consistency across input modalities, which quantifies how consistently a brain region responds to the same concept across paradigms (sentence, word cloud, image) using an fMRI dataset (Pereira et al., 2018). Our experiments show that both language-only and language-vision models predict the signal better in more meaning-consistent areas of the brain, even when these areas are not strongly sensitive to language processing, suggesting that LMs might internally represent cross-modal conceptual meaning.", "date": "2025-08-15", "categories": []}
{"id": "2508.10160", "title": "Pre-trained Transformer-models using chronic invasive electrophysiology for symptom decoding without patient-individual training", "abstract": "Neural decoding of pathological and physiological states can enable patient-individualized closed-loop neuromodulation therapy. Recent advances in pre-trained large-scale foundation models offer the potential for generalized state estimation without patient-individual training. Here we present a foundation model trained on chronic longitudinal deep brain stimulation recordings spanning over 24 days. Adhering to long time-scale symptom fluctuations, we highlight the extended context window of 30 minutes. We present an optimized pre-training loss function for neural electrophysiological data that corrects for the frequency bias of common masked auto-encoder loss functions due to the 1-over-f power law. We show in a downstream task the decoding of Parkinson's disease symptoms with leave-one-subject-out cross-validation without patient-individual training.", "date": "2025-08-13", "categories": []}
{"id": "2508.10450", "title": "From Images to Perception: Emergence of Perceptual Properties by Reconstructing Images", "abstract": "A number of scientists suggested that human visual perception may emerge from image statistics, shaping efficient neural representations in early vision. In this work, a bio-inspired architecture that can accommodate several known facts in the retina-V1 cortex, the PerceptNet, has been end-to-end optimized for different tasks related to image reconstruction: autoencoding, denoising, deblurring, and sparsity regularization. Our results show that the encoder stage (V1-like layer) consistently exhibits the highest correlation with human perceptual judgments on image distortion despite not using perceptual information in the initialization or training. This alignment exhibits an optimum for moderate noise, blur and sparsity. These findings suggest that the visual system may be tuned to remove those particular levels of distortion with that level of sparsity and that biologically inspired models can learn perceptual metrics without human supervision.", "date": "2025-08-14", "categories": []}
{"id": "2407.00397", "title": "Learning Time-Varying Multi-Region Brain Communications via Scalable Markovian Gaussian Processes", "abstract": "Understanding and constructing brain communications that capture dynamic communications across multiple regions is fundamental to modern system neuroscience, yet current methods struggle to find time-varying region-level communications or scale to large neural datasets with long recording durations. We present a novel framework using Markovian Gaussian Processes to learn brain communications with time-varying temporal delays from multi-region neural recordings, named Adaptive Delay Model (ADM). Our method combines Gaussian Processes with State Space Models and employs parallel scan inference algorithms, enabling efficient scaling to large datasets while identifying concurrent communication patterns that evolve over time. This time-varying approach captures how brain region interactions shift dynamically during cognitive processes. Validated on synthetic and multi-region neural recordings datasets, our approach discovers both the directionality and temporal dynamics of neural communication. This work advances our understanding of distributed neural computation and provides a scalable tool for analyzing dynamic brain networks.", "date": "2024-06-29", "categories": []}
{"id": "2503.16567", "title": "Exploring Deep Learning Models for EEG Neural Decoding", "abstract": "Neural decoding is an important method in cognitive neuroscience that aims to decode brain representations from recorded neural activity using a multivariate machine learning model. The THINGS initiative provides a large EEG dataset of 46 subjects watching rapidly shown images. Here, we test the feasibility of using this method for decoding high-level object features using recent deep learning models. We create a derivative dataset from this of living vs non-living entities test 15 different deep learning models with 5 different architectures and compare to a SOTA linear model. We show that the linear model is not able to solve the decoding task, while almost all the deep learning models are successful, suggesting that in some cases non-linear models are needed to decode neural representations. We also run a comparative study of the models' performance on individual object categories, and suggest how artificial neural networks can be used to study brain activity.", "date": "2025-03-20", "categories": []}
{"id": "2508.06501", "title": "Computing with Canonical Microcircuits", "abstract": "The human brain represents the only known example of general intelligence that naturally aligns with human values. On a mere 20-watt power budget, the brain achieves robust learning and adaptive decision-making in ways that continue to elude advanced AI systems. Inspired by the brain, we present a computational architecture based on canonical microcircuits (CMCs) - stereotyped patterns of neurons found ubiquitously throughout the cortex. We implement these circuits as neural ODEs comprising spiny stellate, inhibitory, and pyramidal neurons, forming an 8-dimensional dynamical system with biologically plausible recurrent connections. Our experiments show that even a single CMC node achieves 97.8 percent accuracy on MNIST, while hierarchical configurations - with learnable inter-regional connectivity and recurrent connections - yield improved performance on more complex image benchmarks. Notably, our approach achieves competitive results using substantially fewer parameters than conventional deep learning models. Phase space analysis revealed distinct dynamical trajectories for different input classes, highlighting interpretable, emergent behaviors observed in biological systems. These findings suggest that neuromorphic computing approaches can improve both efficiency and interpretability in artificial neural networks, offering new directions for parameter-efficient architectures grounded in the computational principles of the human brain.", "date": "2025-07-25", "categories": []}
{"id": "2508.07283", "title": "Fine-Tuning Large Language Models Using EEG Microstate Features for Mental Workload Assessment", "abstract": "This study explores the intersection of electroencephalography (EEG) microstates and Large Language Models (LLMs) to enhance the assessment of cognitive load states. By utilizing EEG microstate features, the research aims to fine-tune LLMs for improved predictions of distinct cognitive states, specifically 'Rest' and 'Load'. The experimental design is delineated in four comprehensive stages: dataset collection and preprocessing, microstate segmentation and EEG backfitting, feature extraction paired with prompt engineering, and meticulous LLM model selection and refinement. Employing a supervised learning paradigm, the LLM is trained to identify cognitive load states based on EEG microstate features integrated into prompts, producing accurate discrimination of cognitive load. A curated dataset, linking EEG features to specified cognitive load conditions, underpins the experimental framework. The results indicate a significant improvement in model performance following the proposed fine-tuning, showcasing the potential of EEG-informed LLMs in cognitive neuroscience and cognitive AI applications. This approach not only contributes to the understanding of brain dynamics but also paves the way for advancements in machine learning techniques applicable to cognitive load and cognitive AI research.", "date": "2025-08-10", "categories": []}
{"id": "2508.05341", "title": "Harmonic fractal transformation for modeling complex neuronal effects: from bursting and noise shaping to waveform sensitivity and noise-induced subthreshold spiking", "abstract": "We propose the first fractal frequency mapping, which in a simple form enables to replicate complex neuronal effects. Unlike the conventional filters, which suppress or amplify the input spectral components according to the filter weights, the transformation excites novel components by a fractal recomposition of the input spectra resulting in a formation of spikes at resonant frequencies that are optimal for sampling. This enables high sensitivity detection, robustness to noise and noise-induced signal amplification. The proposed model illustrates that a neuronal functionality can be viewed as a linear summation of spectrum over nonlinearly transformed frequency domain.", "date": "2025-08-07", "categories": []}
{"id": "2410.09206", "title": "pyhgf: A neural network library for predictive coding", "abstract": "Bayesian models of cognition have gained considerable traction in computational neuroscience and psychiatry. Their scopes are now expected to expand rapidly to artificial intelligence, providing general inference frameworks to support embodied, adaptable, and energy-efficient autonomous agents. A central theory in this domain is predictive coding, which posits that learning and behaviour are driven by hierarchical probabilistic inferences about the causes of sensory inputs. Biological realism constrains these networks to rely on simple local computations in the form of precision-weighted predictions and prediction errors. This can make this framework highly efficient, but its implementation comes with unique challenges on the software development side. Embedding such models in standard neural network libraries often becomes limiting, as these libraries' compilation and differentiation backends can force a conceptual separation between optimization algorithms and the systems being optimized. This critically departs from other biological principles such as self-monitoring, self-organisation, cellular growth and functional plasticity. In this paper, we introduce \\texttt{pyhgf}: a Python package backed by JAX and Rust for creating, manipulating and sampling dynamic networks for predictive coding. We improve over other frameworks by enclosing the network components as transparent, modular and malleable variables in the message-passing steps. The resulting graphs can implement arbitrary computational complexities as beliefs propagation. But the transparency of core variables can also translate into inference processes that leverage self-organisation principles, and express structure learning, meta-learning or causal discovery as the consequence of network structural adaptation to surprising inputs. The code, tutorials and documentation are hosted at: https://github.com/ilabcode/pyhgf.", "date": "2024-10-11", "categories": []}
{"id": "2410.06232", "title": "Range, not Independence, Drives Modularity in Biologically Inspired Representations", "abstract": "Why do biological and artificial neurons sometimes modularise, each encoding a single meaningful variable, and sometimes entangle their representation of many variables? In this work, we develop a theory of when biologically inspired networks -- those that are nonnegative and energy efficient -- modularise their representation of source variables (sources). We derive necessary and sufficient conditions on a sample of sources that determine whether the neurons in an optimal biologically-inspired linear autoencoder modularise. Our theory applies to any dataset, extending far beyond the case of statistical independence studied in previous work. Rather we show that sources modularise if their support is ``sufficiently spread''. From this theory, we extract and validate predictions in a variety of empirical studies on how data distribution affects modularisation in nonlinear feedforward and recurrent neural networks trained on supervised and unsupervised tasks. Furthermore, we apply these ideas to neuroscience data, showing that range independence can be used to understand the mixing or modularising of spatial and reward information in entorhinal recordings in seemingly conflicting experiments. Further, we use these results to suggest alternate origins of mixed-selectivity, beyond the predominant theory of flexible nonlinear classification. In sum, our theory prescribes precise conditions on when neural activities modularise, providing tools for inducing and elucidating modular representations in brains and machines.", "date": "2024-10-08", "categories": []}
{"id": "2508.00850", "title": "Gearshift Fellowship: A Next-Generation Neurocomputational Game Platform to Model and Train Human-AI Adaptability", "abstract": "How do we learn when to persist, when to let go, and when to shift gears? Gearshift Fellowship (GF) is the prototype of a new Supertask paradigm designed to model how humans and artificial agents adapt to shifting environment demands. Grounded in cognitive neuroscience, computational psychiatry, economics, and artificial intelligence, Supertasks combine computational neurocognitive modeling with serious gaming. This creates a dynamic, multi-mission environment engineered to assess mechanisms of adaptive behavior across cognitive and social contexts. Computational parameters explain behavior and probe mechanisms by controlling the game environment. Unlike traditional tasks, GF enables neurocognitive modeling of individual differences across perceptual decisions, learning, and meta-cognitive levels. This positions GF as a flexible testbed for understanding how cognitive-affective control processes, learning styles, strategy use, and motivational shifts adapt across contexts and over time. It serves as an experimental platform for scientists, a phenotype-to-mechanism intervention for clinicians, and a training tool for players aiming to strengthen self-regulated learning, mood, and stress resilience. Online study (n = 60, ongoing) results show that GF recovers effects from traditional neuropsychological tasks (construct validity), uncovers novel patterns in how learning differs across contexts and how clinical features map onto distinct adaptations. These findings pave the way for developing in-game interventions that foster self-efficacy and agency to cope with real-world stress and uncertainty. GF builds a new adaptive ecosystem designed to accelerate science, transform clinical care, and foster individual growth. It offers a mirror and training ground where humans and machines co-develop together deeper flexibility and awareness.", "date": "2025-07-11", "categories": []}
{"id": "2508.00531", "title": "The Repeated-Stimulus Confound in Electroencephalography", "abstract": "In neural-decoding studies, recordings of participants' responses to stimuli are used to train models. In recent years, there has been an explosion of publications detailing applications of innovations from deep-learning research to neural-decoding studies. The data-hungry models used in these experiments have resulted in a demand for increasingly large datasets. Consequently, in some studies, the same stimuli are presented multiple times to each participant to increase the number of trials available for use in model training. However, when a decoding model is trained and subsequently evaluated on responses to the same stimuli, stimulus identity becomes a confounder for accuracy. We term this the repeated-stimulus confound. We identify a susceptible dataset, and 16 publications which report model performance based on evaluation procedures affected by the confound. We conducted experiments using models from the affected studies to investigate the likely extent to which results in the literature have been misreported. Our findings suggest that the decoding accuracies of these models were overestimated by between 4.46-7.42%. Our analysis also indicates that per 1% increase in accuracy under the confound, the magnitude of the overestimation increases by 0.26%. The confound not only results in optimistic estimates of decoding performance, but undermines the validity of several claims made within the affected publications. We conducted further experiments to investigate the implications of the confound in alternative contexts. We found that the same methodology used within the affected studies could also be used to justify an array of pseudoscientific claims, such as the existence of extrasensory perception.", "date": "2025-08-01", "categories": []}
{"id": "2507.22229", "title": "TRIBE: TRImodal Brain Encoder for whole-brain fMRI response prediction", "abstract": "Historically, neuroscience has progressed by fragmenting into specialized domains, each focusing on isolated modalities, tasks, or brain regions. While fruitful, this approach hinders the development of a unified model of cognition. Here, we introduce TRIBE, the first deep neural network trained to predict brain responses to stimuli across multiple modalities, cortical areas and individuals. By combining the pretrained representations of text, audio and video foundational models and handling their time-evolving nature with a transformer, our model can precisely model the spatial and temporal fMRI responses to videos, achieving the first place in the Algonauts 2025 brain encoding competition with a significant margin over competitors. Ablations show that while unimodal models can reliably predict their corresponding cortical networks (e.g. visual or auditory networks), they are systematically outperformed by our multimodal model in high-level associative cortices. Currently applied to perception and comprehension, our approach paves the way towards building an integrative model of representations in the human brain. Our code is available at https://github.com/facebookresearch/algonauts-2025.", "date": "2025-07-29", "categories": []}
{"id": "2507.19956", "title": "Predicting Brain Responses To Natural Movies With Multimodal LLMs", "abstract": "We present MedARC's team solution to the Algonauts 2025 challenge. Our pipeline leveraged rich multimodal representations from various state-of-the-art pretrained models across video (V-JEPA2), speech (Whisper), text (Llama 3.2), vision-text (InternVL3), and vision-text-audio (Qwen2.5-Omni). These features extracted from the models were linearly projected to a latent space, temporally aligned to the fMRI time series, and finally mapped to cortical parcels through a lightweight encoder comprising a shared group head plus subject-specific residual heads. We trained hundreds of model variants across hyperparameter settings, validated them on held-out movies and assembled ensembles targeted to each parcel in each subject. Our final submission achieved a mean Pearson's correlation of 0.2085 on the test split of withheld out-of-distribution movies, placing our team in fourth place for the competition. We further discuss a last-minute optimization that would have raised us to second place. Our results highlight how combining features from models trained in different modalities, using a simple architecture consisting of shared-subject and single-subject components, and conducting comprehensive model selection and ensembling improves generalization of encoding models to novel movie stimuli. All code is available on GitHub.", "date": "2025-07-26", "categories": []}
{"id": "2507.17958", "title": "VIBE: Video-Input Brain Encoder for fMRI Response Modeling", "abstract": "We present VIBE, a two-stage Transformer that fuses multi-modal video, audio, and text features to predict fMRI activity. Representations from open-source models (Qwen2.5, BEATs, Whisper, SlowFast, V-JEPA) are merged by a modality-fusion transformer and temporally decoded by a prediction transformer with rotary embeddings. Trained on 65 hours of movie data from the CNeuroMod dataset and ensembled across 20 seeds, VIBE attains mean parcel-wise Pearson correlations of 0.3225 on in-distribution Friends S07 and 0.2125 on six out-of-distribution films. An earlier iteration of the same architecture obtained 0.3198 and 0.2096, respectively, winning Phase-1 and placing second overall in the Algonauts 2025 Challenge.", "date": "2025-07-23", "categories": []}
{"id": "2507.18104", "title": "A Multimodal Seq2Seq Transformer for Predicting Brain Responses to Naturalistic Stimuli", "abstract": "The Algonauts 2025 Challenge called on the community to develop encoding models that predict whole-brain fMRI responses to naturalistic multimodal movies. In this submission, we propose a sequence-to-sequence Transformer that autoregressively predicts fMRI activity from visual, auditory, and language inputs. Stimulus features were extracted using pretrained models including VideoMAE, HuBERT, Qwen, and BridgeTower. The decoder integrates information from prior brain states and current stimuli via dual cross-attention mechanisms that attend to both perceptual information extracted from the stimulus as well as narrative information provided by high-level summaries of the content. One core innovation of our approach is the use of sequences of multimodal context to predict sequences of brain activity, enabling the model to capture long-range temporal structure in both stimuli and neural responses. Another is the combination of a shared encoder with partial subject-specific decoder, which leverages common representational structure across subjects while accounting for individual variability. Our model achieves strong performance on both in-distribution and out-of-distribution data, demonstrating the effectiveness of temporally-aware, multimodal sequence modeling for brain activity prediction. The code is available at https://github.com/Angelneer926/Algonauts_challenge.", "date": "2025-07-24", "categories": []}
{"id": "2507.19052", "title": "Probing Multimodal Fusion in the Brain: The Dominance of Audiovisual Streams in Naturalistic Encoding", "abstract": "Predicting brain activity in response to naturalistic, multimodal stimuli is a key challenge in computational neuroscience. While encoding models are becoming more powerful, their ability to generalize to truly novel contexts remains a critical, often untested, question. In this work, we developed brain encoding models using state-of-the-art visual (X-CLIP) and auditory (Whisper) feature extractors and rigorously evaluated them on both in-distribution (ID) and diverse out-of-distribution (OOD) data. Our results reveal a fundamental trade-off between model complexity and generalization: a higher-capacity attention-based model excelled on ID data, but a simpler linear model was more robust, outperforming a competitive baseline by 18\\% on the OOD set. Intriguingly, we found that linguistic features did not improve predictive accuracy, suggesting that for familiar languages, neural encoding may be dominated by the continuous visual and auditory streams over redundant textual information. Spatially, our approach showed marked performance gains in the auditory cortex, underscoring the benefit of high-fidelity speech representations. Collectively, our findings demonstrate that rigorous OOD testing is essential for building robust neuro-AI models and provides nuanced insights into how model architecture, stimulus characteristics, and sensory hierarchies shape the neural encoding of our rich, multimodal world.", "date": "2025-07-25", "categories": []}
{"id": "2507.15274", "title": "Temporal Basis Function Models for Closed-Loop Neural Stimulation", "abstract": "Closed-loop neural stimulation provides novel therapies for neurological diseases such as Parkinson's disease (PD), but it is not yet clear whether artificial intelligence (AI) techniques can tailor closed-loop stimulation to individual patients or identify new therapies. Progress requires us to address a number of translational issues, including sample efficiency, training time, and minimizing loop latency such that stimulation may be shaped in response to changing brain activity. We propose temporal basis function models (TBFMs) to address these difficulties, and explore this approach in the context of excitatory optogenetic stimulation. We demonstrate the ability of TBF models to provide a single-trial, spatiotemporal forward prediction of the effect of optogenetic stimulation on local field potentials (LFPs) measured in two non-human primates. We further use simulations to demonstrate the use of TBF models for closed-loop stimulation, driving neural activity towards target patterns. The simplicity of TBF models allow them to be sample efficient, rapid to train (2-4min), and low latency (0.2ms) on desktop CPUs. We demonstrate the model on 40 sessions of previously published excitatory optogenetic stimulation data. For each session, the model required 15-20min of data collection to successfully model the remainder of the session. It achieved a prediction accuracy comparable to a baseline nonlinear dynamical systems model that requires hours to train, and superior accuracy to a linear state-space model. In our simulations, it also successfully allowed a closed-loop stimulator to control a neural circuit. Our approach begins to bridge the translational gap between complex AI-based approaches to modeling dynamical systems and the vision of using such forward prediction models to develop novel, clinically useful closed-loop stimulation protocols.", "date": "2025-07-21", "categories": []}
{"id": "2507.11788", "title": "Simulated Language Acquisition in a Biologically Realistic Model of the Brain", "abstract": "Despite tremendous progress in neuroscience, we do not have a compelling narrative for the precise way whereby the spiking of neurons in our brain results in high-level cognitive phenomena such as planning and language. We introduce a simple mathematical formulation of six basic and broadly accepted principles of neuroscience: excitatory neurons, brain areas, random synapses, Hebbian plasticity, local inhibition, and inter-area inhibition. We implement a simulated neuromorphic system based on this formalism, which is capable of basic language acquisition: Starting from a tabula rasa, the system learns, in any language, the semantics of words, their syntactic role (verb versus noun), and the word order of the language, including the ability to generate novel sentences, through the exposure to a modest number of grounded sentences in the same language. We discuss several possible extensions and implications of this result.", "date": "2025-07-15", "categories": []}
{"id": "2507.12009", "title": "Deep Neural Encoder-Decoder Model to Relate fMRI Brain Activity with Naturalistic Stimuli", "abstract": "We propose an end-to-end deep neural encoder-decoder model to encode and decode brain activity in response to naturalistic stimuli using functional magnetic resonance imaging (fMRI) data. Leveraging temporally correlated input from consecutive film frames, we employ temporal convolutional layers in our architecture, which effectively allows to bridge the temporal resolution gap between natural movie stimuli and fMRI acquisitions. Our model predicts activity of voxels in and around the visual cortex and performs reconstruction of corresponding visual inputs from neural activity. Finally, we investigate brain regions contributing to visual decoding through saliency maps. We find that the most contributing regions are the middle occipital area, the fusiform area, and the calcarine, respectively employed in shape perception, complex recognition (in particular face perception), and basic visual features such as edges and contrasts. These functions being strongly solicited are in line with the decoder's capability to reconstruct edges, faces, and contrasts. All in all, this suggests the possibility to probe our understanding of visual processing in films using as a proxy the behaviour of deep learning models such as the one proposed in this paper.", "date": "2025-07-16", "categories": []}
{"id": "2507.12417", "title": "Spontaneous Spatial Cognition Emerges during Egocentric Video Viewing through Non-invasive BCI", "abstract": "Humans possess a remarkable capacity for spatial cognition, allowing for self-localization even in novel or unfamiliar environments. While hippocampal neurons encoding position and orientation are well documented, the large-scale neural dynamics supporting spatial representation, particularly during naturalistic, passive experience, remain poorly understood. Here, we demonstrate for the first time that non-invasive brain-computer interfaces (BCIs) based on electroencephalography (EEG) can decode spontaneous, fine-grained egocentric 6D pose, comprising three-dimensional position and orientation, during passive viewing of egocentric video. Despite EEG's limited spatial resolution and high signal noise, we find that spatially coherent visual input (i.e., continuous and structured motion) reliably evokes decodable spatial representations, aligning with participants' subjective sense of spatial engagement. Decoding performance further improves when visual input is presented at a frame rate of 100 ms per image, suggesting alignment with intrinsic neural temporal dynamics. Using gradient-based backpropagation through a neural decoding model, we identify distinct EEG channels contributing to position -- and orientation specific -- components, revealing a distributed yet complementary neural encoding scheme. These findings indicate that the brain's spatial systems operate spontaneously and continuously, even under passive conditions, challenging traditional distinctions between active and passive spatial cognition. Our results offer a non-invasive window into the automatic construction of egocentric spatial maps and advance our understanding of how the human mind transforms everyday sensory experience into structured internal representations.", "date": "2025-07-16", "categories": []}
{"id": "2407.09488", "title": "Bayesian Theory of Consciousness as Exchangeable Emotion-Cognition Inference", "abstract": "This paper proposes a unified framework in which consciousness emerges as a cycle-consistent, affectively anchored inference process, recursively structured by the interaction of emotion and cognition. Drawing from information theory, optimal transport, and the Bayesian brain hypothesis, we formalize emotion as a low-dimensional structural prior and cognition as a specificity-instantiating update. This emotion-cognition cycle minimizes joint uncertainty by aligning emotionally weighted priors with context-sensitive cognitive appraisals. Subjective experience thus arises as the informational footprint of temporally extended, affect-modulated simulation. We introduce the Exchangeable Integration Theory of Consciousness (EITC), modeling conscious episodes as conditionally exchangeable samples drawn from a latent affective self-model. This latent variable supports integration, via a unified cause-effect structure with nonzero irreducibility, and differentiation, by preserving contextual specificity across episodes. We connect this architecture to the Bayesian theory of consciousness through Rao-Blackwellized inference, which stabilizes inference by marginalizing latent self-structure while enabling adaptive updates. This mechanism ensures coherence, prevents inference collapse, and supports goal-directed simulation. The formal framework builds on De Finetti's exchangeability theorem, integrated information theory, and KL-regularized optimal transport. Overall, consciousness is reframed as a recursive inference process, shaped by emotion, refined by cognition, stabilized through exchangeability, and unified through a latent self-model that integrates experience across time.", "date": "2024-05-17", "categories": []}
{"id": "2507.03222", "title": "The role of gain neuromodulation in layer-5 pyramidal neurons", "abstract": "Biological and artificial learning systems alike confront the plasticity-stability dilemma. In the brain, neuromodulators such as acetylcholine and noradrenaline relieve this tension by tuning neuronal gain and inhibitory gating, balancing segregation and integration of circuits. Fed by dense cholinergic and noradrenergic projections from the ascending arousal system, layer-5 pyramidal neurons in the cerebral cortex offer a relevant substrate for understanding these dynamics. When distal dendritic signals coincide with back-propagating action potentials, calcium plateaus turn a single somatic spike into a high-gain burst, and interneuron inhibition sculpts the output. These properties make layer-5 cells gain-tunable amplifiers that translate neuromodulatory cues into flexible cortical activity. To capture this mechanism we developed a two-compartment Izhikevich model for pyramidal neurons and single-compartment somatostatin (SOM) and parvalbumin (PV) interneurons, linked by Gaussian connectivity and spike-timing-dependent plasticity (STDP). The soma and apical dendrite are so coupled that somatic spikes back-propagate, while dendritic plateaus can switch the soma from regular firing to bursting by shifting reset and adaptation variables. We show that stronger dendritic drive or tighter coupling raise gain by increasing the likelihood of calcium-triggered somatic bursts. In contrast, dendritic-targeted inhibition suppresses gain, while somatic-targeted inhibition raises the firing threshold of neighboring neurons, thus gating neurons output. Notably, bursting accelerates STDP, supporting rapid synaptic reconfiguration and flexibility. This suggests that brief gain pulses driven by neuromodulators could serve as an adaptive two-timescale optimization mechanism, effectively modulating the synaptic weight updates.", "date": "2025-07-03", "categories": []}
{"id": "2507.08402", "title": "SPINT: Spatial Permutation-Invariant Neural Transformer for Consistent Intracortical Motor Decoding", "abstract": "Intracortical Brain-Computer Interfaces (iBCI) aim to decode behavior from neural population activity, enabling individuals with motor impairments to regain motor functions and communication abilities. A key challenge in long-term iBCI is the nonstationarity of neural recordings, where the composition and tuning profiles of the recorded populations are unstable across recording sessions. Existing methods attempt to address this issue by explicit alignment techniques; however, they rely on fixed neural identities and require test-time labels or parameter updates, limiting their generalization across sessions and imposing additional computational burden during deployment. In this work, we introduce SPINT - a Spatial Permutation-Invariant Neural Transformer framework for behavioral decoding that operates directly on unordered sets of neural units. Central to our approach is a novel context-dependent positional embedding scheme that dynamically infers unit-specific identities, enabling flexible generalization across recording sessions. SPINT supports inference on variable-size populations and allows few-shot, gradient-free adaptation using a small amount of unlabeled data from the test session. To further promote model robustness to population variability, we introduce dynamic channel dropout, a regularization method for iBCI that simulates shifts in population composition during training. We evaluate SPINT on three multi-session datasets from the FALCON Benchmark, covering continuous motor decoding tasks in human and non-human primates. SPINT demonstrates robust cross-session generalization, outperforming existing zero-shot and few-shot unsupervised baselines while eliminating the need for test-time alignment and fine-tuning. Our work contributes an initial step toward a robust and scalable neural decoding framework for long-term iBCI applications.", "date": "2025-07-11", "categories": []}
{"id": "2408.05798", "title": "Time Makes Space: Emergence of Place Fields in Networks Encoding Temporally Continuous Sensory Experiences", "abstract": "The vertebrate hippocampus is believed to use recurrent connectivity in area CA3 to support episodic memory recall from partial cues. This brain area also contains place cells, whose location-selective firing fields implement maps supporting spatial memory. Here we show that place cells emerge in networks trained to remember temporally continuous sensory episodes. We model CA3 as a recurrent autoencoder that recalls and reconstructs sensory experiences from noisy and partially occluded observations by agents traversing simulated rooms. The agents move in realistic trajectories modeled from rodents and environments are modeled as high-dimensional sensory experience maps. Training our autoencoder to pattern-complete and reconstruct experiences with a constraint on total activity causes spatially localized firing fields, i.e., place cells, to emerge in the encoding layer. The emergent place fields reproduce key aspects of hippocampal phenomenology: a) remapping (maintenance of and reversion to distinct learned maps in different environments), implemented via repositioning of experience manifolds in the network's hidden layer, b) orthogonality of spatial representations in different arenas, c) robust place field emergence in differently shaped rooms, with single units showing multiple place fields in large or complex spaces, and d) slow representational drift of place fields. We argue that these results arise because continuous traversal of space makes sensory experience temporally continuous. We make testable predictions: a) rapidly changing sensory context will disrupt place fields, b) place fields will form even if recurrent connections are blocked, but reversion to previously learned representations upon remapping will be abolished, c) the dimension of temporally smooth experience sets the dimensionality of place fields, including during virtual navigation of abstract spaces.", "date": "2024-08-11", "categories": []}
{"id": "2507.07157", "title": "Interpretable EEG-to-Image Generation with Semantic Prompts", "abstract": "Decoding visual experience from brain signals offers exciting possibilities for neuroscience and interpretable AI. While EEG is accessible and temporally precise, its limitations in spatial detail hinder image reconstruction. Our model bypasses direct EEG-to-image generation by aligning EEG signals with multilevel semantic captions -- ranging from object-level to abstract themes -- generated by a large language model. A transformer-based EEG encoder maps brain activity to these captions through contrastive learning. During inference, caption embeddings retrieved via projection heads condition a pretrained latent diffusion model for image generation. This text-mediated framework yields state-of-the-art visual decoding on the EEGCVPR dataset, with interpretable alignment to known neurocognitive pathways. Dominant EEG-caption associations reflected the importance of different semantic levels extracted from perceived images. Saliency maps and t-SNE projections reveal semantic topography across the scalp. Our model demonstrates how structured semantic mediation enables cognitively aligned visual decoding from EEG.", "date": "2025-07-09", "categories": []}
{"id": "2507.05888", "title": "Hierarchy or Heterarchy? A Theory of Long-Range Connections for the Sensorimotor Brain", "abstract": "In the traditional understanding of the neocortex, sensory information flows up a hierarchy of regions, with each level processing increasingly complex features. Information also flows down the hierarchy via a different set of connections. Although the hierarchical model has significant support, many anatomical connections do not conform to the standard hierarchical interpretation. In addition, hierarchically arranged regions sometimes respond in parallel, not sequentially as would occur in a hierarchy. This and other evidence suggests that two regions can act in parallel and hierarchically at the same time. Given this flexibility, the word \"heterarchy\" might be a more suitable term to describe neocortical organization. This paper proposes a new interpretation of how sensory and motor information is processed in the neocortex. The key to our proposal is what we call the \"Thousand Brains Theory\", which posits that every cortical column is a sensorimotor learning system. Columns learn by integrating sensory input over multiple movements of a sensor. In this view, even primary and secondary regions, such as V1 and V2, can learn and recognize complete 3D objects. This suggests that the hierarchical connections between regions are used to learn the compositional structure of parent objects composed of smaller child objects. We explain the theory by examining the different types of long-range connections between cortical regions and between the neocortex and thalamus. We describe these connections, and then suggest the specific roles they play in the context of a heterarchy of sensorimotor regions. We also suggest that the thalamus plays an essential role in transforming the pose between objects and sensors. The novel perspective we argue for here has broad implications for both neuroscience and artificial intelligence.", "date": "2025-07-08", "categories": []}
{"id": "2507.02264", "title": "NLP4Neuro: Sequence-to-sequence learning for neural population decoding", "abstract": "Delineating how animal behavior arises from neural activity is a foundational goal of neuroscience. However, as the computations underlying behavior unfold in networks of thousands of individual neurons across the entire brain, this presents challenges for investigating neural roles and computational mechanisms in large, densely wired mammalian brains during behavior. Transformers, the backbones of modern large language models (LLMs), have become powerful tools for neural decoding from smaller neural populations. These modern LLMs have benefited from extensive pre-training, and their sequence-to-sequence learning has been shown to generalize to novel tasks and data modalities, which may also confer advantages for neural decoding from larger, brain-wide activity recordings. Here, we present a systematic evaluation of off-the-shelf LLMs to decode behavior from brain-wide populations, termed NLP4Neuro, which we used to test LLMs on simultaneous calcium imaging and behavior recordings in larval zebrafish exposed to visual motion stimuli. Through NLP4Neuro, we found that LLMs become better at neural decoding when they use pre-trained weights learned from textual natural language data. Moreover, we found that a recent mixture-of-experts LLM, DeepSeek Coder-7b, significantly improved behavioral decoding accuracy, predicted tail movements over long timescales, and provided anatomically consistent highly interpretable readouts of neuron salience. NLP4Neuro demonstrates that LLMs are highly capable of informing brain-wide neural circuit dissection.", "date": "2025-07-03", "categories": []}
{"id": "2507.02311", "title": "Perception Activator: An intuitive and portable framework for brain cognitive exploration", "abstract": "Recent advances in brain-vision decoding have driven significant progress, reconstructing with high fidelity perceived visual stimuli from neural activity, e.g., functional magnetic resonance imaging (fMRI), in the human visual cortex. Most existing methods decode the brain signal using a two-level strategy, i.e., pixel-level and semantic-level. However, these methods rely heavily on low-level pixel alignment yet lack sufficient and fine-grained semantic alignment, resulting in obvious reconstruction distortions of multiple semantic objects. To better understand the brain's visual perception patterns and how current decoding models process semantic objects, we have developed an experimental framework that uses fMRI representations as intervention conditions. By injecting these representations into multi-scale image features via cross-attention, we compare both downstream performance and intermediate feature changes on object detection and instance segmentation tasks with and without fMRI information. Our results demonstrate that incorporating fMRI signals enhances the accuracy of downstream detection and segmentation, confirming that fMRI contains rich multi-object semantic cues and coarse spatial localization information-elements that current models have yet to fully exploit or integrate.", "date": "2025-07-03", "categories": []}
{"id": "2411.17485", "title": "Storing overlapping associative memories on latent manifolds in low-rank spiking networks", "abstract": "Associative memory architectures such as the Hopfield network have long been important conceptual and theoretical models for neuroscience and artificial intelligence. However, translating these abstract models into spiking neural networks has been surprisingly difficult. Indeed, much previous work has been restricted to storing a small number of primarily non-overlapping memories in large networks, thereby limiting their scalability. Here, we revisit the associative memory problem in light of recent advances in understanding spike-based computation. Using a recently-established geometric framework, we show that the spiking activity for a large class of all-inhibitory networks is situated on a low-dimensional, convex, and piecewise-linear manifold, with dynamics that move along the manifold. We then map the associative memory problem onto these dynamics, and demonstrate how the vertices of a hypercubic manifold can be used to store stable, overlapping activity patterns with a direct correspondence to the original Hopfield model. We propose several learning rules, and demonstrate a linear scaling of the storage capacity with the number of neurons, as well as robust pattern completion abilities. Overall, this work serves as a case study to demonstrate the effectiveness of using a geometrical perspective to design dynamics on neural manifolds, with implications for neuroscience and machine learning.", "date": "2024-11-26", "categories": []}
{"id": "2507.00598", "title": "High-resolution spatial memory requires grid-cell-like neural codes", "abstract": "Continuous attractor networks (CANs) are widely used to model how the brain temporarily retains continuous behavioural variables via persistent recurrent activity, such as an animal's position in an environment. However, this memory mechanism is very sensitive to even small imperfections, such as noise or heterogeneity, which are both common in biological systems. Previous work has shown that discretising the continuum into a finite set of discrete attractor states provides robustness to these imperfections, but necessarily reduces the resolution of the represented variable, creating a dilemma between stability and resolution. We show that this stability-resolution dilemma is most severe for CANs using unimodal bump-like codes, as in traditional models. To overcome this, we investigate sparse binary distributed codes based on random feature embeddings, in which neurons have spatially-periodic receptive fields. We demonstrate theoretically and with simulations that such grid-cell-like codes enable CANs to achieve both high stability and high resolution simultaneously. The model extends to embedding arbitrary nonlinear manifolds into a CAN, such as spheres or tori, and generalises linear path integration to integration along freely-programmable on-manifold vector fields. Together, this work provides a theory of how the brain could robustly represent continuous variables with high resolution and perform flexible computations over task-relevant manifolds.", "date": "2025-07-01", "categories": []}
{"id": "2506.22952", "title": "Hierarchical Characterization of Brain Dynamics via State Space-based Vector Quantization", "abstract": "Understanding brain dynamics through functional Magnetic Resonance Imaging (fMRI) remains a fundamental challenge in neuroscience, particularly in capturing how the brain transitions between various functional states. Recently, metastability, which refers to temporarily stable brain states, has offered a promising paradigm to quantify complex brain signals into interpretable, discretized representations. In particular, compared to cluster-based machine learning approaches, tokenization approaches leveraging vector quantization have shown promise in representation learning with powerful reconstruction and predictive capabilities. However, most existing methods ignore brain transition dependencies and lack a quantification of brain dynamics into representative and stable embeddings. In this study, we propose a Hierarchical State space-based Tokenization network, termed HST, which quantizes brain states and transitions in a hierarchical structure based on a state space-based model. We introduce a refined clustered Vector-Quantization Variational AutoEncoder (VQ-VAE) that incorporates quantization error feedback and clustering to improve quantization performance while facilitating metastability with representative and stable token representations. We validate our HST on two public fMRI datasets, demonstrating its effectiveness in quantifying the hierarchical dynamics of the brain and its potential in disease diagnosis and reconstruction performance. Our method offers a promising framework for the characterization of brain dynamics, facilitating the analysis of metastability.", "date": "2025-06-28", "categories": []}
{"id": "2506.23075", "title": "CSBrain: A Cross-scale Spatiotemporal Brain Foundation Model for EEG Decoding", "abstract": "Understanding and decoding brain activity from electroencephalography (EEG) signals is a fundamental challenge in neuroscience and AI, with applications in cognition, emotion recognition, diagnosis, and brain-computer interfaces. While recent EEG foundation models advance generalized decoding via unified architectures and large-scale pretraining, they adopt a scale-agnostic dense modeling paradigm inherited from NLP and vision. This design neglects a core property of neural activity: cross-scale spatiotemporal structure. EEG task patterns span a wide range of temporal and spatial scales, from short bursts to slow rhythms, and from localized cortical responses to distributed interactions. Ignoring this diversity leads to suboptimal representations and weak generalization. We propose CSBrain, a Cross-scale Spatiotemporal Brain foundation model for generalized EEG decoding. CSBrain introduces: (i) Cross-scale Spatiotemporal Tokenization (CST), which aggregates multi-scale features from localized temporal windows and anatomical brain regions into compact scale-aware tokens; and (ii) Structured Sparse Attention (SSA), which captures cross-window and cross-region dependencies, enhancing scale diversity while removing spurious correlations. CST and SSA are alternately stacked to progressively integrate multi-scale dependencies. Experiments on 11 EEG tasks across 16 datasets show that CSBrain consistently outperforms task-specific and foundation model baselines. These results establish cross-scale modeling as a key inductive bias and position CSBrain as a robust backbone for future brain-AI research.", "date": "2025-06-29", "categories": []}
{"id": "2409.04428", "title": "Hybrid Spiking Neural Networks for Low-Power Intra-Cortical Brain-Machine Interfaces", "abstract": "Intra-cortical brain-machine interfaces (iBMIs) have the potential to dramatically improve the lives of people with paraplegia by restoring their ability to perform daily activities. However, current iBMIs suffer from scalability and mobility limitations due to bulky hardware and wiring. Wireless iBMIs offer a solution but are constrained by a limited data rate. To overcome this challenge, we are investigating hybrid spiking neural networks for embedded neural decoding in wireless iBMIs. The networks consist of a temporal convolution-based compression followed by recurrent processing and a final interpolation back to the original sequence length. As recurrent units, we explore gated recurrent units (GRUs), leaky integrate-and-fire (LIF) neurons, and a combination of both - spiking GRUs (sGRUs) and analyze their differences in terms of accuracy, footprint, and activation sparsity. To that end, we train decoders on the \"Nonhuman Primate Reaching with Multichannel Sensorimotor Cortex Electrophysiology\" dataset and evaluate it using the NeuroBench framework, targeting both tracks of the IEEE BioCAS Grand Challenge on Neural Decoding. Our approach achieves high accuracy in predicting velocities of primate reaching movements from multichannel primary motor cortex recordings while maintaining a low number of synaptic operations, surpassing the current baseline models in the NeuroBench framework. This work highlights the potential of hybrid neural networks to facilitate wireless iBMIs with high decoding precision and a substantial increase in the number of monitored neurons, paving the way toward more advanced neuroprosthetic technologies.", "date": "2024-09-06", "categories": []}
{"id": "2410.05266", "title": "Brain Mapping with Dense Features: Grounding Cortical Semantic Selectivity in Natural Images With Vision Transformers", "abstract": "We introduce BrainSAIL, a method for linking neural selectivity with spatially distributed semantic visual concepts in natural scenes. BrainSAIL leverages recent advances in large-scale artificial neural networks, using them to provide insights into the functional topology of the brain. To overcome the challenge presented by the co-occurrence of multiple categories in natural images, BrainSAIL exploits semantically consistent, dense spatial features from pre-trained vision models, building upon their demonstrated ability to robustly predict neural activity. This method derives clean, spatially dense embeddings without requiring any additional training, and employs a novel denoising process that leverages the semantic consistency of images under random augmentations. By unifying the space of whole-image embeddings and dense visual features and then applying voxel-wise encoding models to these features, we enable the identification of specific subregions of each image which drive selectivity patterns in different areas of the higher visual cortex. This provides a powerful tool for dissecting the neural mechanisms that underlie semantic visual processing for natural images. We validate BrainSAIL on cortical regions with known category selectivity, demonstrating its ability to accurately localize and disentangle selectivity to diverse visual concepts. Next, we demonstrate BrainSAIL's ability to characterize high-level visual selectivity to scene properties and low-level visual features such as depth, luminance, and saturation, providing insights into the encoding of complex visual information. Finally, we use BrainSAIL to directly compare the feature selectivity of different brain encoding models across different regions of interest in visual cortex. Our innovative method paves the way for significant advances in mapping and decomposing high-level visual representations in the human brain.", "date": "2024-10-07", "categories": []}
{"id": "2411.05673", "title": "Relationships between the degrees of freedom in the affine Gaussian derivative model for visual receptive fields and 2-D affine image transformations, with application to covariance properties of simple cells in the primary visual cortex", "abstract": "When observing the surface patterns of objects delimited by smooth surfaces, the projections of the surface patterns to the image domain will be subject to substantial variabilities, as induced by variabilities in the geometric viewing conditions, and as generated by either monocular or binocular imaging conditions, or by relative motions between the object and the observer over time. To first order of approximation, the image deformations of such projected surface patterns can be modelled as local linearizations in terms of local 2-D spatial affine transformations.\n  This paper presents a theoretical analysis of relationships between the degrees of freedom in 2-D spatial affine image transformations and the degrees of freedom in the affine Gaussian derivative model for visual receptive fields. For this purpose, we first describe a canonical decomposition of 2-D affine transformations on a product form, closely related to a singular value decomposition, while in closed form, and which reveals the degrees of freedom in terms of (i) uniform scaling transformations, (ii) an overall amount of global rotation, (iii) a complementary non-uniform scaling transformation and (iv) a relative normalization to a preferred symmetry orientation in the image domain. Then, we show how these degrees of freedom relate to the degrees of freedom in the affine Gaussian derivative model.\n  Finally, we use these theoretical results to consider whether we could regard the biological receptive fields in the primary visual cortex of higher mammals as being able to span the degrees of freedom of 2-D spatial affine transformations, based on interpretations of existing neurophysiological experimental results.", "date": "2024-11-08", "categories": []}
{"id": "2506.16009", "title": "Bridging Brain with Foundation Models through Self-Supervised Learning", "abstract": "Foundation models (FMs), powered by self-supervised learning (SSL), have redefined the capabilities of artificial intelligence, demonstrating exceptional performance in domains like natural language processing and computer vision. These advances present a transformative opportunity for brain signal analysis. Unlike traditional supervised learning, which is limited by the scarcity of labeled neural data, SSL offers a promising solution by enabling models to learn meaningful representations from unlabeled data. This is particularly valuable in addressing the unique challenges of brain signals, including high noise levels, inter-subject variability, and low signal-to-noise ratios. This survey systematically reviews the emerging field of bridging brain signals with foundation models through the innovative application of SSL. It explores key SSL techniques, the development of brain-specific foundation models, their adaptation to downstream tasks, and the integration of brain signals with other modalities in multimodal SSL frameworks. The review also covers commonly used evaluation metrics and benchmark datasets that support comparative analysis. Finally, it highlights key challenges and outlines future research directions. This work aims to provide researchers with a structured understanding of this rapidly evolving field and a roadmap for developing generalizable brain foundation models powered by self-supervision.", "date": "2025-06-19", "categories": []}
{"id": "2503.23394", "title": "Spatiotemporal Learning of Brain Dynamics from fMRI Using Frequency-Specific Multi-Band Attention for Cognitive and Psychiatric Applications", "abstract": "Understanding how the brain's complex nonlinear dynamics give rise to cognitive function remains a central challenge in neuroscience. While brain functional dynamics exhibits scale-free and multifractal properties across temporal scales, conventional neuroimaging analytics assume linearity and stationarity, failing to capture frequency-specific neural computations. Here, we introduce Multi-Band Brain Net (MBBN), the first transformer-based framework to explicitly model frequency-specific spatiotemporal brain dynamics from fMRI. MBBN integrates biologically-grounded frequency decomposition with multi-band self-attention mechanisms, enabling discovery of previously undetectable frequency-dependent network interactions. Trained on 49,673 individuals across three large-scale cohorts (UK Biobank, ABCD, ABIDE), MBBN sets a new state-of-the-art in predicting psychiatric and cognitive outcomes (depression, ADHD, ASD), showing particular strength in classification tasks with up to 52.5\\% higher AUROC and provides a novel framework for predicting cognitive intelligence scores. Frequency-resolved analyses uncover disorder-specific signatures: in ADHD, high-frequency fronto-sensorimotor connectivity is attenuated and opercular somatosensory nodes emerge as dynamic hubs; in ASD, orbitofrontal-somatosensory circuits show focal high-frequency disruption together with enhanced ultra-low-frequency coupling between the temporo-parietal junction and prefrontal cortex. By integrating scale-aware neural dynamics with deep learning, MBBN delivers more accurate and interpretable biomarkers, opening avenues for precision psychiatry and developmental neuroscience.", "date": "2025-03-30", "categories": []}
{"id": "2506.05633", "title": "Noninvasive precision modulation of high-level neural population activity via natural vision perturbations", "abstract": "Precise control of neural activity -- modulating target neurons deep in the brain while leaving nearby neurons unaffected -- is an outstanding challenge in neuroscience, generally approached using invasive techniques. This study investigates the possibility of precisely and noninvasively modulating neural activity in the high-level primate ventral visual stream via perturbations on one's natural visual feed. When tested on macaque inferior temporal (IT) neural populations, we found quantitative agreement between the model-predicted and biologically realized effect: strong modulation concentrated on targeted neural sites. We extended this to demonstrate accurate injection of experimenter-chosen neural population patterns via subtle perturbations applied on the background of typical natural visual feeds. These results highlight that current machine-executable models of the ventral stream can now design noninvasive, visually-delivered, possibly imperceptible neural interventions at the resolution of individual neurons.", "date": "2025-06-05", "categories": []}
{"id": "2506.12055", "title": "Towards Unified Neural Decoding with Brain Functional Network Modeling", "abstract": "Recent achievements in implantable brain-computer interfaces (iBCIs) have demonstrated the potential to decode cognitive and motor behaviors with intracranial brain recordings; however, individual physiological and electrode implantation heterogeneities have constrained current approaches to neural decoding within single individuals, rendering interindividual neural decoding elusive. Here, we present Multi-individual Brain Region-Aggregated Network (MIBRAIN), a neural decoding framework that constructs a whole functional brain network model by integrating intracranial neurophysiological recordings across multiple individuals. MIBRAIN leverages self-supervised learning to derive generalized neural prototypes and supports group-level analysis of brain-region interactions and inter-subject neural synchrony. To validate our framework, we recorded stereoelectroencephalography (sEEG) signals from a cohort of individuals performing Mandarin syllable articulation. Both real-time online and offline decoding experiments demonstrated significant improvements in both audible and silent articulation decoding, enhanced decoding accuracy with increased multi-subject data integration, and effective generalization to unseen subjects. Furthermore, neural predictions for regions without direct electrode coverage were validated against authentic neural data. Overall, this framework paves the way for robust neural decoding across individuals and offers insights for practical clinical applications.", "date": "2025-05-30", "categories": []}
{"id": "2506.13450", "title": "A Neural Model for Word Repetition", "abstract": "It takes several years for the developing brain of a baby to fully master word repetition-the task of hearing a word and repeating it aloud. Repeating a new word, such as from a new language, can be a challenging task also for adults. Additionally, brain damage, such as from a stroke, may lead to systematic speech errors with specific characteristics dependent on the location of the brain damage. Cognitive sciences suggest a model with various components for the different processing stages involved in word repetition. While some studies have begun to localize the corresponding regions in the brain, the neural mechanisms and how exactly the brain performs word repetition remain largely unknown. We propose to bridge the gap between the cognitive model of word repetition and neural mechanisms in the human brain by modeling the task using deep neural networks. Neural models are fully observable, allowing us to study the detailed mechanisms in their various substructures and make comparisons with human behavior and, ultimately, the brain. Here, we make first steps in this direction by: (1) training a large set of models to simulate the word repetition task; (2) creating a battery of tests to probe the models for known effects from behavioral studies in humans, and (3) simulating brain damage through ablation studies, where we systematically remove neurons from the model, and repeat the behavioral study to examine the resulting speech errors in the \"patient\" model. Our results show that neural models can mimic several effects known from human research, but might diverge in other aspects, highlighting both the potential and the challenges for future research aimed at developing human-like neural models.", "date": "2025-06-16", "categories": []}
{"id": "2506.11123", "title": "Sparse Autoencoders Bridge The Deep Learning Model and The Brain", "abstract": "We present SAE-BrainMap, a novel framework that directly aligns deep learning visual model representations with voxel-level fMRI responses using sparse autoencoders (SAEs). First, we train layer-wise SAEs on model activations and compute the correlations between SAE unit activations and cortical fMRI signals elicited by the same natural image stimuli with cosine similarity, revealing strong activation correspondence (maximum similarity up to 0.76). Depending on this alignment, we construct a voxel dictionary by optimally assigning the most similar SAE feature to each voxel, demonstrating that SAE units preserve the functional structure of predefined regions of interest (ROIs) and exhibit ROI-consistent selectivity. Finally, we establish fine-grained hierarchical mapping between model layers and the human ventral visual pathway, also by projecting voxel dictionary activations onto individual cortical surfaces, we visualize the dynamic transformation of the visual information in deep learning models. It is found that ViT-B/16$_{CLIP}$ tends to utilize low-level information to generate high-level semantic information in the early layers and reconstructs the low-dimension information later. Our results establish a direct, downstream-task-free bridge between deep neural networks and human visual cortex, offering new insights into model interpretability.", "date": "2025-06-10", "categories": []}
{"id": "2502.03081", "title": "Human-Aligned Image Models Improve Visual Decoding from the Brain", "abstract": "Decoding visual images from brain activity has significant potential for advancing brain-computer interaction and enhancing the understanding of human perception. Recent approaches align the representation spaces of images and brain activity to enable visual decoding. In this paper, we introduce the use of human-aligned image encoders to map brain signals to images. We hypothesize that these models more effectively capture perceptual attributes associated with the rapid visual stimuli presentations commonly used in visual brain data recording experiments. Our empirical results support this hypothesis, demonstrating that this simple modification improves image retrieval accuracy by up to 21% compared to state-of-the-art methods. Comprehensive experiments confirm consistent performance improvements across diverse EEG architectures, image encoders, alignment methods, participants, and brain imaging modalities", "date": "2025-02-05", "categories": []}
{"id": "2501.02333", "title": "On The Causal Network Of Face-selective Regions In Human Brain During Movie Watching", "abstract": "Understanding the causal interactions in some brain tasks, such as face processing, remains a challenging and ambiguous process for researchers. In this study, we address this issue by employing a novel causal discovery method -Directed Acyclic Graphs via M-matrices for Acyclicity (DAGMA)- to investigate the causal structure of the brain's face-selective network and gain deeper insights into its mechanism. Using fMRI data of natural movie stimuli, we extract causal network of face-selective regions and analyze how frames containing faces influence this network. Specifically, our findings reveal that the presence of faces in the stimuli, causally affects the number of identified connections within the network. Additionally, our results highlight the crucial role of subcortical regions in satisfying causal sufficiency, emphasizing it's importance in causal studies of brain. This study provides a new perspective on understanding the causal architecture of the face-selective network of the brain, motivating further research on neural causality.", "date": "2025-01-04", "categories": []}
{"id": "2505.13011", "title": "Unveiling and Steering Connectome Organization with Interpretable Latent Variables", "abstract": "The brain's intricate connectome, a blueprint for its function, presents immense complexity, yet it arises from a compact genetic code, hinting at underlying low-dimensional organizational principles. This work bridges connectomics and representation learning to uncover these principles. We propose a framework that combines subgraph extraction from the Drosophila connectome, FlyWire, with a generative model to derive interpretable low-dimensional representations of neural circuitry. Crucially, an explainability module links these latent dimensions to specific structural features, offering insights into their functional relevance. We validate our approach by demonstrating effective graph reconstruction and, significantly, the ability to manipulate these latent codes to controllably generate connectome subgraphs with predefined properties. This research offers a novel tool for understanding brain architecture and a potential avenue for designing bio-inspired artificial neural networks.", "date": "2025-05-19", "categories": []}
{"id": "2505.20480", "title": "BrainStratify: Coarse-to-Fine Disentanglement of Intracranial Neural Dynamics", "abstract": "Decoding speech directly from neural activity is a central goal in brain-computer interface (BCI) research. In recent years, exciting advances have been made through the growing use of intracranial field potential recordings, such as stereo-ElectroEncephaloGraphy (sEEG) and ElectroCorticoGraphy (ECoG). These neural signals capture rich population-level activity but present key challenges: (i) task-relevant neural signals are sparsely distributed across sEEG electrodes, and (ii) they are often entangled with task-irrelevant neural signals in both sEEG and ECoG. To address these challenges, we introduce a unified Coarse-to-Fine neural disentanglement framework, BrainStratify, which includes (i) identifying functional groups through spatial-context-guided temporal-spatial modeling, and (ii) disentangling distinct neural dynamics within the target functional group using Decoupled Product Quantization (DPQ). We evaluate BrainStratify on two open-source sEEG datasets and one (epidural) ECoG dataset, spanning tasks like vocal production and speech perception. Extensive experiments show that BrainStratify, as a unified framework for decoding speech from intracranial neural signals, significantly outperforms previous decoding methods. Overall, by combining data-driven stratification with neuroscience-inspired modularity, BrainStratify offers a robust and interpretable solution for speech decoding from intracranial recordings.", "date": "2025-05-26", "categories": []}
{"id": "2505.20027", "title": "Multi-modal brain encoding models for multi-modal stimuli", "abstract": "Despite participants engaging in unimodal stimuli, such as watching images or silent videos, recent work has demonstrated that multi-modal Transformer models can predict visual brain activity impressively well, even with incongruent modality representations. This raises the question of how accurately these multi-modal models can predict brain activity when participants are engaged in multi-modal stimuli. As these models grow increasingly popular, their use in studying neural activity provides insights into how our brains respond to such multi-modal naturalistic stimuli, i.e., where it separates and integrates information across modalities through a hierarchy of early sensory regions to higher cognition. We investigate this question by using multiple unimodal and two types of multi-modal models-cross-modal and jointly pretrained-to determine which type of model is more relevant to fMRI brain activity when participants are engaged in watching movies. We observe that both types of multi-modal models show improved alignment in several language and visual regions. This study also helps in identifying which brain regions process unimodal versus multi-modal information. We further investigate the contribution of each modality to multi-modal alignment by carefully removing unimodal features one by one from multi-modal representations, and find that there is additional information beyond the unimodal embeddings that is processed in the visual and language regions. Based on this investigation, we find that while for cross-modal models, their brain alignment is partially attributed to the video modality; for jointly pretrained models, it is partially attributed to both the video and audio modalities. This serves as a strong motivation for the neuroscience community to investigate the interpretability of these models for deepening our understanding of multi-modal information processing in brain.", "date": "2025-05-26", "categories": []}
{"id": "2505.17003", "title": "Sufficient conditions for offline reactivation in recurrent neural networks", "abstract": "During periods of quiescence, such as sleep, neural activity in many brain circuits resembles that observed during periods of task engagement. However, the precise conditions under which task-optimized networks can autonomously reactivate the same network states responsible for online behavior is poorly understood. In this study, we develop a mathematical framework that outlines sufficient conditions for the emergence of neural reactivation in circuits that encode features of smoothly varying stimuli. We demonstrate mathematically that noisy recurrent networks optimized to track environmental state variables using change-based sensory information naturally develop denoising dynamics, which, in the absence of input, cause the network to revisit state configurations observed during periods of online activity. We validate our findings using numerical experiments on two canonical neuroscience tasks: spatial position estimation based on self-motion cues, and head direction estimation based on angular velocity cues. Overall, our work provides theoretical support for modeling offline reactivation as an emergent consequence of task optimization in noisy neural circuits.", "date": "2025-05-22", "categories": []}
{"id": "2406.07726", "title": "A Concise Mathematical Description of Active Inference in Discrete Time", "abstract": "In this paper we present a concise mathematical description of active inference in discrete time. The main part of the paper serves as a basic introduction to the topic, including a detailed example of the action selection mechanism. The appendix discusses the more subtle mathematical details, targeting readers who have already studied the active inference literature but struggle to make sense of the mathematical details and derivations. Throughout, we emphasize precise and standard mathematical notation, ensuring consistency with existing texts and linking all equations to widely used references on active inference. Additionally, we provide Python code that implements the action selection and learning mechanisms described in this paper and is compatible with pymdp environments.", "date": "2024-06-11", "categories": []}
{"id": "2505.11485", "title": "Modeling cognitive processes of natural reading with transformer-based Language Models", "abstract": "Recent advances in Natural Language Processing (NLP) have led to the development of highly sophisticated language models for text generation. In parallel, neuroscience has increasingly employed these models to explore cognitive processes involved in language comprehension. Previous research has shown that models such as N-grams and LSTM networks can partially account for predictability effects in explaining eye movement behaviors, specifically Gaze Duration, during reading. In this study, we extend these findings by evaluating transformer-based models (GPT2, LLaMA-7B, and LLaMA2-7B) to further investigate this relationship. Our results indicate that these architectures outperform earlier models in explaining the variance in Gaze Durations recorded from Rioplantense Spanish readers. However, similar to previous studies, these models still fail to account for the entirety of the variance captured by human predictability. These findings suggest that, despite their advancements, state-of-the-art language models continue to predict language in ways that differ from human readers.", "date": "2025-05-16", "categories": []}
{"id": "2410.11516", "title": "TopoLM: brain-like spatio-functional organization in a topographic language model", "abstract": "Neurons in the brain are spatially organized such that neighbors on tissue often exhibit similar response profiles. In the human language system, experimental studies have observed clusters for syntactic and semantic categories, but the mechanisms underlying this functional organization remain unclear. Here, building on work from the vision literature, we develop TopoLM, a transformer language model with an explicit two-dimensional spatial representation of model units. By combining a next-token prediction objective with a spatial smoothness loss, representations in this model assemble into clusters that correspond to semantically interpretable groupings of text and closely match the functional organization in the brain's language system. TopoLM successfully predicts the emergence of the spatio-functional organization of a cortical language system as well as the organization of functional clusters selective for fine-grained linguistic features empirically observed in human cortex. Our results suggest that the functional organization of the human language system is driven by a unified spatial objective, and provide a functionally and spatially aligned model of language processing in the brain.", "date": "2024-10-15", "categories": []}
{"id": "2501.02982", "title": "A Bio-Inspired Research Paradigm of Collision Perception Neurons Enabling Neuro-Robotic Integration: The LGMD Case", "abstract": "Compared to human vision, locust visual systems excel at rapid and precise collision detection, despite relying on only hundreds of thousands of neurons organized through a few neuropils. This efficiency makes them an attractive model system for developing artificial collision-detecting systems. Specifically, researchers have identified collision-selective neurons in the locust's optic lobe, called lobula giant movement detectors (LGMDs), which respond specifically to approaching objects. Research upon LGMD neurons began in the early 1970s. Initially, due to their large size, these neurons were identified as motion detectors, but their role as looming detectors was recognized over time. Since then, progress in neuroscience, computational modeling of LGMD's visual neural circuits, and LGMD-based robotics have advanced in tandem, each field supporting and driving the others. Today, with a deeper understanding of LGMD neurons, LGMD-based models have significantly improved collision-free navigation in mobile robots including ground and aerial robots. This review highlights recent developments in LGMD research from the perspectives of neuroscience, computational modeling, and robotics. It emphasizes a biologically plausible research paradigm, where insights from neuroscience inform real-world applications, which would in turn validate and advance neuroscience. With strong support from extensive research and growing application demand, this paradigm has reached a mature stage and demonstrates versatility across different areas of neuroscience research, thereby enhancing our understanding of the interconnections between neuroscience, computational modeling, and robotics. Furthermore, this paradigm would shed light upon the modeling and robotic research into other motion-sensitive neurons or neural circuits.", "date": "2025-01-06", "categories": []}
{"id": "2505.06886", "title": "Mice to Machines: Neural Representations from Visual Cortex for Domain Generalization", "abstract": "The mouse is one of the most studied animal models in the field of systems neuroscience. Understanding the generalized patterns and decoding the neural representations that are evoked by the diverse range of natural scene stimuli in the mouse visual cortex is one of the key quests in computational vision. In recent years, significant parallels have been drawn between the primate visual cortex and hierarchical deep neural networks. However, their generalized efficacy in understanding mouse vision has been limited. In this study, we investigate the functional alignment between the mouse visual cortex and deep learning models for object classification tasks. We first introduce a generalized representational learning strategy that uncovers a striking resemblance between the functional mapping of the mouse visual cortex and high-performing deep learning models on both top-down (population-level) and bottom-up (single cell-level) scenarios. Next, this representational similarity across the two systems is further enhanced by the addition of Neural Response Normalization (NeuRN) layer, inspired by the activation profile of excitatory and inhibitory neurons in the visual cortex. To test the performance effect of NeuRN on real-world tasks, we integrate it into deep learning models and observe significant improvements in their robustness against data shifts in domain generalization tasks. Our work proposes a novel framework for comparing the functional architecture of the mouse visual cortex with deep learning models. Our findings carry broad implications for the development of advanced AI models that draw inspiration from the mouse visual cortex, suggesting that these models serve as valuable tools for studying the neural representations of the mouse visual cortex and, as a result, enhancing their performance on real-world tasks.", "date": "2025-05-11", "categories": []}
{"id": "2505.07620", "title": "Higher-Order Convolution Improves Neural Predictivity in the Retina", "abstract": "We present a novel approach to neural response prediction that incorporates higher-order operations directly within convolutional neural networks (CNNs). Our model extends traditional 3D CNNs by embedding higher-order operations within the convolutional operator itself, enabling direct modeling of multiplicative interactions between neighboring pixels across space and time. Our model increases the representational power of CNNs without increasing their depth, therefore addressing the architectural disparity between deep artificial networks and the relatively shallow processing hierarchy of biological visual systems. We evaluate our approach on two distinct datasets: salamander retinal ganglion cell (RGC) responses to natural scenes, and a new dataset of mouse RGC responses to controlled geometric transformations. Our higher-order CNN (HoCNN) achieves superior performance while requiring only half the training data compared to standard architectures, demonstrating correlation coefficients up to 0.75 with neural responses (against 0.80$\\pm$0.02 retinal reliability). When integrated into state-of-the-art architectures, our approach consistently improves performance across different species and stimulus conditions. Analysis of the learned representations reveals that our network naturally encodes fundamental geometric transformations, particularly scaling parameters that characterize object expansion and contraction. This capability is especially relevant for specific cell types, such as transient OFF-alpha and transient ON cells, which are known to detect looming objects and object motion respectively, and where our model shows marked improvement in response prediction. The correlation coefficients for scaling parameters are more than twice as high in HoCNN (0.72) compared to baseline models (0.32).", "date": "2025-05-12", "categories": []}
{"id": "2412.02529", "title": "Active learning of neural population dynamics using two-photon\n  holographic optogenetics", "abstract": "  Recent advances in techniques for monitoring and perturbing neural\npopulations have greatly enhanced our ability to study circuits in the brain.\nIn particular, two-photon holographic optogenetics now enables precise\nphotostimulation of experimenter-specified groups of individual neurons, while\nsimultaneous two-photon calcium imaging enables the measurement of ongoing and\ninduced activity across the neural population. Despite the enormous space of\npotential photostimulation patterns and the time-consuming nature of\nphotostimulation experiments, very little algorithmic work has been done to\ndetermine the most effective photostimulation patterns for identifying the\nneural population dynamics. Here, we develop methods to efficiently select\nwhich neurons to stimulate such that the resulting neural responses will best\ninform a dynamical model of the neural population activity. Using neural\npopulation responses to photostimulation in mouse motor cortex, we demonstrate\nthe efficacy of a low-rank linear dynamical systems model, and develop an\nactive learning procedure which takes advantage of low-rank structure to\ndetermine informative photostimulation patterns. We demonstrate our approach on\nboth real and synthetic data, obtaining in some cases as much as a two-fold\nreduction in the amount of data required to reach a given predictive power. Our\nactive stimulation design method is based on a novel active learning procedure\nfor low-rank regression, which may be of independent interest.\n", "date": "2024-12-03", "categories": []}
{"id": "2505.03856", "title": "An Active Inference Model of Covert and Overt Visual Attention", "abstract": "  The ability to selectively attend to relevant stimuli while filtering out\ndistractions is essential for agents that process complex, high-dimensional\nsensory input. This paper introduces a model of covert and overt visual\nattention through the framework of active inference, utilizing dynamic\noptimization of sensory precisions to minimize free-energy. The model\ndetermines visual sensory precisions based on both current environmental\nbeliefs and sensory input, influencing attentional allocation in both covert\nand overt modalities. To test the effectiveness of the model, we analyze its\nbehavior in the Posner cueing task and a simple target focus task using\ntwo-dimensional(2D) visual data. Reaction times are measured to investigate the\ninterplay between exogenous and endogenous attention, as well as valid and\ninvalid cueing. The results show that exogenous and valid cues generally lead\nto faster reaction times compared to endogenous and invalid cues. Furthermore,\nthe model exhibits behavior similar to inhibition of return, where previously\nattended locations become suppressed after a specific cue-target onset\nasynchrony interval. Lastly, we investigate different aspects of overt\nattention and show that involuntary, reflexive saccades occur faster than\nintentional ones, but at the expense of adaptability.\n", "date": "2025-05-06", "categories": []}
{"id": "2505.03308", "title": "An Active Inference perspective on Neurofeedback Training", "abstract": "  Neurofeedback training (NFT) aims to teach self-regulation of brain activity\nthrough real-time feedback, but suffers from highly variable outcomes and\npoorly understood mechanisms, hampering its validation. To address these\nissues, we propose a formal computational model of the NFT closed loop. Using\nActive Inference, a Bayesian framework modelling perception, action, and\nlearning, we simulate agents interacting with an NFT environment. This enables\nus to test the impact of design choices (e.g., feedback quality, biomarker\nvalidity) and subject factors (e.g., prior beliefs) on training. Simulations\nshow that training effectiveness is sensitive to feedback noise or bias, and to\nprior beliefs (highlighting the importance of guiding instructions), but also\nreveal that perfect feedback is insufficient to guarantee high performance.\nThis approach provides a tool for assessing and predicting NFT variability,\ninterpret empirical data, and potentially develop personalized training\nprotocols.\n", "date": "2025-05-06", "categories": []}
{"id": "2505.02248", "title": "Heterosynaptic Circuits Are Universal Gradient Machines", "abstract": "  We propose a design principle for the learning circuits of the biological\nbrain. The principle states that almost any dendritic weights updated via\nheterosynaptic plasticity can implement a generalized and efficient class of\ngradient-based meta-learning. The theory suggests that a broad class of\nbiologically plausible learning algorithms, together with the standard machine\nlearning optimizers, can be grounded in heterosynaptic circuit motifs. This\nprinciple suggests that the phenomenology of (anti-) Hebbian (HBP) and\nheterosynaptic plasticity (HSP) may emerge from the same underlying dynamics,\nthus providing a unifying explanation. It also suggests an alternative\nperspective of neuroplasticity, where HSP is promoted to the primary learning\nand memory mechanism, and HBP is an emergent byproduct. We present simulations\nthat show that (a) HSP can explain the metaplasticity of neurons, (b) HSP can\nexplain the flexibility of the biology circuits, and (c) gradient learning can\narise quickly from simple evolutionary dynamics that do not compute any\nexplicit gradient. While our primary focus is on biology, the principle also\nimplies a new approach to designing AI training algorithms and physically\nlearnable AI hardware. Conceptually, our result demonstrates that contrary to\nthe common belief, gradient computation may be extremely easy and common in\nnature.\n", "date": "2025-05-04", "categories": []}
{"id": "2502.11141", "title": "Cognitive Neural Architecture Search Reveals Hierarchical Entailment", "abstract": "  Recent research has suggested that the brain is more shallow than previously\nthought, challenging the traditionally assumed hierarchical structure of the\nventral visual pathway. Here, we demonstrate that optimizing convolutional\nnetwork architectures for brain-alignment via evolutionary neural architecture\nsearch results in models with clear representational hierarchies. Despite\nhaving random weights, the identified models achieve brain-alignment scores\nsurpassing even those of pretrained classification models - as measured by both\nregression and representational similarity analysis. Furthermore, through\ntraditional supervised training, architectures optimized for alignment with\nlate ventral regions become competitive classification models. These findings\nsuggest that hierarchical structure is a fundamental mechanism of primate\nvisual processing. Finally, this work demonstrates the potential of neural\narchitecture search as a framework for computational cognitive neuroscience\nresearch that could reduce the field's reliance on manually designed\nconvolutional networks.\n", "date": "2025-02-16", "categories": []}
{"id": "2504.16940", "title": "Better artificial intelligence does not mean better models of biology", "abstract": "  Deep neural networks (DNNs) once showed increasing alignment with primate\nperception and neural responses as they improved on vision benchmarks, raising\nhopes that advances in AI would yield better models of biological vision.\nHowever, we show across three benchmarks that this alignment is now plateauing\n- and in some cases worsening - as DNNs scale to human or superhuman accuracy.\nThis divergence may reflect the adoption of visual strategies that differ from\nthose used by primates. These findings challenge the view that progress in\nartificial intelligence will naturally translate to neuroscience. We argue that\nvision science must chart its own course, developing algorithms grounded in\nbiological visual systems rather than optimizing for benchmarks based on\ninternet-scale datasets.\n", "date": "2025-04-08", "categories": []}
{"id": "2312.14871", "title": "BrainVis: Exploring the Bridge between Brain and Visual Signals via\n  Image Reconstruction", "abstract": "  Analyzing and reconstructing visual stimuli from brain signals effectively\nadvances the understanding of human visual system. However, the EEG signals are\ncomplex and contain significant noise. This leads to substantial limitations in\nexisting works of visual stimuli reconstruction from EEG, such as difficulties\nin aligning EEG embeddings with the fine-grained semantic information and a\nheavy reliance on additional large self-collected dataset for training. To\naddress these challenges, we propose a novel approach called BrainVis. Firstly,\nwe divide the EEG signals into various units and apply a self-supervised\napproach on them to obtain EEG time-domain features, in an attempt to ease the\ntraining difficulty. Additionally, we also propose to utilize the\nfrequency-domain features to enhance the EEG representations. Then, we\nsimultaneously align EEG time-frequency embeddings with the interpolation of\nthe coarse and fine-grained semantics in the CLIP space, to highlight the\nprimary visual components and reduce the cross-modal alignment difficulty.\nFinally, we adopt the cascaded diffusion models to reconstruct images. Using\nonly 10\\% training data of the previous work, our proposed BrainVis outperforms\nstate of the arts in both semantic fidelity reconstruction and generation\nquality. The code is available at https://github.com/RomGai/BrainVis.\n", "date": "2023-12-22", "categories": []}
{"id": "2504.16917", "title": "Application of an attention-based CNN-BiLSTM framework for in vivo\n  two-photon calcium imaging of neuronal ensembles: decoding complex bilateral\n  forelimb movements from unilateral M1", "abstract": "  Decoding behavior, such as movement, from multiscale brain networks remains a\ncentral objective in neuroscience. Over the past decades, artificial\nintelligence and machine learning have played an increasingly significant role\nin elucidating the neural mechanisms underlying motor function. The advancement\nof brain-monitoring technologies, capable of capturing complex neuronal signals\nwith high spatial and temporal resolution, necessitates the development and\napplication of more sophisticated machine learning models for behavioral\ndecoding. In this study, we employ a hybrid deep learning framework, an\nattention-based CNN-BiLSTM model, to decode skilled and complex forelimb\nmovements using signals obtained from in vivo two-photon calcium imaging. Our\nfindings demonstrate that the intricate movements of both ipsilateral and\ncontralateral forelimbs can be accurately decoded from unilateral M1 neuronal\nensembles. These results highlight the efficacy of advanced hybrid deep\nlearning models in capturing the spatiotemporal dependencies of neuronal\nnetworks activity linked to complex movement execution.\n", "date": "2025-04-23", "categories": []}
{"id": "2504.15953", "title": "Visual Place Cell Encoding: A Computational Model for Spatial\n  Representation and Cognitive Mapping", "abstract": "  This paper presents the Visual Place Cell Encoding (VPCE) model, a\nbiologically inspired computational framework for simulating place cell-like\nactivation using visual input. Drawing on evidence that visual landmarks play a\ncentral role in spatial encoding, the proposed VPCE model activates visual\nplace cells by clustering high-dimensional appearance features extracted from\nimages captured by a robot-mounted camera. Each cluster center defines a\nreceptive field, and activation is computed based on visual similarity using a\nradial basis function. We evaluate whether the resulting activation patterns\ncorrelate with key properties of biological place cells, including spatial\nproximity, orientation alignment, and boundary differentiation. Experiments\ndemonstrate that the VPCE can distinguish between visually similar yet\nspatially distinct locations and adapt to environment changes such as the\ninsertion or removal of walls. These results suggest that structured visual\ninput, even in the absence of motion cues or reward-driven learning, is\nsufficient to generate place-cell-like spatial representations and support\nbiologically inspired cognitive mapping.\n", "date": "2025-04-22", "categories": []}
{"id": "2504.14468", "title": "sEEG-based Encoding for Sentence Retrieval: A Contrastive Learning\n  Approach to Brain-Language Alignment", "abstract": "  Interpreting neural activity through meaningful latent representations\nremains a complex and evolving challenge at the intersection of neuroscience\nand artificial intelligence. We investigate the potential of multimodal\nfoundation models to align invasive brain recordings with natural language. We\npresent SSENSE, a contrastive learning framework that projects single-subject\nstereo-electroencephalography (sEEG) signals into the sentence embedding space\nof a frozen CLIP model, enabling sentence-level retrieval directly from brain\nactivity. SSENSE trains a neural encoder on spectral representations of sEEG\nusing InfoNCE loss, without fine-tuning the text encoder. We evaluate our\nmethod on time-aligned sEEG and spoken transcripts from a naturalistic\nmovie-watching dataset. Despite limited data, SSENSE achieves promising\nresults, demonstrating that general-purpose language representations can serve\nas effective priors for neural decoding.\n", "date": "2025-04-20", "categories": []}
{"id": "2405.14139", "title": "Contribute to balance, wire in accordance: Emergence of backpropagation\n  from a simple, bio-plausible neuroplasticity rule", "abstract": "  Backpropagation (BP) has been pivotal in advancing machine learning and\nremains essential in computational applications and comparative studies of\nbiological and artificial neural networks. Despite its widespread use, the\nimplementation of BP in the brain remains elusive, and its biological\nplausibility is often questioned due to inherent issues such as the need for\nsymmetry of weights between forward and backward connections, and the\nrequirement of distinct forward and backward phases of computation. Here, we\nintroduce a novel neuroplasticity rule that offers a potential mechanism for\nimplementing BP in the brain. Similar in general form to the classical Hebbian\nrule, this rule is based on the core principles of maintaining the balance of\nexcitatory and inhibitory inputs as well as on retrograde signaling, and\noperates over three progressively slower timescales: neural firing, retrograde\nsignaling, and neural plasticity. We hypothesize that each neuron possesses an\ninternal state, termed credit, in addition to its firing rate. After achieving\nequilibrium in firing rates, neurons receive credits based on their\ncontribution to the E-I balance of postsynaptic neurons through retrograde\nsignaling. As the network's credit distribution stabilizes, connections from\nthose presynaptic neurons are strengthened that significantly contribute to the\nbalance of postsynaptic neurons. We demonstrate mathematically that our\nlearning rule precisely replicates BP in layered neural networks without any\napproximations. Simulations on artificial neural networks reveal that this rule\ninduces varying community structures in networks, depending on the learning\nrate. This simple theoretical framework presents a biologically plausible\nimplementation of BP, with testable assumptions and predictions that may be\nevaluated through biological experiments.\n", "date": "2024-05-23", "categories": []}
{"id": "2504.09213", "title": "Spiking Neural Network for Intra-cortical Brain Signal Decoding", "abstract": "  Decoding brain signals accurately and efficiently is crucial for\nintra-cortical brain-computer interfaces. Traditional decoding approaches based\non neural activity vector features suffer from low accuracy, whereas deep\nlearning based approaches have high computational cost. To improve both the\ndecoding accuracy and efficiency, this paper proposes a spiking neural network\n(SNN) for effective and energy-efficient intra-cortical brain signal decoding.\nWe also propose a feature fusion approach, which integrates the manually\nextracted neural activity vector features with those extracted by a deep neural\nnetwork, to further improve the decoding accuracy. Experiments in decoding\nmotor-related intra-cortical brain signals of two rhesus macaques demonstrated\nthat our SNN model achieved higher accuracy than traditional artificial neural\nnetworks; more importantly, it was tens or hundreds of times more efficient.\nThe SNN model is very suitable for high precision and low power applications\nlike intra-cortical brain-computer interfaces.\n", "date": "2025-04-12", "categories": []}
{"id": "2504.10028", "title": "Sequence models for by-trial decoding of cognitive strategies from\n  neural data", "abstract": "  Understanding the sequence of cognitive operations that underlie\ndecision-making is a fundamental challenge in cognitive neuroscience.\nTraditional approaches often rely on group-level statistics, which obscure\ntrial-by-trial variations in cognitive strategies. In this study, we introduce\na novel machine learning method that combines Hidden Multivariate Pattern\nanalysis with a Structured State Space Sequence model to decode cognitive\nstrategies from electroencephalography data at the trial level. We apply this\nmethod to a decision-making task, where participants were instructed to\nprioritize either speed or accuracy in their responses. Our results reveal an\nadditional cognitive operation, labeled Confirmation, which seems to occur\npredominantly in the accuracy condition but also frequently in the speed\ncondition. The modeled probability that this operation occurs is associated\nwith higher probability of responding correctly as well as changes of mind, as\nindexed by electromyography data. By successfully modeling cognitive operations\nat the trial level, we provide empirical evidence for dynamic variability in\ndecision strategies, challenging the assumption of homogeneous cognitive\nprocesses within experimental conditions. Our approach shows the potential of\nsequence modeling in cognitive neuroscience to capture trial-level variability\nthat is obscured by aggregate analyses. The introduced method offers a new way\nto detect and understand cognitive strategies in a data-driven manner, with\nimplications for both theoretical research and practical applications in many\nfields.\n", "date": "2025-04-14", "categories": []}
{"id": "2406.02659", "title": "Reanimating Images using Neural Representations of Dynamic Stimuli", "abstract": "  While computer vision models have made incredible strides in static image\nrecognition, they still do not match human performance in tasks that require\nthe understanding of complex, dynamic motion. This is notably true for\nreal-world scenarios where embodied agents face complex and motion-rich\nenvironments. Our approach, BrainNRDS (Brain-Neural Representations of Dynamic\nStimuli), leverages state-of-the-art video diffusion models to decouple static\nimage representation from motion generation, enabling us to utilize fMRI brain\nactivity for a deeper understanding of human responses to dynamic visual\nstimuli. Conversely, we also demonstrate that information about the brain's\nrepresentation of motion can enhance the prediction of optical flow in\nartificial systems. Our novel approach leads to four main findings: (1) Visual\nmotion, represented as fine-grained, object-level resolution optical flow, can\nbe decoded from brain activity generated by participants viewing video stimuli;\n(2) Video encoders outperform image-based models in predicting video-driven\nbrain activity; (3) Brain-decoded motion signals enable realistic video\nreanimation based only on the initial frame of the video; and (4) We extend\nprior work to achieve full video decoding from video-driven brain activity.\nBrainNRDS advances our understanding of how the brain represents spatial and\ntemporal information in dynamic visual scenes. Our findings demonstrate the\npotential of combining brain imaging with video diffusion models for developing\nmore robust and biologically-inspired computer vision systems. We show\nadditional decoding and encoding examples on this site:\nhttps://brain-nrds.github.io/.\n", "date": "2024-06-04", "categories": []}
{"id": "2406.12065", "title": "STNAGNN: Data-driven Spatio-temporal Brain Connectivity beyond FC", "abstract": "  In recent years, graph neural networks (GNNs) have been widely applied in the\nanalysis of brain fMRI, yet defining the connectivity between ROIs remains a\nchallenge in noisy fMRI data. Among all approaches, Functional Connectome (FC)\nis the most popular method. Computed by the correlation coefficients between\nROI time series, FC is a powerful and computationally efficient way to estimate\nROI connectivity. However, it is well known for neglecting structural\nconnections and causality in ROI interactions. Also, FC becomes much more noisy\nin the short spatio-temporal sliding-window subsequences of fMRI. Effective\nConnectome (EC) is proposed as a directional alternative, but is difficult to\naccurately estimate. Furthermore, for optimal GNN performance, usually only a\nsmall percentage of the strongest connections are selected as sparse edges,\nresulting in oversimplification of complex brain connections. To tackle these\nchallenges, we propose the Spatio-Temporal Node Attention Graph Neural Network\n(STNAGNN) as a data-driven alternative that combines sparse predefined FC with\ndense data-driven spatio-temporal connections, allowing for flexible and\nspatio-temporal learning of ROI interaction patterns.\n", "date": "2024-06-17", "categories": []}
{"id": "2410.05454", "title": "Meta-Dynamical State Space Models for Integrative Neural Data Analysis", "abstract": "  Learning shared structure across environments facilitates rapid learning and\nadaptive behavior in neural systems. This has been widely demonstrated and\napplied in machine learning to train models that are capable of generalizing to\nnovel settings. However, there has been limited work exploiting the shared\nstructure in neural activity during similar tasks for learning latent dynamics\nfrom neural recordings. Existing approaches are designed to infer dynamics from\na single dataset and cannot be readily adapted to account for statistical\nheterogeneities across recordings. In this work, we hypothesize that similar\ntasks admit a corresponding family of related solutions and propose a novel\napproach for meta-learning this solution space from task-related neural\nactivity of trained animals. Specifically, we capture the variabilities across\nrecordings on a low-dimensional manifold which concisely parametrizes this\nfamily of dynamics, thereby facilitating rapid learning of latent dynamics\ngiven new recordings. We demonstrate the efficacy of our approach on few-shot\nreconstruction and forecasting of synthetic dynamical systems, and neural\nrecordings from the motor cortex during different arm reaching tasks.\n", "date": "2024-10-07", "categories": []}
{"id": "2502.14671", "title": "Explanations of Large Language Models Explain Language Representations\n  in the Brain", "abstract": "  Large language models (LLMs) not only exhibit human-like performance but also\nshare computational principles with the brain's language processing mechanisms.\nWhile prior research has focused on mapping LLMs' internal representations to\nneural activity, we propose a novel approach using explainable AI (XAI) to\nstrengthen this link. Applying attribution methods, we quantify the influence\nof preceding words on LLMs' next-word predictions and use these explanations to\npredict fMRI data from participants listening to narratives. We find that\nattribution methods robustly predict brain activity across the language\nnetwork, revealing a hierarchical pattern: explanations from early layers align\nwith the brain's initial language processing stages, while later layers\ncorrespond to more advanced stages. Additionally, layers with greater influence\non next-word prediction$\\unicode{x2014}$reflected in higher attribution\nscores$\\unicode{x2014}$demonstrate stronger brain alignment. These results\nunderscore XAI's potential for exploring the neural basis of language and\nsuggest brain alignment for assessing the biological plausibility of\nexplanation methods.\n", "date": "2025-02-20", "categories": []}
{"id": "2406.03044", "title": "Population Transformer: Learning Population-level Representations of\n  Neural Activity", "abstract": "  We present a self-supervised framework that learns population-level codes for\narbitrary ensembles of neural recordings at scale. We address key challenges in\nscaling models with neural time-series data, namely, sparse and variable\nelectrode distribution across subjects and datasets. The Population Transformer\n(PopT) stacks on top of pretrained temporal embeddings and enhances downstream\ndecoding by enabling learned aggregation of multiple spatially-sparse data\nchannels. The pretrained PopT lowers the amount of data required for downstream\ndecoding experiments, while increasing accuracy, even on held-out subjects and\ntasks. Compared to end-to-end methods, this approach is computationally\nlightweight, while achieving similar or better decoding performance. We further\nshow how our framework is generalizable to multiple time-series embeddings and\nneural data modalities. Beyond decoding, we interpret the pretrained and\nfine-tuned PopT models to show how they can be used to extract neuroscience\ninsights from large amounts of data. We release our code as well as a\npretrained PopT to enable off-the-shelf improvements in multi-channel\nintracranial data decoding and interpretability. Code is available at\nhttps://github.com/czlwang/PopulationTransformer.\n", "date": "2024-06-05", "categories": []}
{"id": "2407.09538", "title": "A Dynamic Systems Approach to Modelling Human-Machine Rhythm Interaction", "abstract": "  In exploring the simulation of human rhythmic perception and synchronization\ncapabilities, this study introduces a computational model inspired by the\nphysical and biological processes underlying rhythm processing. Utilizing a\nreservoir computing framework that simulates the function of cerebellum, the\nmodel features a dual-neuron classification and incorporates parameters to\nmodulate information transfer, reflecting biological neural network\ncharacteristics. Our findings demonstrate the model's ability to accurately\nperceive and adapt to rhythmic patterns within the human perceptible range,\nexhibiting behavior closely aligned with human rhythm interaction. By\nincorporating fine-tuning mechanisms and delay-feedback, the model enables\ncontinuous learning and precise rhythm prediction. The introduction of\ncustomized settings further enhances its capacity to stimulate diverse human\nrhythmic behaviors, underscoring the potential of this architecture in temporal\ncognitive task modeling and the study of rhythm synchronization and prediction\nin artificial and biological systems. Therefore, our model is capable of\ntransparently modelling cognitive theories that elucidate the dynamic processes\nby which the brain generates rhythm-related behavior.\n", "date": "2024-06-26", "categories": []}
{"id": "2503.19923", "title": "Mapping fMRI Signal and Image Stimuli in an Artificial Neural Network\n  Latent Space: Bringing Artificial and Natural Minds Together", "abstract": "  The goal of this study is to investigate whether latent space representations\nof visual stimuli and fMRI data share common information. Decoding and\nreconstructing stimuli from fMRI data remains a challenge in AI and\nneuroscience, with significant implications for understanding neural\nrepresentations and improving the interpretability of Artificial Neural\nNetworks (ANNs). In this preliminary study, we investigate the feasibility of\nsuch reconstruction by examining the similarity between the latent spaces of\none autoencoder (AE) and one vision transformer (ViT) trained on fMRI and image\ndata, respectively. Using representational similarity analysis (RSA), we found\nthat the latent spaces of the two domains appear different. However, these\ninitial findings are inconclusive, and further research is needed to explore\nthis relationship more thoroughly.\n", "date": "2025-03-12", "categories": []}
{"id": "2503.15978", "title": "A Survey on fMRI-based Brain Decoding for Reconstructing Multimodal\n  Stimuli", "abstract": "  In daily life, we encounter diverse external stimuli, such as images, sounds,\nand videos. As research in multimodal stimuli and neuroscience advances,\nfMRI-based brain decoding has become a key tool for understanding brain\nperception and its complex cognitive processes. Decoding brain signals to\nreconstruct stimuli not only reveals intricate neural mechanisms but also\ndrives progress in AI, disease treatment, and brain-computer interfaces. Recent\nadvancements in neuroimaging and image generation models have significantly\nimproved fMRI-based decoding. While fMRI offers high spatial resolution for\nprecise brain activity mapping, its low temporal resolution and signal noise\npose challenges. Meanwhile, techniques like GANs, VAEs, and Diffusion Models\nhave enhanced reconstructed image quality, and multimodal pre-trained models\nhave boosted cross-modal decoding tasks. This survey systematically reviews\nrecent progress in fMRI-based brain decoding, focusing on stimulus\nreconstruction from passive brain signals. It summarizes datasets, relevant\nbrain regions, and categorizes existing methods by model structure.\nAdditionally, it evaluates model performance and discusses their effectiveness.\nFinally, it identifies key challenges and proposes future research directions,\noffering valuable insights for the field. For more information and resources\nrelated to this survey, visit https://github.com/LpyNow/BrainDecodingImage.\n", "date": "2025-03-20", "categories": []}
{"id": "2503.16085", "title": "Allostatic Control of Persistent States in Spiking Neural Networks for\n  perception and computation", "abstract": "  We introduce a novel model for updating perceptual beliefs about the\nenvironment by extending the concept of Allostasis to the control of internal\nrepresentations. Allostasis is a fundamental regulatory mechanism observed in\nanimal physiology that orchestrates responses to maintain a dynamic equilibrium\nin bodily needs and internal states. In this paper, we focus on an application\nin numerical cognition, where a bump of activity in an attractor network is\nused as a spatial numerical representation. While existing neural networks can\nmaintain persistent states, to date, there is no unified framework for\ndynamically controlling spatial changes in neuronal activity in response to\nenvironmental changes. To address this, we couple a well known allostatic\nmicrocircuit, the Hammel model, with a ring attractor, resulting in a Spiking\nNeural Network architecture that can modulate the location of the bump as a\nfunction of some reference input. This localized activity in turn is used as a\nperceptual belief in a simulated subitization task a quick enumeration process\nwithout counting. We provide a general procedure to fine-tune the model and\ndemonstrate the successful control of the bump location. We also study the\nresponse time in the model with respect to changes in parameters and compare it\nwith biological data. Finally, we analyze the dynamics of the network to\nunderstand the selectivity and specificity of different neurons to distinct\ncategories present in the input. The results of this paper, particularly the\nmechanism for moving persistent states, are not limited to numerical cognition\nbut can be applied to a wide range of tasks involving similar representations.\n", "date": "2025-03-20", "categories": []}
{"id": "2406.05002", "title": "Deep Jansen-Rit Parameter Inference for Model-Driven Analysis of Brain\n  Activity", "abstract": "  Accurately modeling effective connectivity (EC) is critical for understanding\nhow the brain processes and integrates sensory information. Yet, it remains a\nformidable challenge due to complex neural dynamics and noisy measurements such\nas those obtained from the electroencephalogram (EEG). Model-driven EC infers\nlocal (within a brain region) and global (between brain regions) EC parameters\nby fitting a generative model of neural activity onto experimental data. This\napproach offers a promising route for various applications, including\ninvestigating neurodevelopmental disorders. However, current approaches fail to\nscale to whole-brain analyses and are highly noise-sensitive. In this work, we\nemploy three deep-learning architectures--a transformer, a long short-term\nmemory (LSTM) network, and a convolutional neural network and bidirectional\nLSTM (CNN-BiLSTM) network--for inverse modeling and compare their performance\nwith simulation-based inference in estimating the Jansen-Rit neural mass model\n(JR-NMM) parameters from simulated EEG data under various noise conditions. We\ndemonstrate a reliable estimation of key local parameters, such as synaptic\ngains and time constants. However, other parameters like local JR-NMM\nconnectivity cannot be evaluated reliably from evoked-related potentials (ERP).\nWe also conduct a sensitivity analysis to characterize the influence of JR-NMM\nparameters on ERP and evaluate their learnability. Our results show the\nfeasibility of deep-learning approaches to estimate the subset of learnable\nJR-NMM parameters.\n", "date": "2024-06-07", "categories": []}
{"id": "2503.12509", "title": "A Reservoir-based Model for Human-like Perception of Complex Rhythm\n  Pattern", "abstract": "  Rhythm is a fundamental aspect of human behaviour, present from infancy and\ndeeply embedded in cultural practices. Rhythm anticipation is a spontaneous\ncognitive process that typically occurs before the onset of actual beats. While\nmost research in both neuroscience and artificial intelligence has focused on\nmetronome-based rhythm tasks, studies investigating the perception of complex\nmusical rhythm patterns remain limited. To address this gap, we propose a\nhierarchical oscillator-based model to better understand the perception of\ncomplex musical rhythms in biological systems. The model consists of two types\nof coupled neurons that generate oscillations, with different layers tuned to\nrespond to distinct perception levels. We evaluate the model using several\nrepresentative rhythm patterns spanning the upper, middle, and lower bounds of\nhuman musical perception. Our findings demonstrate that, while maintaining a\nhigh degree of synchronization accuracy, the model exhibits human-like rhythmic\nbehaviours. Additionally, the beta band neuronal activity in the model mirrors\npatterns observed in the human brain, further validating the biological\nplausibility of the approach.\n", "date": "2025-03-16", "categories": []}
{"id": "2503.10518", "title": "Why the Brain Cannot Be a Digital Computer: History-Dependence and the\n  Computational Limits of Consciousness", "abstract": "  This paper presents a novel information-theoretic proof demonstrating that\nthe human brain as currently understood cannot function as a classical digital\ncomputer. Through systematic quantification of distinguishable conscious states\nand their historical dependencies, we establish that the minimum information\nrequired to specify a conscious state exceeds the physical information capacity\nof the human brain by a significant factor. Our analysis calculates the\nbit-length requirements for representing consciously distinguishable sensory\n\"stimulus frames\" and demonstrates that consciousness exhibits mandatory\ntemporal-historical dependencies that multiply these requirements beyond the\nbrain's storage capabilities. This mathematical approach offers new insights\ninto the fundamental limitations of computational models of consciousness and\nsuggests that non-classical information processing mechanisms may be necessary\nto account for conscious experience.\n", "date": "2025-03-13", "categories": []}
{"id": "2503.08608", "title": "A Grid Cell-Inspired Structured Vector Algebra for Cognitive Maps", "abstract": "  The entorhinal-hippocampal formation is the mammalian brain's navigation\nsystem, encoding both physical and abstract spaces via grid cells. This system\nis well-studied in neuroscience, and its efficiency and versatility make it\nattractive for applications in robotics and machine learning. While continuous\nattractor networks (CANs) successfully model entorhinal grid cells for encoding\nphysical space, integrating both continuous spatial and abstract spatial\ncomputations into a unified framework remains challenging. Here, we attempt to\nbridge this gap by proposing a mechanistic model for versatile information\nprocessing in the entorhinal-hippocampal formation inspired by CANs and Vector\nSymbolic Architectures (VSAs), a neuro-symbolic computing framework. The novel\ngrid-cell VSA (GC-VSA) model employs a spatially structured encoding scheme\nwith 3D neuronal modules mimicking the discrete scales and orientations of grid\ncell modules, reproducing their characteristic hexagonal receptive fields. In\nexperiments, the model demonstrates versatility in spatial and abstract tasks:\n(1) accurate path integration for tracking locations, (2) spatio-temporal\nrepresentation for querying object locations and temporal relations, and (3)\nsymbolic reasoning using family trees as a structured test case for\nhierarchical relationships.\n", "date": "2025-03-11", "categories": []}
{"id": "2503.03784", "title": "Neural Models of Task Adaptation: A Tutorial on Spiking Networks for\n  Executive Control", "abstract": "  Understanding cognitive flexibility and task-switching mechanisms in neural\nsystems requires biologically plausible computational models. This tutorial\npresents a step-by-step approach to constructing a spiking neural network (SNN)\nthat simulates task-switching dynamics within the cognitive control network.\nThe model incorporates biologically realistic features, including lateral\ninhibition, adaptive synaptic weights through unsupervised Spike\nTiming-Dependent Plasticity (STDP), and precise neuronal parameterization\nwithin physiologically relevant ranges. The SNN is implemented using Leaky\nIntegrate-and-Fire (LIF) neurons, which represent excitatory (glutamatergic)\nand inhibitory (GABAergic) populations. We utilize two real-world datasets as\ntasks, demonstrating how the network learns and dynamically switches between\nthem. Experimental design follows cognitive psychology paradigms to analyze\nneural adaptation, synaptic weight modifications, and emergent behaviors such\nas Long-Term Potentiation (LTP), Long-Term Depression (LTD), and Task-Set\nReconfiguration (TSR). Through a series of structured experiments, this\ntutorial illustrates how variations in task-switching intervals affect\nperformance and multitasking efficiency. The results align with empirically\nobserved neuronal responses, offering insights into the computational\nunderpinnings of executive function. By following this tutorial, researchers\ncan develop and extend biologically inspired SNN models for studying cognitive\nprocesses and neural adaptation.\n", "date": "2025-03-05", "categories": []}
{"id": "2503.03545", "title": "Revisiting the Role of Relearning in Semantic Dementia", "abstract": "  Patients with semantic dementia (SD) present with remarkably consistent\natrophy of neurons in the anterior temporal lobe and behavioural impairments,\nsuch as graded loss of category knowledge. While relearning of lost knowledge\nhas been shown in acute brain injuries such as stroke, it has not been widely\nsupported in chronic cognitive diseases such as SD. Previous research has shown\nthat deep linear artificial neural networks exhibit stages of semantic learning\nakin to humans. Here, we use a deep linear network to test the hypothesis that\nrelearning during disease progression rather than particular atrophy cause the\nspecific behavioural patterns associated with SD. After training the network to\ngenerate the common semantic features of various hierarchically organised\nobjects, neurons are successively deleted to mimic atrophy while retraining the\nmodel. The model with relearning and deleted neurons reproduced errors specific\nto SD, including prototyping errors and cross-category confusions. This\nsuggests that relearning is necessary for artificial neural networks to\nreproduce the behavioural patterns associated with SD in the absence of\n\\textit{output} non-linearities. Our results support a theory of SD progression\nthat results from continuous relearning of lost information. Future research\nshould revisit the role of relearning as a contributing factor to cognitive\ndiseases.\n", "date": "2025-03-05", "categories": []}
{"id": "2501.02950", "title": "Key-value memory in the brain", "abstract": "  Classical models of memory in psychology and neuroscience rely on\nsimilarity-based retrieval of stored patterns, where similarity is a function\nof retrieval cues and the stored patterns. While parsimonious, these models do\nnot allow distinct representations for storage and retrieval, despite their\ndistinct computational demands. Key-value memory systems, in contrast,\ndistinguish representations used for storage (values) and those used for\nretrieval (keys). This allows key-value memory systems to optimize\nsimultaneously for fidelity in storage and discriminability in retrieval. We\nreview the computational foundations of key-value memory, its role in modern\nmachine learning systems, related ideas from psychology and neuroscience,\napplications to a number of empirical puzzles, and possible biological\nimplementations.\n", "date": "2025-01-06", "categories": []}
{"id": "2503.01867", "title": "Neural Manifolds and Cognitive Consistency: A New Approach to Memory\n  Consolidation in Artificial Systems", "abstract": "  We introduce a novel mathematical framework that unifies neural population\ndynamics, hippocampal sharp wave-ripple (SpWR) generation, and cognitive\nconsistency constraints inspired by Heider's theory. Our model leverages\nlow-dimensional manifold representations to capture structured neural drift and\nincorporates a balance energy function to enforce coherent synaptic\ninteractions, effectively simulating the memory consolidation processes\nobserved in biological systems. Simulation results demonstrate that our\napproach not only reproduces key features of SpWR events but also enhances\nnetwork interpretability. This work paves the way for scalable neuromorphic\narchitectures that bridge neuroscience and artificial intelligence, offering\nmore robust and adaptive learning mechanisms for future intelligent systems.\n", "date": "2025-02-25", "categories": []}
{"id": "2503.02618", "title": "ZAPBench: A Benchmark for Whole-Brain Activity Prediction in Zebrafish", "abstract": "  Data-driven benchmarks have led to significant progress in key scientific\nmodeling domains including weather and structural biology. Here, we introduce\nthe Zebrafish Activity Prediction Benchmark (ZAPBench) to measure progress on\nthe problem of predicting cellular-resolution neural activity throughout an\nentire vertebrate brain. The benchmark is based on a novel dataset containing\n4d light-sheet microscopy recordings of over 70,000 neurons in a larval\nzebrafish brain, along with motion stabilized and voxel-level cell\nsegmentations of these data that facilitate development of a variety of\nforecasting methods. Initial results from a selection of time series and\nvolumetric video modeling approaches achieve better performance than naive\nbaseline methods, but also show room for further improvement. The specific\nbrain used in the activity recording is also undergoing synaptic-level\nanatomical mapping, which will enable future integration of detailed structural\ninformation into forecasting methods.\n", "date": "2025-03-04", "categories": []}
{"id": "2410.02094", "title": "Tracking objects that change in appearance with phase synchrony", "abstract": "  Objects we encounter often change appearance as we interact with them.\nChanges in illumination (shadows), object pose, or the movement of non-rigid\nobjects can drastically alter available image features. How do biological\nvisual systems track objects as they change? One plausible mechanism involves\nattentional mechanisms for reasoning about the locations of objects\nindependently of their appearances -- a capability that prominent neuroscience\ntheories have associated with computing through neural synchrony. Here, we\ndescribe a novel deep learning circuit that can learn to precisely control\nattention to features separately from their location in the world through\nneural synchrony: the complex-valued recurrent neural network (CV-RNN). Next,\nwe compare object tracking in humans, the CV-RNN, and other deep neural\nnetworks (DNNs), using FeatureTracker: a large-scale challenge that asks\nobservers to track objects as their locations and appearances change in\nprecisely controlled ways. While humans effortlessly solved FeatureTracker,\nstate-of-the-art DNNs did not. In contrast, our CV-RNN behaved similarly to\nhumans on the challenge, providing a computational proof-of-concept for the\nrole of phase synchronization as a neural substrate for tracking\nappearance-morphing objects as they move about.\n", "date": "2024-10-02", "categories": []}
{"id": "2503.00073", "title": "Forecasting Whole-Brain Neuronal Activity from Volumetric Video", "abstract": "  Large-scale neuronal activity recordings with fluorescent calcium indicators\nare increasingly common, yielding high-resolution 2D or 3D videos. Traditional\nanalysis pipelines reduce this data to 1D traces by segmenting regions of\ninterest, leading to inevitable information loss. Inspired by the success of\ndeep learning on minimally processed data in other domains, we investigate the\npotential of forecasting neuronal activity directly from volumetric videos. To\ncapture long-range dependencies in high-resolution volumetric whole-brain\nrecordings, we design a model with large receptive fields, which allow it to\nintegrate information from distant regions within the brain. We explore the\neffects of pre-training and perform extensive model selection, analyzing\nspatio-temporal trade-offs for generating accurate forecasts. Our model\noutperforms trace-based forecasting approaches on ZAPBench, a recently proposed\nbenchmark on whole-brain activity prediction in zebrafish, demonstrating the\nadvantages of preserving the spatial structure of neuronal activity.\n", "date": "2025-02-27", "categories": []}
{"id": "2301.11955", "title": "Adaptive whitening in neural populations with gain-modulating\n  interneurons", "abstract": "  Statistical whitening transformations play a fundamental role in many\ncomputational systems, and may also play an important role in biological\nsensory systems. Existing neural circuit models of adaptive whitening operate\nby modifying synaptic interactions; however, such modifications would seem both\ntoo slow and insufficiently reversible. Motivated by the extensive neuroscience\nliterature on gain modulation, we propose an alternative model that adaptively\nwhitens its responses by modulating the gains of individual neurons. Starting\nfrom a novel whitening objective, we derive an online algorithm that whitens\nits outputs by adjusting the marginal variances of an overcomplete set of\nprojections. We map the algorithm onto a recurrent neural network with fixed\nsynaptic weights and gain-modulating interneurons. We demonstrate numerically\nthat sign-constraining the gains improves robustness of the network to\nill-conditioned inputs, and a generalization of the circuit achieves a form of\nlocal whitening in convolutional populations, such as those found throughout\nthe visual or auditory systems.\n", "date": "2023-01-27", "categories": []}
{"id": "2406.07843", "title": "Self-Attention-Based Contextual Modulation Improves Neural System\n  Identification", "abstract": "  Convolutional neural networks (CNNs) have been shown to be state-of-the-art\nmodels for visual cortical neurons. Cortical neurons in the primary visual\ncortex are sensitive to contextual information mediated by extensive horizontal\nand feedback connections. Standard CNNs integrate global contextual information\nto model contextual modulation via two mechanisms: successive convolutions and\na fully connected readout layer. In this paper, we find that self-attention\n(SA), an implementation of non-local network mechanisms, can improve neural\nresponse predictions over parameter-matched CNNs in two key metrics: tuning\ncurve correlation and peak tuning. We introduce peak tuning as a metric to\nevaluate a model's ability to capture a neuron's top feature preference. We\nfactorize networks to assess each context mechanism, revealing that information\nin the local receptive field is most important for modeling overall tuning, but\nsurround information is critically necessary for characterizing the tuning\npeak. We find that self-attention can replace posterior spatial-integration\nconvolutions when learned incrementally, and is further enhanced in the\npresence of a fully connected readout layer, suggesting that the two context\nmechanisms are complementary. Finally, we find that decomposing receptive field\nlearning and contextual modulation learning in an incremental manner may be an\neffective and robust mechanism for learning surround-center interactions.\n", "date": "2024-06-12", "categories": []}
{"id": "2405.16865", "title": "On Conformal Isometry of Grid Cells: Learning Distance-Preserving\n  Position Embedding", "abstract": "  This paper investigates the conformal isometry hypothesis as a potential\nexplanation for the hexagonal periodic patterns in grid cell response maps. We\nposit that grid cell activities form a high-dimensional vector in neural space,\nencoding the agent's position in 2D physical space. As the agent moves, this\nvector rotates within a 2D manifold in the neural space, driven by a recurrent\nneural network. The conformal hypothesis proposes that this neural manifold is\na conformal isometric embedding of 2D physical space, where local physical\ndistance is preserved by the embedding up to a scaling factor (or unit of\nmetric). Such distance-preserving position embedding is indispensable for path\nplanning in navigation, especially planning local straight path segments. We\nconduct numerical experiments to show that this hypothesis leads to the\nhexagonal grid firing patterns by learning maximally distance-preserving\nposition embedding, agnostic to the choice of the recurrent neural network.\nFurthermore, we present a theoretical explanation of why hexagon periodic\npatterns emerge by minimizing our loss function by showing that hexagon flat\ntorus is maximally distance preserving.\n", "date": "2024-05-27", "categories": []}
{"id": "2410.13866", "title": "Associative memory and dead neurons", "abstract": "  In \"Large Associative Memory Problem in Neurobiology and Machine Learning,\"\nDmitry Krotov and John Hopfield introduced a general technique for the\nsystematic construction of neural ordinary differential equations with\nnon-increasing energy or Lyapunov function. We study this energy function and\nidentify that it is vulnerable to the problem of dead neurons. Each point in\nthe state space where the neuron dies is contained in a non-compact region with\nconstant energy. In these flat regions, energy function alone does not\ncompletely determine all degrees of freedom and, as a consequence, can not be\nused to analyze stability or find steady states or basins of attraction. We\nperform a direct analysis of the dynamical system and show how to resolve\nproblems caused by flat directions corresponding to dead neurons: (i) all\ninformation about the state vector at a fixed point can be extracted from the\nenergy and Hessian matrix (of Lagrange function), (ii) it is enough to analyze\nstability in the range of Hessian matrix, (iii) if steady state touching flat\nregion is stable the whole flat region is the basin of attraction. The analysis\nof the Hessian matrix can be complicated for realistic architectures, so we\nshow that for a slightly altered dynamical system (with the same structure of\nsteady states), one can derive a diverse family of Lyapunov functions that do\nnot have flat regions corresponding to dead neurons. In addition, these energy\nfunctions allow one to use Lagrange functions with Hessian matrices that are\nnot necessarily positive definite and even consider architectures with\nnon-symmetric feedforward and feedback connections.\n", "date": "2024-10-02", "categories": []}
{"id": "2502.16238", "title": "Brain-Model Evaluations Need the NeuroAI Turing Test", "abstract": "  What makes an artificial system a good model of intelligence? The classical\ntest proposed by Alan Turing focuses on behavior, requiring that an artificial\nagent's behavior be indistinguishable from that of a human. While behavioral\nsimilarity provides a strong starting point, two systems with very different\ninternal representations can produce the same outputs. Thus, in modeling\nbiological intelligence, the field of NeuroAI often aims to go beyond\nbehavioral similarity and achieve representational convergence between a\nmodel's activations and the measured activity of a biological system. This\nposition paper argues that the standard definition of the Turing Test is\nincomplete for NeuroAI, and proposes a stronger framework called the ``NeuroAI\nTuring Test'', a benchmark that extends beyond behavior alone and\n\\emph{additionally} requires models to produce internal neural representations\nthat are empirically indistinguishable from those of a brain up to measured\nindividual variability, i.e. the differences between a computational model and\nthe brain is no more than the difference between one brain and another brain.\nWhile the brain is not necessarily the ceiling of intelligence, it remains the\nonly universally agreed-upon example, making it a natural reference point for\nevaluating computational models. By proposing this framework, we aim to shift\nthe discourse from loosely defined notions of brain inspiration to a systematic\nand testable standard centered on both behavior and internal representations,\nproviding a clear benchmark for neuroscientific modeling and AI development.\n", "date": "2025-02-22", "categories": []}
{"id": "2502.18725", "title": "Talking to the brain: Using Large Language Models as Proxies to Model\n  Brain Semantic Representation", "abstract": "  Traditional psychological experiments utilizing naturalistic stimuli face\nchallenges in manual annotation and ecological validity. To address this, we\nintroduce a novel paradigm leveraging multimodal large language models (LLMs)\nas proxies to extract rich semantic information from naturalistic images\nthrough a Visual Question Answering (VQA) strategy for analyzing human visual\nsemantic representation. LLM-derived representations successfully predict\nestablished neural activity patterns measured by fMRI (e.g., faces, buildings),\nvalidating its feasibility and revealing hierarchical semantic organization\nacross cortical regions. A brain semantic network constructed from LLM-derived\nrepresentations identifies meaningful clusters reflecting functional and\ncontextual associations. This innovative methodology offers a powerful solution\nfor investigating brain semantic organization with naturalistic stimuli,\novercoming limitations of traditional annotation methods and paving the way for\nmore ecologically valid explorations of human cognition.\n", "date": "2025-02-26", "categories": []}
{"id": "2502.14731", "title": "Beyond Performance Scores: Directed Functional Connectivity as a\n  Brain-Based Biomarker for Motor Skill Learning and Retention", "abstract": "  Motor skill acquisition in fields like surgery, robotics, and sports involves\nlearning complex task sequences through extensive training. Traditional\nperformance metrics, like execution time and error rates, offer limited insight\nas they fail to capture the neural mechanisms underlying skill learning and\nretention. This study introduces directed functional connectivity (dFC),\nderived from electroencephalography (EEG), as a novel brain-based biomarker for\nassessing motor skill learning and retention. For the first time, dFC is\napplied as a biomarker to map the stages of the Fitts and Posner motor learning\nmodel, offering new insights into the neural mechanisms underlying skill\nacquisition and retention. Unlike traditional measures, it captures both the\nstrength and direction of neural information flow, providing a comprehensive\nunderstanding of neural adaptations across different learning stages. The\nanalysis demonstrates that dFC can effectively identify and track the\nprogression through various stages of the Fitts and Posner model. Furthermore,\nits stability over a six-week washout period highlights its utility in\nmonitoring long-term retention. No significant changes in dFC were observed in\na control group, confirming that the observed neural adaptations were specific\nto training and not due to external factors. By offering a granular view of the\nlearning process at the group and individual levels, dFC facilitates the\ndevelopment of personalized, targeted training protocols aimed at enhancing\noutcomes in fields where precision and long-term retention are critical, such\nas surgical education. These findings underscore the value of dFC as a robust\nbiomarker that complements traditional performance metrics, providing a deeper\nunderstanding of motor skill learning and retention.\n", "date": "2025-02-20", "categories": []}
{"id": "2502.10425", "title": "Neuron Platonic Intrinsic Representation From Dynamics Using Contrastive\n  Learning", "abstract": "  The Platonic Representation Hypothesis suggests a universal,\nmodality-independent reality representation behind different data modalities.\nInspired by this, we view each neuron as a system and detect its multi-segment\nactivity data under various peripheral conditions. We assume there's a\ntime-invariant representation for the same neuron, reflecting its intrinsic\nproperties like molecular profiles, location, and morphology. The goal of\nobtaining these intrinsic neuronal representations has two criteria: (I)\nsegments from the same neuron should have more similar representations than\nthose from different neurons; (II) the representations must generalize well to\nout-of-domain data. To meet these, we propose the NeurPIR (Neuron Platonic\nIntrinsic Representation) framework. It uses contrastive learning, with\nsegments from the same neuron as positive pairs and those from different\nneurons as negative pairs. In implementation, we use VICReg, which focuses on\npositive pairs and separates dissimilar samples via regularization. We tested\nour method on Izhikevich model-simulated neuronal population dynamics data. The\nresults accurately identified neuron types based on preset hyperparameters. We\nalso applied it to two real-world neuron dynamics datasets with neuron type\nannotations from spatial transcriptomics and neuron locations. Our model's\nlearned representations accurately predicted neuron types and locations and\nwere robust on out-of-domain data (from unseen animals). This shows the\npotential of our approach for understanding neuronal systems and future\nneuroscience research.\n", "date": "2025-02-06", "categories": []}
{"id": "2502.12536", "title": "An Algorithm Board in Neural Decoding", "abstract": "  Understanding the mechanisms of neural encoding and decoding has always been\na highly interesting research topic in fields such as neuroscience and\ncognitive intelligence. In prior studies, some researchers identified a\nsymmetry in neural data decoded by unsupervised methods in motor scenarios and\nconstructed a cognitive learning system based on this pattern (i.e., symmetry).\nNevertheless, the distribution state of the data flow that significantly\ninfluences neural decoding positions still remains a mystery within the system,\nwhich further restricts the enhancement of the system's interpretability. Based\non this, this paper mainly explores changes in the distribution state within\nthe system from the machine learning and mathematical statistics perspectives.\nIn the experiment, we assessed the correctness of this symmetry using various\ntools and indicators commonly utilized in mathematics and statistics. According\nto the experimental results, the normal distribution (or Gaussian distribution)\nplays a crucial role in the decoding of prediction positions within the system.\nEventually, an algorithm board similar to the Galton board was built to serve\nas the mathematical foundation of the discovered symmetry.\n", "date": "2025-02-18", "categories": []}
{"id": "2412.09115", "title": "Vision CNNs trained to estimate spatial latents learned similar\n  ventral-stream-aligned representations", "abstract": "  Studies of the functional role of the primate ventral visual stream have\ntraditionally focused on object categorization, often ignoring -- despite much\nprior evidence -- its role in estimating \"spatial\" latents such as object\nposition and pose. Most leading ventral stream models are derived by optimizing\nnetworks for object categorization, which seems to imply that the ventral\nstream is also derived under such an objective. Here, we explore an alternative\nhypothesis: Might the ventral stream be optimized for estimating spatial\nlatents? And a closely related question: How different -- if at all -- are\nrepresentations learned from spatial latent estimation compared to\ncategorization? To ask these questions, we leveraged synthetic image datasets\ngenerated by a 3D graphic engine and trained convolutional neural networks\n(CNNs) to estimate different combinations of spatial and category latents. We\nfound that models trained to estimate just a few spatial latents achieve neural\nalignment scores comparable to those trained on hundreds of categories, and the\nspatial latent performance of models strongly correlates with their neural\nalignment. Spatial latent and category-trained models have very similar -- but\nnot identical -- internal representations, especially in their early and middle\nlayers. We provide evidence that this convergence is partly driven by\nnon-target latent variability in the training data, which facilitates the\nimplicit learning of representations of those non-target latents. Taken\ntogether, these results suggest that many training objectives, such as spatial\nlatents, can lead to similar models aligned neurally with the ventral stream.\nThus, one should not assume that the ventral stream is optimized for object\ncategorization only. As a field, we need to continue to sharpen our measures of\ncomparing models to brains to better understand the functional roles of the\nventral stream.\n", "date": "2024-12-12", "categories": []}
{"id": "2502.10955", "title": "A recurrent vision transformer shows signatures of primate visual\n  attention", "abstract": "  Attention is fundamental to both biological and artificial intelligence, yet\nresearch on animal attention and AI self attention remains largely\ndisconnected. We propose a Recurrent Vision Transformer (Recurrent ViT) that\nintegrates self-attention with recurrent memory, allowing both current inputs\nand stored information to guide attention allocation. Trained solely via sparse\nreward feedback on a spatially cued orientation change detection task, a\nparadigm used in primate studies, our model exhibits primate like signatures of\nattention, including improved accuracy and faster responses for cued stimuli\nthat scale with cue validity. Analysis of self-attention maps reveals dynamic\nspatial prioritization with reactivation prior to expected changes, and\ntargeted perturbations produce performance shifts similar to those observed in\nprimate frontal eye fields and superior colliculus. These findings demonstrate\nthat incorporating recurrent feedback into self attention can capture key\naspects of primate visual attention.\n", "date": "2025-02-16", "categories": []}
{"id": "2311.03131", "title": "Decoding Neuronal Networks: A Reservoir Computing Approach for\n  Predicting Connectivity and Functionality", "abstract": "  In this study, we address the challenge of analyzing electrophysiological\nmeasurements in neuronal networks. Our computational model, based on the\nReservoir Computing Network (RCN) architecture, deciphers spatio-temporal data\nobtained from electrophysiological measurements of neuronal cultures. By\nreconstructing the network structure on a macroscopic scale, we reveal the\nconnectivity between neuronal units. Notably, our model outperforms common\nmethods like Cross-Correlation and Transfer-Entropy in predicting the network's\nconnectivity map. Furthermore, we experimentally validate its ability to\nforecast network responses to specific inputs, including localized optogenetic\nstimuli.\n", "date": "2023-11-06", "categories": []}
{"id": "2502.07800", "title": "neuro2voc: Decoding Vocalizations from Neural Activity", "abstract": "  Accurate decoding of neural spike trains and relating them to motor output is\na challenging task due to the inherent sparsity and length in neural spikes and\nthe complexity of brain circuits. This master project investigates experimental\nmethods for decoding zebra finch motor outputs (in both discrete syllables and\ncontinuous spectrograms), from invasive neural recordings obtained from\nNeuropixels.\n  There are three major achievements: (1) XGBoost with SHAP analysis trained on\nspike rates revealed neuronal interaction patterns crucial for syllable\nclassification. (2) Novel method (tokenizing neural data with GPT2) and\narchitecture (Mamba2) demonstrated potential for decoding of syllables using\nspikes. (3) A combined contrastive learning-VAE framework successfully\ngenerated spectrograms from binned neural data.\n  This work establishes a promising foundation for neural decoding of complex\nmotor outputs and offers several novel methodological approaches for processing\nsparse neural data.\n", "date": "2025-02-02", "categories": []}
{"id": "2502.04892", "title": "A Foundational Brain Dynamics Model via Stochastic Optimal Control", "abstract": "  We introduce a foundational model for brain dynamics that utilizes stochastic\noptimal control (SOC) and amortized inference. Our method features a\ncontinuous-discrete state space model (SSM) that can robustly handle the\nintricate and noisy nature of fMRI signals. To address computational\nlimitations, we implement an approximation strategy grounded in the SOC\nframework. Additionally, we present a simulation-free latent dynamics approach\nthat employs locally linear approximations, facilitating efficient and scalable\ninference. For effective representation learning, we derive an Evidence Lower\nBound (ELBO) from the SOC formulation, which integrates smoothly with recent\nadvancements in self-supervised learning (SSL), thereby promoting robust and\ntransferable representations. Pre-trained on extensive datasets such as the\nUKB, our model attains state-of-the-art results across a variety of downstream\ntasks, including demographic prediction, trait analysis, disease diagnosis, and\nprognosis. Moreover, evaluating on external datasets such as HCP-A, ABIDE, and\nADHD200 further validates its superior abilities and resilience across\ndifferent demographic and clinical distributions. Our foundational model\nprovides a scalable and efficient approach for deciphering brain dynamics,\nopening up numerous applications in neuroscience.\n", "date": "2025-02-07", "categories": []}
{"id": "2410.13872", "title": "BLEND: Behavior-guided Neural Population Dynamics Modeling via\n  Privileged Knowledge Distillation", "abstract": "  Modeling the nonlinear dynamics of neuronal populations represents a key\npursuit in computational neuroscience. Recent research has increasingly focused\non jointly modeling neural activity and behavior to unravel their\ninterconnections. Despite significant efforts, these approaches often\nnecessitate either intricate model designs or oversimplified assumptions. Given\nthe frequent absence of perfectly paired neural-behavioral datasets in\nreal-world scenarios when deploying these models, a critical yet understudied\nresearch question emerges: how to develop a model that performs well using only\nneural activity as input at inference, while benefiting from the insights\ngained from behavioral signals during training?\n  To this end, we propose BLEND, the behavior-guided neural population dynamics\nmodeling framework via privileged knowledge distillation. By considering\nbehavior as privileged information, we train a teacher model that takes both\nbehavior observations (privileged features) and neural activities (regular\nfeatures) as inputs. A student model is then distilled using only neural\nactivity. Unlike existing methods, our framework is model-agnostic and avoids\nmaking strong assumptions about the relationship between behavior and neural\nactivity. This allows BLEND to enhance existing neural dynamics modeling\narchitectures without developing specialized models from scratch. Extensive\nexperiments across neural population activity modeling and transcriptomic\nneuron identity prediction tasks demonstrate strong capabilities of BLEND,\nreporting over 50% improvement in behavioral decoding and over 15% improvement\nin transcriptomic neuron identity prediction after behavior-guided\ndistillation. Furthermore, we empirically explore various behavior-guided\ndistillation strategies within the BLEND framework and present a comprehensive\nanalysis of effectiveness and implications for model performance.\n", "date": "2024-10-02", "categories": []}
{"id": "2412.17404", "title": "BrainMAP: Learning Multiple Activation Pathways in Brain Networks", "abstract": "  Functional Magnetic Resonance Image (fMRI) is commonly employed to study\nhuman brain activity, since it offers insight into the relationship between\nfunctional fluctuations and human behavior. To enhance analysis and\ncomprehension of brain activity, Graph Neural Networks (GNNs) have been widely\napplied to the analysis of functional connectivities (FC) derived from fMRI\ndata, due to their ability to capture the synergistic interactions among brain\nregions. However, in the human brain, performing complex tasks typically\ninvolves the activation of certain pathways, which could be represented as\npaths across graphs. As such, conventional GNNs struggle to learn from these\npathways due to the long-range dependencies of multiple pathways. To address\nthese challenges, we introduce a novel framework BrainMAP to learn Multiple\nActivation Pathways in Brain networks. BrainMAP leverages sequential models to\nidentify long-range correlations among sequentialized brain regions and\nincorporates an aggregation module based on Mixture of Experts (MoE) to learn\nfrom multiple pathways. Our comprehensive experiments highlight BrainMAP's\nsuperior performance. Furthermore, our framework enables explanatory analyses\nof crucial brain regions involved in tasks. Our code is provided at\nhttps://github.com/LzyFischer/Graph-Mamba.\n", "date": "2024-12-23", "categories": []}
{"id": "2501.06278", "title": "Aligning Brain Activity with Advanced Transformer Models: Exploring the\n  Role of Punctuation in Semantic Processing", "abstract": "  This research examines the congruence between neural activity and advanced\ntransformer models, emphasizing the semantic significance of punctuation in\ntext understanding. Utilizing an innovative approach originally proposed by\nToneva and Wehbe, we evaluate four advanced transformer models RoBERTa,\nDistiliBERT, ALBERT, and ELECTRA against neural activity data. Our findings\nindicate that RoBERTa exhibits the closest alignment with neural activity,\nsurpassing BERT in accuracy. Furthermore, we investigate the impact of\npunctuation removal on model performance and neural alignment, revealing that\nBERT's accuracy enhances in the absence of punctuation. This study contributes\nto the comprehension of how neural networks represent language and the\ninfluence of punctuation on semantic processing within the human brain.\n", "date": "2025-01-10", "categories": []}
{"id": "2501.06326", "title": "On Creating A Brain-To-Text Decoder", "abstract": "  Brain decoding has emerged as a rapidly advancing and extensively utilized\ntechnique within neuroscience. This paper centers on the application of raw\nelectroencephalogram (EEG) signals for decoding human brain activity, offering\na more expedited and efficient methodology for enhancing our understanding of\nthe human brain. The investigation specifically scrutinizes the efficacy of\nbrain-computer interfaces (BCI) in deciphering neural signals associated with\nspeech production, with particular emphasis on the impact of vocabulary size,\nelectrode density, and training data on the framework's performance. The study\nreveals the competitive word error rates (WERs) achievable on the Librispeech\nbenchmark through pre-training on unlabelled data for speech processing.\nFurthermore, the study evaluates the efficacy of voice recognition under\nconfigurations with limited labeled data, surpassing previous state-of-the-art\ntechniques while utilizing significantly fewer labels. Additionally, the\nresearch provides a comprehensive analysis of error patterns in voice\nrecognition and the influence of model size and unlabelled training data. It\nunderscores the significance of factors such as vocabulary size and electrode\ndensity in enhancing BCI performance, advocating for an increase in\nmicroelectrodes and refinement of language models.\n", "date": "2025-01-10", "categories": []}
{"id": "2501.07440", "title": "Attention when you need", "abstract": "  Being attentive to task-relevant features can improve task performance, but\npaying attention comes with its own metabolic cost. Therefore, strategic\nallocation of attention is crucial in performing the task efficiently. This\nwork aims to understand this strategy. Recently, de Gee et al. conducted\nexperiments involving mice performing an auditory sustained attention-value\ntask. This task required the mice to exert attention to identify whether a\nhigh-order acoustic feature was present amid the noise. By varying the trial\nduration and reward magnitude, the task allows us to investigate how an agent\nshould strategically deploy their attention to maximize their benefits and\nminimize their costs. In our work, we develop a reinforcement learning-based\nnormative model of the mice to understand how it balances attention cost\nagainst its benefits. The model is such that at each moment the mice can choose\nbetween two levels of attention and decide when to take costly actions that\ncould obtain rewards. Our model suggests that efficient use of attentional\nresources involves alternating blocks of high attention with blocks of low\nattention. In the extreme case where the agent disregards sensory input during\nlow attention states, we see that high attention is used rhythmically. Our\nmodel provides evidence about how one should deploy attention as a function of\ntask utility, signal statistics, and how attention affects sensory evidence.\n", "date": "2025-01-13", "categories": []}
{"id": "2501.15322", "title": "Scaling laws for decoding images from brain activity", "abstract": "  Generative AI has recently propelled the decoding of images from brain\nactivity. How do these approaches scale with the amount and type of neural\nrecordings? Here, we systematically compare image decoding from four types of\nnon-invasive devices: electroencephalography (EEG), magnetoencephalography\n(MEG), high-field functional Magnetic Resonance Imaging (3T fMRI) and\nultra-high field (7T) fMRI. For this, we evaluate decoding models on the\nlargest benchmark to date, encompassing 8 public datasets, 84 volunteers, 498\nhours of brain recording and 2.3 million brain responses to natural images.\nUnlike previous work, we focus on single-trial decoding performance to simulate\nreal-time settings. This systematic comparison reveals three main findings.\nFirst, the most precise neuroimaging devices tend to yield the best decoding\nperformances, when the size of the training sets are similar. However, the gain\nenabled by deep learning - in comparison to linear models - is obtained with\nthe noisiest devices. Second, we do not observe any plateau of decoding\nperformance as the amount of training data increases. Rather, decoding\nperformance scales log-linearly with the amount of brain recording. Third, this\nscaling law primarily depends on the amount of data per subject. However,\nlittle decoding gain is observed by increasing the number of subjects. Overall,\nthese findings delineate the path most suitable to scale the decoding of images\nfrom non-invasive brain recordings.\n", "date": "2025-01-25", "categories": []}
{"id": "2402.16996", "title": "Towards Decoding Brain Activity During Passive Listening of Speech", "abstract": "  The aim of the study is to investigate the complex mechanisms of speech\nperception and ultimately decode the electrical changes in the brain accruing\nwhile listening to speech. We attempt to decode heard speech from intracranial\nelectroencephalographic (iEEG) data using deep learning methods. The goal is to\naid the advancement of brain-computer interface (BCI) technology for speech\nsynthesis, and, hopefully, to provide an additional perspective on the\ncognitive processes of speech perception. This approach diverges from the\nconventional focus on speech production and instead chooses to investigate\nneural representations of perceived speech. This angle opened up a complex\nperspective, potentially allowing us to study more sophisticated neural\npatterns. Leveraging the power of deep learning models, the research aimed to\nestablish a connection between these intricate neural activities and the\ncorresponding speech sounds. Despite the approach not having achieved a\nbreakthrough yet, the research sheds light on the potential of decoding neural\nactivity during speech perception. Our current efforts can serve as a\nfoundation, and we are optimistic about the potential of expanding and\nimproving upon this work to move closer towards more advanced BCIs, better\nunderstanding of processes underlying perceived speech and its relation to\nspoken speech.\n", "date": "2024-02-26", "categories": []}
{"id": "2406.19087", "title": "Dimensions underlying the representational alignment of deep neural\n  networks with humans", "abstract": "  Determining the similarities and differences between humans and artificial\nintelligence (AI) is an important goal both in computational cognitive\nneuroscience and machine learning, promising a deeper understanding of human\ncognition and safer, more reliable AI systems. Much previous work comparing\nrepresentations in humans and AI has relied on global, scalar measures to\nquantify their alignment. However, without explicit hypotheses, these measures\nonly inform us about the degree of alignment, not the factors that determine\nit. To address this challenge, we propose a generic framework to compare human\nand AI representations, based on identifying latent representational dimensions\nunderlying the same behavior in both domains. Applying this framework to humans\nand a deep neural network (DNN) model of natural images revealed a\nlow-dimensional DNN embedding of both visual and semantic dimensions. In\ncontrast to humans, DNNs exhibited a clear dominance of visual over semantic\nproperties, indicating divergent strategies for representing images. While\nin-silico experiments showed seemingly consistent interpretability of DNN\ndimensions, a direct comparison between human and DNN representations revealed\nsubstantial differences in how they process images. By making representations\ndirectly comparable, our results reveal important challenges for\nrepresentational alignment and offer a means for improving their comparability.\n", "date": "2024-06-27", "categories": []}
{"id": "2501.12615", "title": "GATE: Adaptive Learning with Working Memory by Information Gating in\n  Multi-lamellar Hippocampal Formation", "abstract": "  Hippocampal formation (HF) can rapidly adapt to varied environments and build\nflexible working memory (WM). To mirror the HF's mechanism on generalization\nand WM, we propose a model named Generalization and Associative Temporary\nEncoding (GATE), which deploys a 3-D multi-lamellar dorsoventral (DV)\narchitecture, and learns to build up internally representation from externally\ndriven information layer-wisely. In each lamella, regions of HF:\nEC3-CA1-EC5-EC3 forms a re-entrant loop that discriminately maintains\ninformation by EC3 persistent activity, and selectively readouts the retained\ninformation by CA1 neurons. CA3 and EC5 further provides gating function that\ncontrols these processes. After learning complex WM tasks, GATE forms neuron\nrepresentations that align with experimental records, including splitter, lap,\nevidence, trace, delay-active cells, as well as conventional place cells.\nCrucially, DV architecture in GATE also captures information, range from\ndetailed to abstract, which enables a rapid generalization ability when cue,\nenvironment or task changes, with learned representations inherited. GATE\npromises a viable framework for understanding the HF's flexible memory\nmechanisms and for progressively developing brain-inspired intelligent systems.\n", "date": "2025-01-22", "categories": []}
{"id": "2501.08334", "title": "High-throughput digital twin framework for predicting neurite\n  deterioration using MetaFormer attention", "abstract": "  Neurodevelopmental disorders (NDDs) cover a variety of conditions, including\nautism spectrum disorder, attention-deficit/hyperactivity disorder, and\nepilepsy, which impair the central and peripheral nervous systems. Their high\ncomorbidity and complex etiologies present significant challenges for accurate\ndiagnosis and effective treatments. Conventional clinical and experimental\nstudies are time-intensive, burdening research progress considerably. This\npaper introduces a high-throughput digital twin framework for modeling neurite\ndeteriorations associated with NDDs, integrating synthetic data generation,\nexperimental images, and machine learning (ML) models. The synthetic data\ngenerator utilizes an isogeometric analysis (IGA)-based phase field model to\ncapture diverse neurite deterioration patterns such as neurite retraction,\natrophy, and fragmentation while mitigating the limitations of scarce\nexperimental data. The ML model utilizes MetaFormer-based gated spatiotemporal\nattention architecture with deep temporal layers and provides fast predictions.\nThe framework effectively captures long-range temporal dependencies and\nintricate morphological transformations with average errors of 1.9641% and\n6.0339% for synthetic and experimental neurite deterioration, respectively.\nSeamlessly integrating simulations, experiments, and ML, the digital twin\nframework can guide researchers to make informed experimental decisions by\npredicting potential experimental outcomes, significantly reducing costs and\nsaving valuable time. It can also advance our understanding of neurite\ndeterioration and provide a scalable solution for exploring complex\nneurological mechanisms, contributing to the development of targeted\ntreatments.\n", "date": "2024-12-18", "categories": []}
{"id": "2311.09308", "title": "Divergences between Language Models and Human Brains", "abstract": "  Do machines and humans process language in similar ways? Recent research has\nhinted at the affirmative, showing that human neural activity can be\neffectively predicted using the internal representations of language models\n(LMs). Although such results are thought to reflect shared computational\nprinciples between LMs and human brains, there are also clear differences in\nhow LMs and humans represent and use language. In this work, we systematically\nexplore the divergences between human and machine language processing by\nexamining the differences between LM representations and human brain responses\nto language as measured by Magnetoencephalography (MEG) across two datasets in\nwhich subjects read and listened to narrative stories. Using an LLM-based\ndata-driven approach, we identify two domains that LMs do not capture well:\nsocial/emotional intelligence and physical commonsense. We validate these\nfindings with human behavioral experiments and hypothesize that the gap is due\nto insufficient representations of social/emotional and physical knowledge in\nLMs. Our results show that fine-tuning LMs on these domains can improve their\nalignment with human brain responses.\n", "date": "2023-11-15", "categories": []}
{"id": "2501.06192", "title": "A Computational Model of Learning and Memory Using Structurally Dynamic\n  Cellular Automata", "abstract": "  In the fields of computation and neuroscience, much is still unknown about\nthe underlying computations that enable key cognitive functions including\nlearning, memory, abstraction and behavior. This paper proposes a mathematical\nand computational model of learning and memory based on a small set of\nbio-plausible functions that include coincidence detection, signal modulation,\nand reward/penalty mechanisms. Our theoretical approach proposes that these\nbasic functions are sufficient to establish and modulate an information space\nover which computation can be carried out, generating signal gradients usable\nfor inference and behavior. The computational method used to test this is a\nstructurally dynamic cellular automaton with continuous-valued cell states and\na series of recursive steps propagating over an undirected graph with the\nmemory function embedded entirely in the creation and modulation of graph\nedges. The experimental results show: that the toy model can make near-optimal\nchoices to re-discover a reward state after a single training run; that it can\navoid complex penalty configurations; that signal modulation and network\nplasticity can generate exploratory behaviors in sparse reward environments;\nthat the model generates context-dependent memory representations; and that it\nexhibits high computational efficiency because of its minimal, single-pass\ntraining requirements combined with flexible and contextual memory\nrepresentation.\n", "date": "2024-12-20", "categories": []}
{"id": "2410.08224", "title": "A Survey of Spatio-Temporal EEG data Analysis: from Models to\n  Applications", "abstract": "  In recent years, the field of electroencephalography (EEG) analysis has\nwitnessed remarkable advancements, driven by the integration of machine\nlearning and artificial intelligence. This survey aims to encapsulate the\nlatest developments, focusing on emerging methods and technologies that are\npoised to transform our comprehension and interpretation of brain activity. We\ndelve into self-supervised learning methods that enable the robust\nrepresentation of brain signals, which are fundamental for a variety of\ndownstream applications. We also explore emerging discriminative methods,\nincluding graph neural networks (GNN), foundation models, and large language\nmodels (LLMs)-based approaches. Furthermore, we examine generative technologies\nthat harness EEG data to produce images or text, offering novel perspectives on\nbrain activity visualization and interpretation. The survey provides an\nextensive overview of these cutting-edge techniques, their current\napplications, and the profound implications they hold for future research and\nclinical practice. The relevant literature and open-source materials have been\ncompiled and are consistently being refreshed at\n\\url{https://github.com/wpf535236337/LLMs4TS}\n", "date": "2024-09-26", "categories": []}
{"id": "2501.03246", "title": "Bridging Auditory Perception and Language Comprehension through\n  MEG-Driven Encoding Models", "abstract": "  Understanding the neural mechanisms behind auditory and linguistic processing\nis key to advancing cognitive neuroscience. In this study, we use\nMagnetoencephalography (MEG) data to analyze brain responses to spoken language\nstimuli. We develop two distinct encoding models: an audio-to-MEG encoder,\nwhich uses time-frequency decompositions (TFD) and wav2vec2 latent space\nrepresentations, and a text-to-MEG encoder, which leverages CLIP and GPT-2\nembeddings. Both models successfully predict neural activity, demonstrating\nsignificant correlations between estimated and observed MEG signals. However,\nthe text-to-MEG model outperforms the audio-based model, achieving higher\nPearson Correlation (PC) score. Spatially, we identify that auditory-based\nembeddings (TFD and wav2vec2) predominantly activate lateral temporal regions,\nwhich are responsible for primary auditory processing and the integration of\nauditory signals. In contrast, textual embeddings (CLIP and GPT-2) primarily\nengage the frontal cortex, particularly Broca's area, which is associated with\nhigher-order language processing, including semantic integration and language\nproduction, especially in the 8-30 Hz frequency range. The strong involvement\nof these regions suggests that auditory stimuli are processed through more\ndirect sensory pathways, while linguistic information is encoded via networks\nthat integrate meaning and cognitive control. Our results reveal distinct\nneural pathways for auditory and linguistic information processing, with higher\nencoding accuracy for text representations in the frontal regions. These\ninsights refine our understanding of the brain's functional architecture in\nprocessing auditory and textual information, offering quantitative advancements\nin the modelling of neural responses to complex language stimuli.\n", "date": "2024-12-22", "categories": []}
{"id": "2307.10246", "title": "Deep Neural Networks and Brain Alignment: Brain Encoding and Decoding\n  (Survey)", "abstract": "  Can artificial intelligence unlock the secrets of the human brain? How do the\ninner mechanisms of deep learning models relate to our neural circuits? Is it\npossible to enhance AI by tapping into the power of brain recordings? These\ncaptivating questions lie at the heart of an emerging field at the intersection\nof neuroscience and artificial intelligence. Our survey dives into this\nexciting domain, focusing on human brain recording studies and cutting-edge\ncognitive neuroscience datasets that capture brain activity during natural\nlanguage processing, visual perception, and auditory experiences. We explore\ntwo fundamental approaches: encoding models, which attempt to generate brain\nactivity patterns from sensory inputs; and decoding models, which aim to\nreconstruct our thoughts and perceptions from neural signals. These techniques\nnot only promise breakthroughs in neurological diagnostics and brain-computer\ninterfaces but also offer a window into the very nature of cognition. In this\nsurvey, we first discuss popular representations of language, vision, and\nspeech stimuli, and present a summary of neuroscience datasets. We then review\nhow the recent advances in deep learning transformed this field, by\ninvestigating the popular deep learning based encoding and decoding\narchitectures, noting their benefits and limitations across different sensory\nmodalities. From text to images, speech to videos, we investigate how these\nmodels capture the brain's response to our complex, multimodal world. While our\nprimary focus is on human studies, we also highlight the crucial role of animal\nmodels in advancing our understanding of neural mechanisms. Throughout, we\nmention the ethical implications of these powerful technologies, addressing\nconcerns about privacy and cognitive liberty. We conclude with a summary and\ndiscussion of future trends in this rapidly evolving field.\n", "date": "2023-07-17", "categories": []}
{"id": "2407.07059", "title": "Differentiable Optimization of Similarity Scores Between Models and\n  Brains", "abstract": "  How do we know if two systems - biological or artificial - process\ninformation in a similar way? Similarity measures such as linear regression,\nCentered Kernel Alignment (CKA), Normalized Bures Similarity (NBS), and angular\nProcrustes distance, are often used to quantify this similarity. However, it is\ncurrently unclear what drives high similarity scores and even what constitutes\na \"good\" score. Here, we introduce a novel tool to investigate these questions\nby differentiating through similarity measures to directly maximize the score.\nSurprisingly, we find that high similarity scores do not guarantee encoding\ntask-relevant information in a manner consistent with neural data; and this is\nparticularly acute for CKA and even some variations of cross-validated and\nregularized linear regression. We find no consistent threshold for a good\nsimilarity score - it depends on both the measure and the dataset. In addition,\nsynthetic datasets optimized to maximize similarity scores initially learn the\nhighest variance principal component of the target dataset, but some methods\nlike angular Procrustes capture lower variance dimensions much earlier than\nmethods like CKA. To shed light on this, we mathematically derive the\nsensitivity of CKA, angular Procrustes, and NBS to the variance of principal\ncomponent dimensions, and explain the emphasis CKA places on high variance\ncomponents. Finally, by jointly optimizing multiple similarity measures, we\ncharacterize their allowable ranges and reveal that some similarity measures\nare more constraining than others. While current measures offer a seemingly\nstraightforward way to quantify the similarity between neural systems, our work\nunderscores the need for careful interpretation. We hope the tools we developed\nwill be used by practitioners to better understand current and future\nsimilarity measures.\n", "date": "2024-07-09", "categories": []}
{"id": "2412.19814", "title": "Predicting Human Brain States with Transformer", "abstract": "  The human brain is a complex and highly dynamic system, and our current\nknowledge of its functional mechanism is still very limited. Fortunately, with\nfunctional magnetic resonance imaging (fMRI), we can observe blood oxygen\nlevel-dependent (BOLD) changes, reflecting neural activity, to infer brain\nstates and dynamics. In this paper, we ask the question of whether the brain\nstates rep-resented by the regional brain fMRI can be predicted. Due to the\nsuccess of self-attention and the transformer architecture in sequential\nauto-regression problems (e.g., language modelling or music generation), we\nexplore the possi-bility of the use of transformers to predict human brain\nresting states based on the large-scale high-quality fMRI data from the human\nconnectome project (HCP). Current results have shown that our model can\naccurately predict the brain states up to 5.04s with the previous 21.6s.\nFurthermore, even though the prediction error accumulates for the prediction of\na longer time period, the gen-erated fMRI brain states reflect the architecture\nof functional connectome. These promising initial results demonstrate the\npossibility of developing gen-erative models for fMRI data using self-attention\nthat learns the functional or-ganization of the human brain. Our code is\navailable at: https://github.com/syf0122/brain_state_pred.\n", "date": "2024-12-11", "categories": []}
{"id": "2412.19999", "title": "Comprehensive Review of EEG-to-Output Research: Decoding Neural Signals\n  into Images, Videos, and Audio", "abstract": "  Electroencephalography (EEG) is an invaluable tool in neuroscience, offering\ninsights into brain activity with high temporal resolution. Recent advancements\nin machine learning and generative modeling have catalyzed the application of\nEEG in reconstructing perceptual experiences, including images, videos, and\naudio. This paper systematically reviews EEG-to-output research, focusing on\nstate-of-the-art generative methods, evaluation metrics, and data challenges.\nUsing PRISMA guidelines, we analyze 1800 studies and identify key trends,\nchallenges, and opportunities in the field. The findings emphasize the\npotential of advanced models such as Generative Adversarial Networks (GANs),\nVariational Autoencoders (VAEs), and Transformers, while highlighting the\npressing need for standardized datasets and cross-subject generalization. A\nroadmap for future research is proposed that aims to improve decoding accuracy\nand broadening real-world applications.\n", "date": "2024-12-28", "categories": []}
{"id": "2412.20090", "title": "From Worms to Mice: Homeostasis Maybe All You Need", "abstract": "  In this brief and speculative commentary, we explore ideas inspired by neural\nnetworks in machine learning, proposing that a simple neural XOR motif,\ninvolving both excitatory and inhibitory connections, may provide the basis for\na relevant mode of plasticity in neural circuits of living organisms, with\nhomeostasis as the sole guiding principle. This XOR motif simply signals the\ndiscrepancy between incoming signals and reference signals, thereby providing a\nbasis for a loss function in learning neural circuits, and at the same time\nregulating homeostasis by halting the propagation of these incoming signals.\nThe core motif uses a 4:1 ratio of excitatory to inhibitory neurons, and\nsupports broader neural patterns such as the well-known 'winner takes all'\n(WTA) mechanism. We examined the prevalence of the XOR motif in the published\nconnectomes of various organisms with increasing complexity, and found that it\nranges from tens (in C. elegans) to millions (in several Drosophila neuropils)\nand more than tens of millions (in mouse V1 visual cortex). If validated, our\nhypothesis identifies two of the three key components in analogy to machine\nlearning models: the architecture and the loss function. And we propose that a\nrelevant type of biological neural plasticity is simply driven by a basic\ncontrol or regulatory system, which has persisted and adapted despite the\nincreasing complexity of organisms throughout evolution.\n", "date": "2024-12-28", "categories": []}
{"id": "2412.21188", "title": "Sparse chaos in cortical circuits", "abstract": "  Nerve impulses, the currency of information flow in the brain, are generated\nby an instability of the neuronal membrane potential dynamics. Neuronal\ncircuits exhibit collective chaos that appears essential for learning, memory,\nsensory processing, and motor control. However, the factors controlling the\nnature and intensity of collective chaos in neuronal circuits are not well\nunderstood. Here we use computational ergodic theory to demonstrate that basic\nfeatures of nerve impulse generation profoundly affect collective chaos in\nneuronal circuits. Numerically exact calculations of Lyapunov spectra,\nKolmogorov-Sinai-entropy, and upper and lower bounds on attractor dimension\nshow that changes in nerve impulse generation in individual neurons moderately\nimpact information encoding rates but qualitatively transform phase space\nstructure. Specifically, we find a drastic reduction in the number of unstable\nmanifolds, Kolmogorov-Sinai entropy, and attractor dimension. Beyond a critical\npoint, marked by the simultaneous breakdown of the diffusion approximation, a\npeak in the largest Lyapunov exponent, and a localization transition of the\nleading covariant Lyapunov vector, networks exhibit sparse chaos: prolonged\nperiods of near stable dynamics interrupted by short bursts of intense chaos.\nAnalysis of large, more realistically structured networks supports the\ngenerality of these findings. In cortical circuits, biophysical properties\nappear tuned to this regime of sparse chaos. Our results reveal a close link\nbetween fundamental aspects of single-neuron biophysics and the collective\ndynamics of cortical circuits, suggesting that nerve impulse generation\nmechanisms are adapted to enhance circuit controllability and information flow.\n", "date": "2024-12-30", "categories": []}
{"id": "2408.12804", "title": "Universal dimensions of visual representation", "abstract": "  Do neural network models of vision learn brain-aligned representations\nbecause they share architectural constraints and task objectives with\nbiological vision or because they learn universal features of natural image\nprocessing? We characterized the universality of hundreds of thousands of\nrepresentational dimensions from visual neural networks with varied\nconstruction. We found that networks with varied architectures and task\nobjectives learn to represent natural images using a shared set of latent\ndimensions, despite appearing highly distinct at a surface level. Next, by\ncomparing these networks with human brain representations measured with fMRI,\nwe found that the most brain-aligned representations in neural networks are\nthose that are universal and independent of a network's specific\ncharacteristics. Remarkably, each network can be reduced to fewer than ten of\nits most universal dimensions with little impact on its representational\nsimilarity to the human brain. These results suggest that the underlying\nsimilarities between artificial and biological vision are primarily governed by\na core set of universal image representations that are convergently learned by\ndiverse systems.\n", "date": "2024-08-23", "categories": []}
{"id": "2411.03630", "title": "RTify: Aligning Deep Neural Networks with Human Behavioral Decisions", "abstract": "  Current neural network models of primate vision focus on replicating overall\nlevels of behavioral accuracy, often neglecting perceptual decisions' rich,\ndynamic nature. Here, we introduce a novel computational framework to model the\ndynamics of human behavioral choices by learning to align the temporal dynamics\nof a recurrent neural network (RNN) to human reaction times (RTs). We describe\nan approximation that allows us to constrain the number of time steps an RNN\ntakes to solve a task with human RTs. The approach is extensively evaluated\nagainst various psychophysics experiments. We also show that the approximation\ncan be used to optimize an \"ideal-observer\" RNN model to achieve an optimal\ntradeoff between speed and accuracy without human data. The resulting model is\nfound to account well for human RT data. Finally, we use the approximation to\ntrain a deep learning implementation of the popular Wong-Wang decision-making\nmodel. The model is integrated with a convolutional neural network (CNN) model\nof visual processing and evaluated using both artificial and natural image\nstimuli. Overall, we present a novel framework that helps align current vision\nmodels with human behavior, bringing us closer to an integrated model of human\nvision.\n", "date": "2024-11-06", "categories": []}
{"id": "2412.19487", "title": "UniBrain: A Unified Model for Cross-Subject Brain Decoding", "abstract": "  Brain decoding aims to reconstruct original stimuli from fMRI signals,\nproviding insights into interpreting mental content. Current approaches rely\nheavily on subject-specific models due to the complex brain processing\nmechanisms and the variations in fMRI signals across individuals. Therefore,\nthese methods greatly limit the generalization of models and fail to capture\ncross-subject commonalities. To address this, we present UniBrain, a unified\nbrain decoding model that requires no subject-specific parameters. Our approach\nincludes a group-based extractor to handle variable fMRI signal lengths, a\nmutual assistance embedder to capture cross-subject commonalities, and a\nbilevel feature alignment scheme for extracting subject-invariant features. We\nvalidate our UniBrain on the brain decoding benchmark, achieving comparable\nperformance to current state-of-the-art subject-specific models with extremely\nfewer parameters. We also propose a generalization benchmark to encourage the\ncommunity to emphasize cross-subject commonalities for more general brain\ndecoding. Our code is available at https://github.com/xiaoyao3302/UniBrain.\n", "date": "2024-12-27", "categories": []}
{"id": "2412.17824", "title": "Ensemble Machine Learning Model for Inner Speech Recognition: A\n  Subject-Specific Investigation", "abstract": "  Inner speech recognition has gained enormous interest in recent years due to\nits applications in rehabilitation, developing assistive technology, and\ncognitive assessment. However, since language and speech productions are a\ncomplex process, for which identifying speech components has remained a\nchallenging task. Different approaches were taken previously to reach this\ngoal, but new approaches remain to be explored. Also, a subject-oriented\nanalysis is necessary to understand the underlying brain dynamics during inner\nspeech production, which can bring novel methods to neurological research. A\npublicly available dataset, Thinking Out Loud Dataset, has been used to develop\na Machine Learning (ML)-based technique to classify inner speech using\n128-channel surface EEG signals. The dataset is collected on a Spanish cohort\nof ten subjects while uttering four words (Arriba, Abajo, Derecha, and\nIzquierda) by each participant. Statistical methods were employed to detect and\nremove motion artifacts from the Electroencephalography (EEG) signals. A large\nnumber (191 per channel) of time-, frequency- and time-frequency-domain\nfeatures were extracted. Eight feature selection algorithms are explored, and\nthe best feature selection technique is selected for subsequent evaluations.\nThe performance of six ML algorithms is evaluated, and an ensemble model is\nproposed. Deep Learning (DL) models are also explored, and the results are\ncompared with the classical ML approach. The proposed ensemble model, by\nstacking the five best logistic regression models, generated an overall\naccuracy of 81.13% and an F1 score of 81.12% in the classification of four\ninner speech words using surface EEG signals. The proposed framework with the\nproposed ensemble of classical ML models shows promise in the classification of\ninner speech using surface EEG signals.\n", "date": "2024-12-09", "categories": []}
{"id": "2412.18354", "title": "The Thousand Brains Project: A New Paradigm for Sensorimotor\n  Intelligence", "abstract": "  Artificial intelligence has advanced rapidly in the last decade, driven\nprimarily by progress in the scale of deep-learning systems. Despite these\nadvances, the creation of intelligent systems that can operate effectively in\ndiverse, real-world environments remains a significant challenge. In this white\npaper, we outline the Thousand Brains Project, an ongoing research effort to\ndevelop an alternative, complementary form of AI, derived from the operating\nprinciples of the neocortex. We present an early version of a thousand-brains\nsystem, a sensorimotor agent that is uniquely suited to quickly learn a wide\nrange of tasks and eventually implement any capabilities the human neocortex\nhas. Core to its design is the use of a repeating computational unit, the\nlearning module, modeled on the cortical columns found in mammalian brains.\nEach learning module operates as a semi-independent unit that can model entire\nobjects, represents information through spatially structured reference frames,\nand both estimates and is able to effect movement in the world. Learning is a\nquick, associative process, similar to Hebbian learning in the brain, and\nleverages inductive biases around the spatial structure of the world to enable\nrapid and continual learning. Multiple learning modules can interact with one\nanother both hierarchically and non-hierarchically via a \"cortical messaging\nprotocol\" (CMP), creating more abstract representations and supporting\nmultimodal integration. We outline the key principles motivating the design of\nthousand-brains systems and provide details about the implementation of Monty,\nour first instantiation of such a system. Code can be found at\nhttps://github.com/thousandbrainsproject/tbp.monty, along with more detailed\ndocumentation at https://thousandbrainsproject.readme.io/.\n", "date": "2024-12-24", "categories": []}
{"id": "2412.16773", "title": "Fast Multi-Group Gaussian Process Factor Models", "abstract": "  Gaussian processes are now commonly used in dimensionality reduction\napproaches tailored to neuroscience, especially to describe changes in\nhigh-dimensional neural activity over time. As recording capabilities expand to\ninclude neuronal populations across multiple brain areas, cortical layers, and\ncell types, interest in extending Gaussian process factor models to\ncharacterize multi-population interactions has grown. However, the cubic\nruntime scaling of current methods with the length of experimental trials and\nthe number of recorded populations (groups) precludes their application to\nlarge-scale multi-population recordings. Here, we improve this scaling from\ncubic to linear in both trial length and group number. We present two\napproximate approaches to fitting multi-group Gaussian process factor models\nbased on (1) inducing variables and (2) the frequency domain. Empirically, both\nmethods achieved orders of magnitude speed-up with minimal impact on\nstatistical performance, in simulation and on neural recordings of hundreds of\nneurons across three brain areas. The frequency domain approach, in particular,\nconsistently provided the greatest runtime benefits with the fewest trade-offs\nin statistical performance. We further characterize the estimation biases\nintroduced by the frequency domain approach and demonstrate effective\nstrategies to mitigate them. This work enables a powerful class of analysis\ntechniques to keep pace with the growing scale of multi-population recordings,\nopening new avenues for exploring brain function.\n", "date": "2024-12-21", "categories": []}
{"id": "2410.14697", "title": "Learning Cortico-Muscular Dependence through Orthonormal Decomposition\n  of Density Ratios", "abstract": "  The cortico-spinal neural pathway is fundamental for motor control and\nmovement execution, and in humans it is typically studied using concurrent\nelectroencephalography (EEG) and electromyography (EMG) recordings. However,\ncurrent approaches for capturing high-level and contextual connectivity between\nthese recordings have important limitations. Here, we present a novel\napplication of statistical dependence estimators based on orthonormal\ndecomposition of density ratios to model the relationship between cortical and\nmuscle oscillations. Our method extends from traditional scalar-valued measures\nby learning eigenvalues, eigenfunctions, and projection spaces of density\nratios from realizations of the signal, addressing the interpretability,\nscalability, and local temporal dependence of cortico-muscular connectivity. We\nexperimentally demonstrate that eigenfunctions learned from cortico-muscular\nconnectivity can accurately classify movements and subjects. Moreover, they\nreveal channel and temporal dependencies that confirm the activation of\nspecific EEG channels during movement. Our code is available at\nhttps://github.com/bohu615/corticomuscular-eigen-encoder.\n", "date": "2024-10-04", "categories": []}
{"id": "2412.15560", "title": "Predicting Artificial Neural Network Representations to Learn\n  Recognition Model for Music Identification from Brain Recordings", "abstract": "  Recent studies have demonstrated that the representations of artificial\nneural networks (ANNs) can exhibit notable similarities to cortical\nrepresentations when subjected to identical auditory sensory inputs. In these\nstudies, the ability to predict cortical representations is probed by\nregressing from ANN representations to cortical representations. Building upon\nthis concept, our approach reverses the direction of prediction: we utilize ANN\nrepresentations as a supervisory signal to train recognition models using noisy\nbrain recordings obtained through non-invasive measurements. Specifically, we\nfocus on constructing a recognition model for music identification, where\nelectroencephalography (EEG) brain recordings collected during music listening\nserve as input. By training an EEG recognition model to predict ANN\nrepresentations-representations associated with music identification-we\nobserved a substantial improvement in classification accuracy. This study\nintroduces a novel approach to developing recognition models for brain\nrecordings in response to external auditory stimuli. It holds promise for\nadvancing brain-computer interfaces (BCI), neural decoding techniques, and our\nunderstanding of music cognition. Furthermore, it provides new insights into\nthe relationship between auditory brain activity and ANN representations.\n", "date": "2024-12-20", "categories": []}
{"id": "2412.15620", "title": "Modeling Autonomous Shifts Between Focus State and Mind-Wandering Using\n  a Predictive-Coding-Inspired Variational RNN Model", "abstract": "  The current study investigates possible neural mechanisms underling\nautonomous shifts between focus state and mind-wandering by conducting model\nsimulation experiments. On this purpose, we modeled perception processes of\ncontinuous sensory sequences using our previous proposed variational RNN model\nwhich was developed based on the free energy principle. The current study\nextended this model by introducing an adaptation mechanism of a meta-level\nparameter, referred to as the meta-prior $\\mathbf{w}$, which regulates the\ncomplexity term in the free energy. Our simulation experiments demonstrated\nthat autonomous shifts between focused perception and mind-wandering take place\nwhen $\\mathbf{w}$ switches between low and high values associated with decrease\nand increase of the average reconstruction error over the past window. In\nparticular, high $\\mathbf{w}$ prioritized top-down predictions while low\n$\\mathbf{w}$ emphasized bottom-up sensations. This paper explores how our\nexperiment results align with existing studies and highlights their potential\nfor future research.\n", "date": "2024-12-20", "categories": []}
{"id": "2311.07315", "title": "An introduction to reinforcement learning for neuroscience", "abstract": "  Reinforcement learning (RL) has a rich history in neuroscience, from early\nwork on dopamine as a reward prediction error signal (Schultz et al., 1997) to\nrecent work proposing that the brain could implement a form of 'distributional\nreinforcement learning' popularized in machine learning (Dabney et al., 2020).\nThere has been a close link between theoretical advances in reinforcement\nlearning and neuroscience experiments throughout this literature, and the\ntheories describing the experimental data have therefore become increasingly\ncomplex. Here, we provide an introduction and mathematical background to many\nof the methods that have been used in systems neroscience. We start with an\noverview of the RL problem and classical temporal difference algorithms,\nfollowed by a discussion of 'model-free', 'model-based', and intermediate RL\nalgorithms. We then introduce deep reinforcement learning and discuss how this\nframework has led to new insights in neuroscience. This includes a particular\nfocus on meta-reinforcement learning (Wang et al., 2018) and distributional RL\n(Dabney et al., 2020). Finally, we discuss potential shortcomings of the RL\nformalism for neuroscience and highlight open questions in the field. Code that\nimplements the methods discussed and generates the figures is also provided.\n", "date": "2023-11-13", "categories": []}
{"id": "2404.13282", "title": "Wills Aligner: Multi-Subject Collaborative Brain Visual Decoding", "abstract": "  Decoding visual information from human brain activity has seen remarkable\nadvancements in recent research. However, the diversity in cortical\nparcellation and fMRI patterns across individuals has prompted the development\nof deep learning models tailored to each subject. The personalization limits\nthe broader applicability of brain visual decoding in real-world scenarios. To\naddress this issue, we introduce Wills Aligner, a novel approach designed to\nachieve multi-subject collaborative brain visual decoding. Wills Aligner begins\nby aligning the fMRI data from different subjects at the anatomical level. It\nthen employs delicate mixture-of-brain-expert adapters and a meta-learning\nstrategy to account for individual fMRI pattern differences. Additionally,\nWills Aligner leverages the semantic relation of visual stimuli to guide the\nlearning of inter-subject commonality, enabling visual decoding for each\nsubject to draw insights from other subjects' data. We rigorously evaluate our\nWills Aligner across various visual decoding tasks, including classification,\ncross-modal retrieval, and image reconstruction. The experimental results\ndemonstrate that Wills Aligner achieves promising performance.\n", "date": "2024-04-20", "categories": []}
{"id": "2412.06839", "title": "A Neural Model of Rule Discovery with Relatively Short-Term Sequence\n  Memory", "abstract": "  This report proposes a neural cognitive model for discovering regularities in\nevent sequences. In a fluid intelligence task, the subject is required to\ndiscover regularities from relatively short-term memory of the first-seen task.\nSome fluid intelligence tasks require discovering regularities in event\nsequences. Thus, a neural network model was constructed to explain fluid\nintelligence or regularity discovery in event sequences with relatively\nshort-term memory. The model was implemented and tested with delayed\nmatch-to-sample tasks.\n", "date": "2024-12-07", "categories": []}
{"id": "2412.07102", "title": "Primary visual cortex contributes to color constancy by predicting\n  rather than discounting the illuminant: evidence from a computational study", "abstract": "  Color constancy (CC) is an important ability of the human visual system to\nstably perceive the colors of objects despite considerable changes in the color\nof the light illuminating them. While increasing evidence from the field of\nneuroscience supports that multiple levels of the visual system contribute to\nthe realization of CC, how the primary visual cortex (V1) plays role in CC is\nnot fully resolved. In specific, double-opponent (DO) neurons in V1 have been\nthought to contribute to realizing a degree of CC, but the computational\nmechanism is not clear. We build an electrophysiologically based V1 neural\nmodel to learn the color of the light source from a natural image dataset with\nthe ground truth illuminants as the labels. Based on the qualitative and\nquantitative analysis of the responsive properties of the learned model\nneurons, we found that both the spatial structures and color weights of the\nreceptive fields of the learned model neurons are quite similar to those of the\nsimple and DO neurons recorded in V1. Computationally, DO cells perform more\nrobustly than the simple cells in V1 for illuminant prediction. Therefore, this\nwork provides computational evidence supporting that V1 DO neurons serve to\nrealize color constancy by encoding the illuminant,which is contradictory to\nthe common hypothesis that V1 contributes to CC by discounting the illuminant\nusing its DO cells. This evidence is expected to not only help resolve the\nvisual mechanisms of CC, but also provide inspiration to develop more effective\ncomputer vision models.\n", "date": "2024-12-10", "categories": []}
{"id": "2401.17671", "title": "Contextual Feature Extraction Hierarchies Converge in Large Language\n  Models and the Brain", "abstract": "  Recent advancements in artificial intelligence have sparked interest in the\nparallels between large language models (LLMs) and human neural processing,\nparticularly in language comprehension. While prior research has established\nsimilarities in the representation of LLMs and the brain, the underlying\ncomputational principles that cause this convergence, especially in the context\nof evolving LLMs, remain elusive. Here, we examined a diverse selection of\nhigh-performance LLMs with similar parameter sizes to investigate the factors\ncontributing to their alignment with the brain's language processing\nmechanisms. We find that as LLMs achieve higher performance on benchmark tasks,\nthey not only become more brain-like as measured by higher performance when\npredicting neural responses from LLM embeddings, but also their hierarchical\nfeature extraction pathways map more closely onto the brain's while using fewer\nlayers to do the same encoding. We also compare the feature extraction pathways\nof the LLMs to each other and identify new ways in which high-performing models\nhave converged toward similar hierarchical processing mechanisms. Finally, we\nshow the importance of contextual information in improving model performance\nand brain similarity. Our findings reveal the converging aspects of language\nprocessing in the brain and LLMs and offer new directions for developing models\nthat align more closely with human cognitive processing.\n", "date": "2024-01-31", "categories": []}
{"id": "2405.14473", "title": "Poisson Variational Autoencoder", "abstract": "  Variational autoencoders (VAEs) employ Bayesian inference to interpret\nsensory inputs, mirroring processes that occur in primate vision across both\nventral (Higgins et al., 2021) and dorsal (Vafaii et al., 2023) pathways.\nDespite their success, traditional VAEs rely on continuous latent variables,\nwhich deviates sharply from the discrete nature of biological neurons. Here, we\ndeveloped the Poisson VAE (P-VAE), a novel architecture that combines\nprinciples of predictive coding with a VAE that encodes inputs into discrete\nspike counts. Combining Poisson-distributed latent variables with predictive\ncoding introduces a metabolic cost term in the model loss function, suggesting\na relationship with sparse coding which we verify empirically. Additionally, we\nanalyze the geometry of learned representations, contrasting the P-VAE to\nalternative VAE models. We find that the P-VAE encodes its inputs in relatively\nhigher dimensions, facilitating linear separability of categories in a\ndownstream classification task with a much better (5x) sample efficiency. Our\nwork provides an interpretable computational framework to study brain-like\nsensory processing and paves the way for a deeper understanding of perception\nas an inferential process.\n", "date": "2024-05-23", "categories": []}
{"id": "2411.03541", "title": "Do Mice Grok? Glimpses of Hidden Progress During Overtraining in Sensory\n  Cortex", "abstract": "  Does learning of task-relevant representations stop when behavior stops\nchanging? Motivated by recent theoretical advances in machine learning and the\nintuitive observation that human experts continue to learn from practice even\nafter mastery, we hypothesize that task-specific representation learning can\ncontinue, even when behavior plateaus. In a novel reanalysis of recently\npublished neural data, we find evidence for such learning in posterior piriform\ncortex of mice following continued training on a task, long after behavior\nsaturates at near-ceiling performance (\"overtraining\"). This learning is marked\nby an increase in decoding accuracy from piriform neural populations and\nimproved performance on held-out generalization tests. We demonstrate that\nclass representations in cortex continue to separate during overtraining, so\nthat examples that were incorrectly classified at the beginning of overtraining\ncan abruptly be correctly classified later on, despite no changes in behavior\nduring that time. We hypothesize this hidden yet rich learning takes the form\nof approximate margin maximization; we validate this and other predictions in\nthe neural data, as well as build and interpret a simple synthetic model that\nrecapitulates these phenomena. We conclude by showing how this model of\nlate-time feature learning implies an explanation for the empirical puzzle of\novertraining reversal in animal learning, where task-specific representations\nare more robust to particular task changes because the learned features can be\nreused.\n", "date": "2024-11-05", "categories": []}
{"id": "2411.16075", "title": "The brain versus AI: World-model-based versatile circuit computation\n  underlying diverse functions in the neocortex and cerebellum", "abstract": "  AI's significant recent advances using general-purpose circuit computations\noffer a potential window into how the neocortex and cerebellum of the brain are\nable to achieve a diverse range of functions across sensory, cognitive, and\nmotor domains, despite their uniform circuit structures. However, comparing the\nbrain and AI is challenging unless clear similarities exist, and past reviews\nhave been limited to comparison of brain-inspired vision AI and the visual\nneocortex. Here, to enable comparisons across diverse functional domains, we\nsubdivide circuit computation into three elements -- circuit structure,\ninput/outputs, and the learning algorithm -- and evaluate the similarities for\neach element. With this novel approach, we identify wide-ranging similarities\nand convergent evolution in the brain and AI, providing new insights into key\nconcepts in neuroscience. Furthermore, inspired by processing mechanisms of AI,\nwe propose a new theory that integrates established neuroscience theories,\nparticularly the theories of internal models and the mirror neuron system. Both\nthe neocortex and cerebellum predict future world events from past information\nand learn from prediction errors, thereby acquiring models of the world. These\nmodels enable three core processes: (1) Prediction -- generating future\ninformation, (2) Understanding -- interpreting the external world via\ncompressed and abstracted sensory information, and (3) Generation --\nrepurposing the future-information generation mechanism to produce other types\nof outputs. The universal application of these processes underlies the ability\nof the neocortex and cerebellum to accomplish diverse functions with uniform\ncircuits. Our systematic approach, insights, and theory promise groundbreaking\nadvances in understanding the brain.\n", "date": "2024-11-25", "categories": []}
{"id": "2410.09614", "title": "Exploring Behavior-Relevant and Disentangled Neural Dynamics with\n  Generative Diffusion Models", "abstract": "  Understanding the neural basis of behavior is a fundamental goal in\nneuroscience. Current research in large-scale neuro-behavioral data analysis\noften relies on decoding models, which quantify behavioral information in\nneural data but lack details on behavior encoding. This raises an intriguing\nscientific question: ``how can we enable in-depth exploration of neural\nrepresentations in behavioral tasks, revealing interpretable neural dynamics\nassociated with behaviors''. However, addressing this issue is challenging due\nto the varied behavioral encoding across different brain regions and mixed\nselectivity at the population level. To tackle this limitation, our approach,\nnamed ``BeNeDiff'', first identifies a fine-grained and disentangled neural\nsubspace using a behavior-informed latent variable model. It then employs\nstate-of-the-art generative diffusion models to synthesize behavior videos that\ninterpret the neural dynamics of each latent factor. We validate the method on\nmulti-session datasets containing widefield calcium imaging recordings across\nthe dorsal cortex. Through guiding the diffusion model to activate individual\nlatent factors, we verify that the neural dynamics of latent factors in the\ndisentangled neural subspace provide interpretable quantifications of the\nbehaviors of interest. At the same time, the neural subspace in BeNeDiff\ndemonstrates high disentanglement and neural reconstruction quality.\n", "date": "2024-10-12", "categories": []}
{"id": "2411.10458", "title": "Neural decoding from stereotactic EEG: accounting for electrode\n  variability across subjects", "abstract": "  Deep learning based neural decoding from stereotactic electroencephalography\n(sEEG) would likely benefit from scaling up both dataset and model size. To\nachieve this, combining data across multiple subjects is crucial. However, in\nsEEG cohorts, each subject has a variable number of electrodes placed at\ndistinct locations in their brain, solely based on clinical needs. Such\nheterogeneity in electrode number/placement poses a significant challenge for\ndata integration, since there is no clear correspondence of the neural activity\nrecorded at distinct sites between individuals. Here we introduce seegnificant:\na training framework and architecture that can be used to decode behavior\nacross subjects using sEEG data. We tokenize the neural activity within\nelectrodes using convolutions and extract long-term temporal dependencies\nbetween tokens using self-attention in the time dimension. The 3D location of\neach electrode is then mixed with the tokens, followed by another\nself-attention in the electrode dimension to extract effective spatiotemporal\nneural representations. Subject-specific heads are then used for downstream\ndecoding tasks. Using this approach, we construct a multi-subject model trained\non the combined data from 21 subjects performing a behavioral task. We\ndemonstrate that our model is able to decode the trial-wise response time of\nthe subjects during the behavioral task solely from neural data. We also show\nthat the neural representations learned by pretraining our model across\nindividuals can be transferred in a few-shot manner to new subjects. This work\nintroduces a scalable approach towards sEEG data integration for multi-subject\nmodel training, paving the way for cross-subject generalization for sEEG\ndecoding.\n", "date": "2024-11-01", "categories": []}
{"id": "2411.08197", "title": "What Representational Similarity Measures Imply about Decodable\n  Information", "abstract": "  Neural responses encode information that is useful for a variety of\ndownstream tasks. A common approach to understand these systems is to build\nregression models or ``decoders'' that reconstruct features of the stimulus\nfrom neural responses. Popular neural network similarity measures like centered\nkernel alignment (CKA), canonical correlation analysis (CCA), and Procrustes\nshape distance, do not explicitly leverage this perspective and instead\nhighlight geometric invariances to orthogonal or affine transformations when\ncomparing representations. Here, we show that many of these measures can, in\nfact, be equivalently motivated from a decoding perspective. Specifically,\nmeasures like CKA and CCA quantify the average alignment between optimal linear\nreadouts across a distribution of decoding tasks. We also show that the\nProcrustes shape distance upper bounds the distance between optimal linear\nreadouts and that the converse holds for representations with low participation\nratio. Overall, our work demonstrates a tight link between the geometry of\nneural representations and the ability to linearly decode information. This\nperspective suggests new ways of measuring similarity between neural systems\nand also provides novel, unifying interpretations of existing measures.\n", "date": "2024-11-12", "categories": []}
{"id": "2411.08221", "title": "SynapsNet: Enhancing Neuronal Population Dynamics Modeling via Learning\n  Functional Connectivity", "abstract": "  The availability of large-scale neuronal population datasets necessitates new\nmethods to model population dynamics and extract interpretable, scientifically\ntranslatable insights. Existing deep learning methods often overlook the\nbiological mechanisms underlying population activity and thus exhibit\nsuboptimal performance with neuronal data and provide little to no\ninterpretable information about neurons and their interactions. In response, we\nintroduce SynapsNet, a novel deep-learning framework that effectively models\npopulation dynamics and functional interactions between neurons. Within this\nbiologically realistic framework, each neuron, characterized by a latent\nembedding, sends and receives currents through directed connections. A shared\ndecoder uses the input current, previous neuronal activity, neuron embedding,\nand behavioral data to predict the population activity in the next time step.\nUnlike common sequential models that treat population activity as a\nmultichannel time series, SynapsNet applies its decoder to each neuron\n(channel) individually, with the learnable functional connectivity serving as\nthe sole pathway for information flow between neurons. Our experiments,\nconducted on mouse cortical activity from publicly available datasets and\nrecorded using the two most common population recording modalities (Ca imaging\nand Neuropixels) across three distinct tasks, demonstrate that SynapsNet\nconsistently outperforms existing models in forecasting population activity.\nAdditionally, our experiments on both real and synthetic data showed that\nSynapsNet accurately learns functional connectivity that reveals predictive\ninteractions between neurons.\n", "date": "2024-11-12", "categories": []}
{"id": "2312.11436", "title": "Layerwise complexity-matched learning yields an improved model of\n  cortical area V2", "abstract": "  Human ability to recognize complex visual patterns arises through\ntransformations performed by successive areas in the ventral visual cortex.\nDeep neural networks trained end-to-end for object recognition approach human\ncapabilities, and offer the best descriptions to date of neural responses in\nthe late stages of the hierarchy. But these networks provide a poor account of\nthe early stages, compared to traditional hand-engineered models, or models\noptimized for coding efficiency or prediction. Moreover, the gradient\nbackpropagation used in end-to-end learning is generally considered to be\nbiologically implausible. Here, we overcome both of these limitations by\ndeveloping a bottom-up self-supervised training methodology that operates\nindependently on successive layers. Specifically, we maximize feature\nsimilarity between pairs of locally-deformed natural image patches, while\ndecorrelating features across patches sampled from other images. Crucially, the\ndeformation amplitudes are adjusted proportionally to receptive field sizes in\neach layer, thus matching the task complexity to the capacity at each stage of\nprocessing. In comparison with architecture-matched versions of previous\nmodels, we demonstrate that our layerwise complexity-matched learning (LCL)\nformulation produces a two-stage model (LCL-V2) that is better aligned with\nselectivity properties and neural activity in primate area V2. We demonstrate\nthat the complexity-matched learning paradigm is responsible for much of the\nemergence of the improved biological alignment. Finally, when the two-stage\nmodel is used as a fixed front-end for a deep network trained to perform object\nrecognition, the resultant model (LCL-V2Net) is significantly better than\nstandard end-to-end self-supervised, supervised, and adversarially-trained\nmodels in terms of generalization to out-of-distribution tasks and alignment\nwith human behavior.\n", "date": "2023-12-18", "categories": []}
{"id": "2407.11024", "title": "A mathematical framework of intelligence and consciousness based on\n  Riemannian Geometry", "abstract": "  Understanding intelligence is a central pursuit in neuroscience, cognitive\nscience, and artificial intelligence. Intelligence encompasses learning,\nproblem-solving, creativity, and even consciousness. Recent advancements in\ngeometric analysis have revealed new insights into high-dimensional information\nrepresentation and organisation, exposing intrinsic data structures and dynamic\nprocesses within neural and artificial systems. However, a comprehensive\nframework that unifies the static and dynamic aspects of intelligence is still\nlacking. This manuscript proposes a mathematical framework based on Riemannian\ngeometry to describe the structure and dynamics of intelligence and\nconsciousness. Intelligence elements are conceptualised as tokens embedded in a\nhigh-dimensional space. The learned token embeddings capture the\ninterconnections of tokens across various scenarios and tasks, forming\nmanifolds in the intelligence space. Thought flow is depicted as the sequential\nactivation of tokens along geodesics within these manifolds. During the\nnavigation of geodesics, consciousness, as a self-referential process,\nperceives the thought flow, evaluates it against predictions, and provides\nfeedback through prediction errors, adjusting the geodesic: non-zero prediction\nerrors, such as learning, lead to the restructuring of the curved manifolds,\nthus changing the geodesic of thought flow. This dynamic interaction integrates\nnew information, evolves the geometry and facilitates learning. The geometry of\nintelligence guides consciousness, and consciousness structures the geometry of\nintelligence. By integrating geometric concepts, this proposed theory offers a\nunified, mathematically framework for describing the structure and dynamics of\nintelligence and consciousness. Applicable to biological and artificial\nintelligence, this framework may pave the way for future research and empirical\nvalidation.\n", "date": "2024-07-02", "categories": []}
{"id": "2301.07016", "title": "Consciousness is entailed by compositional learning of new causal\n  structures in deep predictive processing systems", "abstract": "  Machine learning algorithms have achieved superhuman performance in specific\ncomplex domains. However, learning online from few examples and compositional\nlearning for efficient generalization across domains remain elusive. In humans,\nsuch learning includes specific declarative memory formation and is closely\nassociated with consciousness. Predictive processing has been advanced as a\nprincipled Bayesian framework for understanding the cortex as implementing deep\ngenerative models for both sensory perception and action control. However,\npredictive processing offers little direct insight into fast compositional\nlearning or of the separation between conscious and unconscious contents. Here,\npropose that access consciousness arises as a consequence of a particular\nlearning mechanism operating within a predictive processing system. We extend\npredictive processing by adding online, single-example new structure learning\nvia hierarchical binding of unpredicted inferences. This system learns new\ncauses by quickly connecting together novel combinations of perceptions, which\nmanifests as working memories that can become short- and long-term declarative\nmemories retrievable by associative recall. The contents of such bound\nrepresentations are unified yet differentiated, can be maintained by selective\nattention and are globally available. The proposed learning process explains\ncontrast and masking manipulations, postdictive perceptual integration, and\nother paradigm cases of consciousness research. 'Phenomenal conscious\nexperience' is how the learning system transparently models its own\nfunctioning, giving rise to perceptual illusions underlying the meta-problem of\nconsciousness. Our proposal naturally unifies the feature binding, recurrent\nprocessing, predictive processing, and global workspace theories of\nconsciousness.\n", "date": "2023-01-17", "categories": []}
{"id": "2411.02783", "title": "BrainBits: How Much of the Brain are Generative Reconstruction Methods\n  Using?", "abstract": "  When evaluating stimuli reconstruction results it is tempting to assume that\nhigher fidelity text and image generation is due to an improved understanding\nof the brain or more powerful signal extraction from neural recordings.\nHowever, in practice, new reconstruction methods could improve performance for\nat least three other reasons: learning more about the distribution of stimuli,\nbecoming better at reconstructing text or images in general, or exploiting\nweaknesses in current image and/or text evaluation metrics. Here we disentangle\nhow much of the reconstruction is due to these other factors vs. productively\nusing the neural recordings. We introduce BrainBits, a method that uses a\nbottleneck to quantify the amount of signal extracted from neural recordings\nthat is actually necessary to reproduce a method's reconstruction fidelity. We\nfind that it takes surprisingly little information from the brain to produce\nreconstructions with high fidelity. In these cases, it is clear that the priors\nof the methods' generative models are so powerful that the outputs they produce\nextrapolate far beyond the neural signal they decode. Given that reconstructing\nstimuli can be improved independently by either improving signal extraction\nfrom the brain or by building more powerful generative models, improving the\nlatter may fool us into thinking we are improving the former. We propose that\nmethods should report a method-specific random baseline, a reconstruction\nceiling, and a curve of performance as a function of bottleneck size, with the\nultimate goal of using more of the neural recordings.\n", "date": "2024-11-05", "categories": []}
{"id": "2410.22283", "title": "Leveraging Recurrent Neural Networks for Predicting Motor Movements from\n  Primate Motor Cortex Neural Recordings", "abstract": "  This paper presents an efficient deep learning solution for decoding motor\nmovements from neural recordings in non-human primates. An Autoencoder Gated\nRecurrent Unit (AEGRU) model was adopted as the model architecture for this\ntask. The autoencoder is only used during the training stage to achieve better\ngeneralization. Together with the preprocessing techniques, our model achieved\n0.71 $R^2$ score, surpassing the baseline models in Neurobench and is ranked\nfirst for $R^2$ in the IEEE BioCAS 2024 Grand Challenge on Neural Decoding.\nModel pruning is also applied leading to a reduction of 41.4% of the\nmultiply-accumulate (MAC) operations with little change in the $R^2$ score\ncompared to the unpruned model.\n", "date": "2024-10-29", "categories": []}
{"id": "2410.23812", "title": "Graph Neural Networks Uncover Geometric Neural Representations in\n  Reinforcement-Based Motor Learning", "abstract": "  Graph Neural Networks (GNN) can capture the geometric properties of neural\nrepresentations in EEG data. Here we utilise those to study how\nreinforcement-based motor learning affects neural activity patterns during\nmotor planning, leveraging the inherent graph structure of EEG channels to\ncapture the spatial relationships in brain activity. By exploiting\ntask-specific symmetries, we define different pretraining strategies that not\nonly improve model performance across all participant groups but also validate\nthe robustness of the geometric representations. Explainability analysis based\non the graph structures reveals consistent group-specific neural signatures\nthat persist across pretraining conditions, suggesting stable geometric\nstructures in the neural representations associated with motor learning and\nfeedback processing. These geometric patterns exhibit partial invariance to\ncertain task space transformations, indicating symmetries that enable\ngeneralisation across conditions while maintaining specificity to individual\nlearning strategies. This work demonstrates how GNNs can uncover the effects of\nprevious outcomes on motor planning, in a complex real-world task, providing\ninsights into the geometric principles governing neural representations. Our\nexperimental design bridges the gap between controlled experiments and\necologically valid scenarios, offering new insights into the organisation of\nneural representations during naturalistic motor learning, which may open\navenues for exploring fundamental principles governing brain activity in\ncomplex tasks.\n", "date": "2024-10-31", "categories": []}
{"id": "2410.19838", "title": "Non-invasive Neural Decoding in Source Reconstructed Brain Space", "abstract": "  Non-invasive brainwave decoding is usually done using\nMagneto/Electroencephalography (MEG/EEG) sensor measurements as inputs. This\nmakes combining datasets and building models with inductive biases difficult as\nmost datasets use different scanners and the sensor arrays have a nonintuitive\nspatial structure. In contrast, fMRI scans are acquired directly in brain\nspace, a voxel grid with a typical structured input representation. By using\nestablished techniques to reconstruct the sensors' sources' neural activity it\nis possible to decode from voxels for MEG data as well. We show that this\nenables spatial inductive biases, spatial data augmentations, better\ninterpretability, zero-shot generalisation between datasets, and data\nharmonisation.\n", "date": "2024-10-20", "categories": []}
{"id": "2410.18395", "title": "A contrastive-learning approach for auditory attention detection", "abstract": "  Carrying conversations in multi-sound environments is one of the more\nchallenging tasks, since the sounds overlap across time and frequency making it\ndifficult to understand a single sound source. One proposed approach to help\nisolate an attended speech source is through decoding the electroencephalogram\n(EEG) and identifying the attended audio source using statistical or machine\nlearning techniques. However, the limited amount of data in comparison to other\nmachine learning problems and the distributional shift between different EEG\nrecordings emphasizes the need for a self supervised approach that works with\nlimited data to achieve a more robust solution. In this paper, we propose a\nmethod based on self supervised learning to minimize the difference between the\nlatent representations of an attended speech signal and the corresponding EEG\nsignal. This network is further finetuned for the auditory attention\nclassification task. We compare our results with previously published methods\nand achieve state-of-the-art performance on the validation set.\n", "date": "2024-10-24", "categories": []}
{"id": "2309.11028", "title": "The Topology and Geometry of Neural Representations", "abstract": "  A central question for neuroscience is how to characterize brain\nrepresentations of perceptual and cognitive content. An ideal characterization\nshould distinguish different functional regions with robustness to noise and\nidiosyncrasies of individual brains that do not correspond to computational\ndifferences. Previous studies have characterized brain representations by their\nrepresentational geometry, which is defined by the representational\ndissimilarity matrix (RDM), a summary statistic that abstracts from the roles\nof individual neurons (or responses channels) and characterizes the\ndiscriminability of stimuli. Here we explore a further step of abstraction:\nfrom the geometry to the topology of brain representations. We propose\ntopological representational similarity analysis (tRSA), an extension of\nrepresentational similarity analysis (RSA) that uses a family of\ngeo-topological summary statistics that generalizes the RDM to characterize the\ntopology while de-emphasizing the geometry. We evaluate this new family of\nstatistics in terms of the sensitivity and specificity for model selection\nusing both simulations and fMRI data. In the simulations, the ground truth is a\ndata-generating layer representation in a neural network model and the models\nare the same and other layers in different model instances (trained from\ndifferent random seeds). In fMRI, the ground truth is a visual area and the\nmodels are the same and other areas measured in different subjects. Results\nshow that topology-sensitive characterizations of population codes are robust\nto noise and interindividual variability and maintain excellent sensitivity to\nthe unique representational signatures of different neural network layers and\nbrain regions. These methods enable researchers to calibrate comparisons among\nrepresentations in brains and models to be sensitive to the geometry, the\ntopology, or a combination of both.\n", "date": "2023-09-20", "categories": []}
{"id": "2410.11886", "title": "Are Grid Cells Hexagonal for Performance or by Convenience?", "abstract": "  This paper investigates whether the hexagonal structure of grid cells\nprovides any performance benefits or if it merely represents a biologically\nconvenient configuration. Utilizing the Vector-HaSH content addressable memory\nmodel as a model of the grid cell -- place cell network of the mammalian brain,\nwe compare the performance of square and hexagonal grid cells in tasks of\nstoring and retrieving spatial memories. Our experiments across different path\ntypes, path lengths and grid configurations, reveal that hexagonal grid cells\nperform similarly to square grid cells with respect to spatial representation\nand memory recall. Our results show comparable accuracy and robustness across\ndifferent datasets and noise levels on images to recall. These findings suggest\nthat the brain's use of hexagonal grids may be more a matter of biological\nconvenience and ease of implementation rather than because they provide\nsuperior performance over square grid cells (which are easier to implement in\nsilico).\n", "date": "2024-10-11", "categories": []}
{"id": "2410.11503", "title": "Network Representation Learning for Biophysical Neural Network Analysis", "abstract": "  The analysis of biophysical neural networks (BNNs) has been a longstanding\nfocus in computational neuroscience. A central yet unresolved challenge in BNN\nanalysis lies in deciphering the correlations between neuronal and synaptic\ndynamics, their connectivity patterns, and learning process. To address this,\nwe introduce a novel BNN analysis framework grounded in network representation\nlearning (NRL), which leverages attention scores to uncover intricate\ncorrelations between network components and their features. Our framework\nintegrates a new computational graph (CG)-based BNN representation, a\nbio-inspired graph attention network (BGAN) that enables multiscale correlation\nanalysis across BNN representations, and an extensive BNN dataset. The CG-based\nrepresentation captures key computational features, information flow, and\nstructural relationships underlying neuronal and synaptic dynamics, while BGAN\nreflects the compositional structure of neurons, including dendrites, somas,\nand axons, as well as bidirectional information flows between BNN components.\nThe dataset comprises publicly available models from ModelDB, reconstructed\nusing the Python and standardized in NeuroML format, and is augmented with data\nderived from canonical neuron and synapse models. To our knowledge, this study\nis the first to apply an NRL-based approach to the full spectrum of BNNs and\ntheir analysis.\n", "date": "2024-10-15", "categories": []}
{"id": "2410.07503", "title": "Modeling Alzheimer's Disease: From Memory Loss to Plaque & Tangles\n  Formation", "abstract": "  We employ the Hopfield model as a simplified framework to explore both the\nmemory deficits and the biochemical processes characteristic of Alzheimer's\ndisease. By simulating neuronal death and synaptic degradation through\nincreasing the number of stored patterns and introducing noise into the\nsynaptic weights, we demonstrate hallmark symptoms of dementia, including\nmemory loss, confusion, and delayed retrieval times. As the network's capacity\nis exceeded, retrieval errors increase, mirroring the cognitive confusion\nobserved in Alzheimer's patients. Additionally, we simulate the impact of\nsynaptic degradation by varying the sparsity of the weight matrix, showing\nimpaired memory recall and reduced retrieval success as noise levels increase.\nFurthermore, we extend our model to connect memory loss with biochemical\nprocesses linked to Alzheimer's. By simulating the role of reduced insulin\nsensitivity over time, we show how it can trigger increased calcium influx into\nmitochondria, leading to misfolded proteins and the formation of amyloid\nplaques. These findings, modeled over time, suggest that both neuronal\ndegradation and metabolic factors contribute to the progressive decline seen in\nAlzheimer's disease. Our work offers a computational framework for\nunderstanding the dual impact of synaptic and metabolic dysfunction in\nneurodegenerative diseases.\n", "date": "2024-10-10", "categories": []}
{"id": "2410.04497", "title": "Generalizability analysis of deep learning predictions of human brain\n  responses to augmented and semantically novel visual stimuli", "abstract": "  The purpose of this work is to investigate the soundness and utility of a\nneural network-based approach as a framework for exploring the impact of image\nenhancement techniques on visual cortex activation. In a preliminary study, we\nprepare a set of state-of-the-art brain encoding models, selected among the top\n10 methods that participated in The Algonauts Project 2023 Challenge [16]. We\nanalyze their ability to make valid predictions about the effects of various\nimage enhancement techniques on neural responses. Given the impossibility of\nacquiring the actual data due to the high costs associated with brain imaging\nprocedures, our investigation builds up on a series of experiments.\nSpecifically, we analyze the ability of brain encoders to estimate the cerebral\nreaction to various augmentations by evaluating the response to augmentations\ntargeting objects (i.e., faces and words) with known impact on specific areas.\nMoreover, we study the predicted activation in response to objects unseen\nduring training, exploring the impact of semantically out-of-distribution\nstimuli. We provide relevant evidence for the generalization ability of the\nmodels forming the proposed framework, which appears to be promising for the\nidentification of the optimal visual augmentation filter for a given task,\nmodel-driven design strategies as well as for AR and VR applications.\n", "date": "2024-10-06", "categories": []}
{"id": "2410.02870", "title": "Individuation of 3D perceptual units from neurogeometry of binocular\n  cells", "abstract": "  We model the functional architecture of the early stages of three-dimensional\nvision by extending the neurogeometric sub-Riemannian model for stereo-vision\nintroduced in \\cite{BCSZ23}. A new framework for correspondence is introduced\nthat integrates a neural-based algorithm to achieve stereo correspondence\nlocally while, simultaneously, organizing the corresponding points into global\nperceptual units. The result is an effective scene segmentation. We achieve\nthis using harmonic analysis on the sub-Riemannian structure and show, in a\ncomparison against Riemannian distance, that the sub-Riemannian metric is\ncentral to the solution.\n", "date": "2024-10-03", "categories": []}
{"id": "2212.00596", "title": "Language models and brains align due to more than next-word prediction\n  and word-level information", "abstract": "  Pretrained language models have been shown to significantly predict brain\nrecordings of people comprehending language. Recent work suggests that the\nprediction of the next word is a key mechanism that contributes to this\nalignment. What is not yet understood is whether prediction of the next word is\nnecessary for this observed alignment or simply sufficient, and whether there\nare other shared mechanisms or information that are similarly important. In\nthis work, we take a step towards understanding the reasons for brain alignment\nvia two simple perturbations in popular pretrained language models. These\nperturbations help us design contrasts that can control for different types of\ninformation. By contrasting the brain alignment of these differently perturbed\nmodels, we show that improvements in alignment with brain recordings are due to\nmore than improvements in next-word prediction and word-level information.\n", "date": "2022-12-01", "categories": []}
{"id": "2410.02430", "title": "Predictive Attractor Models", "abstract": "  Sequential memory, the ability to form and accurately recall a sequence of\nevents or stimuli in the correct order, is a fundamental prerequisite for\nbiological and artificial intelligence as it underpins numerous cognitive\nfunctions (e.g., language comprehension, planning, episodic memory formation,\netc.) However, existing methods of sequential memory suffer from catastrophic\nforgetting, limited capacity, slow iterative learning procedures, low-order\nMarkov memory, and, most importantly, the inability to represent and generate\nmultiple valid future possibilities stemming from the same context. Inspired by\nbiologically plausible neuroscience theories of cognition, we propose\n\\textit{Predictive Attractor Models (PAM)}, a novel sequence memory\narchitecture with desirable generative properties. PAM is a streaming model\nthat learns a sequence in an online, continuous manner by observing each input\n\\textit{only once}. Additionally, we find that PAM avoids catastrophic\nforgetting by uniquely representing past context through lateral inhibition in\ncortical minicolumns, which prevents new memories from overwriting previously\nlearned knowledge. PAM generates future predictions by sampling from a union\nset of predicted possibilities; this generative ability is realized through an\nattractor model trained alongside the predictor. We show that PAM is trained\nwith local computations through Hebbian plasticity rules in a biologically\nplausible framework. Other desirable traits (e.g., noise tolerance, CPU-based\nlearning, capacity scaling) are discussed throughout the paper. Our findings\nsuggest that PAM represents a significant step forward in the pursuit of\nbiologically plausible and computationally efficient sequential memory models,\nwith broad implications for cognitive science and artificial intelligence\nresearch.\n", "date": "2024-10-03", "categories": []}
{"id": "2405.17663", "title": "Finding Shared Decodable Concepts and their Negations in the Brain", "abstract": "  Prior work has offered evidence for functional localization in the brain;\ndifferent anatomical regions preferentially activate for certain types of\nvisual input. For example, the fusiform face area preferentially activates for\nvisual stimuli that include a face. However, the spectrum of visual semantics\nis extensive, and only a few semantically-tuned patches of cortex have so far\nbeen identified in the human brain. Using a multimodal (natural language and\nimage) neural network architecture (CLIP) we train a highly accurate\ncontrastive model that maps brain responses during naturalistic image viewing\nto CLIP embeddings. We then use a novel adaptation of the DBSCAN clustering\nalgorithm to cluster the parameters of these participant-specific contrastive\nmodels. This reveals what we call Shared Decodable Concepts (SDCs): clusters in\nCLIP space that are decodable from common sets of voxels across multiple\nparticipants.\n  Examining the images most and least associated with each SDC cluster gives us\nadditional insight into the semantic properties of each SDC. We note SDCs for\npreviously reported visual features (e.g. orientation tuning in early visual\ncortex) as well as visual semantic concepts such as faces, places and bodies.\nIn cases where our method finds multiple clusters for a visuo-semantic concept,\nthe least associated images allow us to dissociate between confounding factors.\nFor example, we discovered two clusters of food images, one driven by color,\nthe other by shape. We also uncover previously unreported areas such as regions\nof extrastriate body area (EBA) tuned for legs/hands and sensitivity to\nnumerosity in right intraparietal sulcus, and more. Thus, our\ncontrastive-learning methodology better characterizes new and existing\nvisuo-semantic representations in the brain by leveraging multimodal neural\nnetwork representations and a novel adaptation of clustering algorithms.\n", "date": "2024-05-27", "categories": []}
{"id": "2409.19407", "title": "Brain-JEPA: Brain Dynamics Foundation Model with Gradient Positioning\n  and Spatiotemporal Masking", "abstract": "  We introduce Brain-JEPA, a brain dynamics foundation model with the\nJoint-Embedding Predictive Architecture (JEPA). This pioneering model achieves\nstate-of-the-art performance in demographic prediction, disease\ndiagnosis/prognosis, and trait prediction through fine-tuning. Furthermore, it\nexcels in off-the-shelf evaluations (e.g., linear probing) and demonstrates\nsuperior generalizability across different ethnic groups, surpassing the\nprevious large model for brain activity significantly. Brain-JEPA incorporates\ntwo innovative techniques: Brain Gradient Positioning and Spatiotemporal\nMasking. Brain Gradient Positioning introduces a functional coordinate system\nfor brain functional parcellation, enhancing the positional encoding of\ndifferent Regions of Interest (ROIs). Spatiotemporal Masking, tailored to the\nunique characteristics of fMRI data, addresses the challenge of heterogeneous\ntime-series patches. These methodologies enhance model performance and advance\nour understanding of the neural circuits underlying cognition. Overall,\nBrain-JEPA is paving the way to address pivotal questions of building brain\nfunctional coordinate system and masking brain activity at the AI-neuroscience\ninterface, and setting a potentially new paradigm in brain activity analysis\nthrough downstream adaptation.\n", "date": "2024-09-28", "categories": []}
{"id": "2409.20174", "title": "Modelando procesos cognitivos de la lectura natural con GPT-2", "abstract": "  The advancement of the Natural Language Processing field has enabled the\ndevelopment of language models with a great capacity for generating text. In\nrecent years, Neuroscience has been using these models to better understand\ncognitive processes. In previous studies, we found that models like Ngrams and\nLSTM networks can partially model Predictability when used as a co-variable to\nexplain readers' eye movements. In the present work, we further this line of\nresearch by using GPT-2 based models. The results show that this architecture\nachieves better outcomes than its predecessors.\n", "date": "2024-09-30", "categories": []}
{"id": "2406.19708", "title": "A Differentiable Approach to Multi-scale Brain Modeling", "abstract": "  We present a multi-scale differentiable brain modeling workflow utilizing\nBrainPy, a unique differentiable brain simulator that combines accurate brain\nsimulation with powerful gradient-based optimization. We leverage this\ncapability of BrainPy across different brain scales. At the single-neuron\nlevel, we implement differentiable neuron models and employ gradient methods to\noptimize their fit to electrophysiological data. On the network level, we\nincorporate connectomic data to construct biologically constrained network\nmodels. Finally, to replicate animal behavior, we train these models on\ncognitive tasks using gradient-based learning rules. Experiments demonstrate\nthat our approach achieves superior performance and speed in fitting\ngeneralized leaky integrate-and-fire and Hodgkin-Huxley single neuron models.\nAdditionally, training a biologically-informed network of excitatory and\ninhibitory spiking neurons on working memory tasks successfully replicates\nobserved neural activity and synaptic weight distributions. Overall, our\ndifferentiable multi-scale simulation approach offers a promising tool to\nbridge neuroscience data across electrophysiological, anatomical, and\nbehavioral scales.\n", "date": "2024-06-28", "categories": []}
{"id": "2409.13470", "title": "Deterministic versus stochastic dynamical classifiers: opposing random\n  adversarial attacks with noise", "abstract": "  The Continuous-Variable Firing Rate (CVFR) model, widely used in neuroscience\nto describe the intertangled dynamics of excitatory biological neurons, is here\ntrained and tested as a veritable dynamically assisted classifier. To this end\nthe model is supplied with a set of planted attractors which are\nself-consistently embedded in the inter-nodes coupling matrix, via its spectral\ndecomposition. Learning to classify amounts to sculp the basin of attraction of\nthe imposed equilibria, directing different items towards the corresponding\ndestination target, which reflects the class of respective pertinence. A\nstochastic variant of the CVFR model is also studied and found to be robust to\naversarial random attacks, which corrupt the items to be classified. This\nremarkable finding is one of the very many surprising effects which arise when\nnoise and dynamical attributes are made to mutually resonate.\n", "date": "2024-09-20", "categories": []}
{"id": "2409.13471", "title": "Stimulus-to-Stimulus Learning in RNNs with Cortical Inductive Biases", "abstract": "  Animals learn to predict external contingencies from experience through a\nprocess of conditioning. A natural mechanism for conditioning is stimulus\nsubstitution, whereby the neuronal response to a stimulus with no prior\nbehavioral significance becomes increasingly identical to that generated by a\nbehaviorally significant stimulus it reliably predicts. We propose a recurrent\nneural network model of stimulus substitution which leverages two forms of\ninductive bias pervasive in the cortex: representational inductive bias in the\nform of mixed stimulus representations, and architectural inductive bias in the\nform of two-compartment pyramidal neurons that have been shown to serve as a\nfundamental unit of cortical associative learning. The properties of these\nneurons allow for a biologically plausible learning rule that implements\nstimulus substitution, utilizing only information available locally at the\nsynapses. We show that the model generates a wide array of conditioning\nphenomena, and can learn large numbers of associations with an amount of\ntraining commensurate with animal experiments, without relying on parameter\nfine-tuning for each individual experimental task. In contrast, we show that\ncommonly used Hebbian rules fail to learn generic stimulus-stimulus\nassociations with mixed selectivity, and require task-specific parameter\nfine-tuning. Our framework highlights the importance of multi-compartment\nneuronal processing in the cortex, and showcases how it might confer cortical\nanimals the evolutionary edge.\n", "date": "2024-09-20", "categories": []}
{"id": "2409.13622", "title": "pAE: An Efficient Autoencoder Architecture for Modeling the Lateral\n  Geniculate Nucleus by Integrating Feedforward and Feedback Streams in Human\n  Visual System", "abstract": "  The visual cortex is a vital part of the brain, responsible for\nhierarchically identifying objects. Understanding the role of the lateral\ngeniculate nucleus (LGN) as a prior region of the visual cortex is crucial when\nprocessing visual information in both bottom-up and top-down pathways. When\nvisual stimuli reach the retina, they are transmitted to the LGN area for\ninitial processing before being sent to the visual cortex for further\nprocessing. In this study, we introduce a deep convolutional model that closely\napproximates human visual information processing. We aim to approximate the\nfunction for the LGN area using a trained shallow convolutional model which is\ndesigned based on a pruned autoencoder (pAE) architecture. The pAE model\nattempts to integrate feed forward and feedback streams from/to the V1 area\ninto the problem. This modeling framework encompasses both temporal and\nnon-temporal data feeding modes of the visual stimuli dataset containing\nnatural images captured by a fixed camera in consecutive frames, featuring two\ncategories: images with animals (in motion), and images without animals.\nSubsequently, we compare the results of our proposed deep-tuned model with\nwavelet filter bank methods employing Gabor and biorthogonal wavelet functions.\nOur experiments reveal that the proposed method based on the deep-tuned model\nnot only achieves results with high similarity in comparison with human\nbenchmarks but also performs significantly better than other models. The pAE\nmodel achieves the final 99.26% prediction performance and demonstrates a\nnotable improvement of around 28% over human results in the temporal mode.\n", "date": "2024-09-20", "categories": []}
{"id": "2409.12567", "title": "Model calibration using a parallel differential evolution algorithm in\n  computational neuroscience: simulation of stretch induced nerve deficit", "abstract": "  Neuronal damage, in the form of both brain and spinal cord injuries, is one\nof the major causes of disability and death in young adults worldwide. One way\nto assess the direct damage occurring after a mechanical insult is the\nsimulation of the neuronal cells functional deficits following the mechanical\nevent. In this study, we use a coupled mechanical electrophysiological model\nwith several free parameters that are required to be calibrated against\nexperimental results. The calibration is carried out by means of an\nevolutionary algorithm (differential evolution, DE) that needs to evaluate each\nconfiguration of parameters on six different damage cases, each of them taking\nseveral minutes to compute. To minimise the simulation time of the parameter\ntuning for the DE, the stretch of one unique fixed-diameter axon with a\nsimplified triggering process is used to speed up the calculations. The model\nis then leveraged for the parameter optimization of the more realistic bundle\nof independent axons, an impractical configuration to run on a single processor\ncomputer. To this end, we have developed a parallel implementation based on\nOpenMP that runs on a multi-processor taking advantage of all the available\ncomputational power. The parallel DE algorithm obtains good results,\noutperforming the best effort achieved by published manual calibration, in a\nfraction of the time. While not being able to fully capture the experimental\nresults, the resulting nerve model provides a complex averaging framework for\nnerve damage simulation able to simulate gradual axonal functional alteration\nin a bundle.\n", "date": "2024-09-19", "categories": []}
{"id": "2409.02241", "title": "What Makes a Face Look like a Hat: Decoupling Low-level and High-level\n  Visual Properties with Image Triplets", "abstract": "  In visual decision making, high-level features, such as object categories,\nhave a strong influence on choice. However, the impact of low-level features on\nbehavior is less understood partly due to the high correlation between high-\nand low-level features in the stimuli presented (e.g., objects of the same\ncategory are more likely to share low-level features). To disentangle these\neffects, we propose a method that de-correlates low- and high-level visual\nproperties in a novel set of stimuli. Our method uses two Convolutional Neural\nNetworks (CNNs) as candidate models of the ventral visual stream: the CORnet-S\nthat has high neural predictivity in high-level, IT-like responses and the\nVGG-16 that has high neural predictivity in low-level responses. Triplets\n(root, image1, image2) of stimuli are parametrized by the level of low- and\nhigh-level similarity of images extracted from the different layers. These\nstimuli are then used in a decision-making task where participants are tasked\nto choose the most similar-to-the-root image. We found that different networks\nshow differing abilities to predict the effects of low-versus-high-level\nsimilarity: while CORnet-S outperforms VGG-16 in explaining human choices based\non high-level similarity, VGG-16 outperforms CORnet-S in explaining human\nchoices based on low-level similarity. Using Brain-Score, we observed that the\nbehavioral prediction abilities of different layers of these networks\nqualitatively corresponded to their ability to explain neural activity at\ndifferent levels of the visual hierarchy. In summary, our algorithm for\nstimulus set generation enables the study of how different representations in\nthe visual stream affect high-level cognitive behaviors.\n", "date": "2024-09-03", "categories": []}
{"id": "2409.05279", "title": "BrainDecoder: Style-Based Visual Decoding of EEG Signals", "abstract": "  Decoding neural representations of visual stimuli from electroencephalography\n(EEG) offers valuable insights into brain activity and cognition. Recent\nadvancements in deep learning have significantly enhanced the field of visual\ndecoding of EEG, primarily focusing on reconstructing the semantic content of\nvisual stimuli. In this paper, we present a novel visual decoding pipeline\nthat, in addition to recovering the content, emphasizes the reconstruction of\nthe style, such as color and texture, of images viewed by the subject. Unlike\nprevious methods, this ``style-based'' approach learns in the CLIP spaces of\nimage and text separately, facilitating a more nuanced extraction of\ninformation from EEG signals. We also use captions for text alignment simpler\nthan previously employed, which we find work better. Both quantitative and\nqualitative evaluations show that our method better preserves the style of\nvisual stimuli and extracts more fine-grained semantic information from neural\nsignals. Notably, it achieves significant improvements in quantitative results\nand sets a new state-of-the-art on the popular Brain2Image dataset.\n", "date": "2024-09-09", "categories": []}
{"id": "2409.02327", "title": "Generative Principal Component Regression via Variational Inference", "abstract": "  The ability to manipulate complex systems, such as the brain, to modify\nspecific outcomes has far-reaching implications, particularly in the treatment\nof psychiatric disorders. One approach to designing appropriate manipulations\nis to target key features of predictive models. While generative latent\nvariable models, such as probabilistic principal component analysis (PPCA), is\na powerful tool for identifying targets, they struggle incorporating\ninformation relevant to low-variance outcomes into the latent space. When\nstimulation targets are designed on the latent space in such a scenario, the\nintervention can be suboptimal with minimal efficacy. To address this problem,\nwe develop a novel objective based on supervised variational autoencoders\n(SVAEs) that enforces such information is represented in the latent space. The\nnovel objective can be used with linear models, such as PPCA, which we refer to\nas generative principal component regression (gPCR). We show in simulations\nthat gPCR dramatically improves target selection in manipulation as compared to\nstandard PCR and SVAEs. As part of these simulations, we develop a metric for\ndetecting when relevant information is not properly incorporated into the\nloadings. We then show in two neural datasets related to stress and social\nbehavior in which gPCR dramatically outperforms PCR in predictive performance\nand that SVAEs exhibit low incorporation of relevant information into the\nloadings. Overall, this work suggests that our method significantly improves\ntarget selection for manipulation using latent variable models over competitor\ninference schemes.\n", "date": "2024-09-03", "categories": []}
{"id": "2409.02390", "title": "Neural Dynamics Model of Visual Decision-Making: Learning from Human\n  Experts", "abstract": "  Uncovering the fundamental neural correlates of biological intelligence,\ndeveloping mathematical models, and conducting computational simulations are\ncritical for advancing new paradigms in artificial intelligence (AI). In this\nstudy, we implemented a comprehensive visual decision-making model that spans\nfrom visual input to behavioral output, using a neural dynamics modeling\napproach. Drawing inspiration from the key components of the dorsal visual\npathway in primates, our model not only aligns closely with human behavior but\nalso reflects neural activities in primates, and achieving accuracy comparable\nto convolutional neural networks (CNNs). Moreover, magnetic resonance imaging\n(MRI) identified key neuroimaging features such as structural connections and\nfunctional connectivity that are associated with performance in perceptual\ndecision-making tasks. A neuroimaging-informed fine-tuning approach was\nintroduced and applied to the model, leading to performance improvements that\nparalleled the behavioral variations observed among subjects. Compared to\nclassical deep learning models, our model more accurately replicates the\nbehavioral performance of biological intelligence, relying on the structural\ncharacteristics of biological neural networks rather than extensive training\ndata, and demonstrating enhanced resilience to perturbation.\n", "date": "2024-09-04", "categories": []}
{"id": "2409.02709", "title": "How does the brain compute with probabilities?", "abstract": "  This perspective piece is the result of a Generative Adversarial\nCollaboration (GAC) tackling the question `How does neural activity represent\nprobability distributions?'. We have addressed three major obstacles to\nprogress on answering this question: first, we provide a unified language for\ndefining competing hypotheses. Second, we explain the fundamentals of three\nprominent proposals for probabilistic computations -- Probabilistic Population\nCodes (PPCs), Distributed Distributional Codes (DDCs), and Neural Sampling\nCodes (NSCs) -- and describe similarities and differences in that common\nlanguage. Third, we review key empirical data previously taken as evidence for\nat least one of these proposal, and describe how it may or may not be\nexplainable by alternative proposals. Finally, we describe some key challenges\nin resolving the debate, and propose potential directions to address them\nthrough a combination of theory and experiments.\n", "date": "2024-09-01", "categories": []}
{"id": "2409.01762", "title": "Decoding finger velocity from cortical spike trains with recurrent\n  spiking neural networks", "abstract": "  Invasive cortical brain-machine interfaces (BMIs) can significantly improve\nthe life quality of motor-impaired patients. Nonetheless, externally mounted\npedestals pose an infection risk, which calls for fully implanted systems. Such\nsystems, however, must meet strict latency and energy constraints while\nproviding reliable decoding performance. While recurrent spiking neural\nnetworks (RSNNs) are ideally suited for ultra-low-power, low-latency processing\non neuromorphic hardware, it is unclear whether they meet the above\nrequirements. To address this question, we trained RSNNs to decode finger\nvelocity from cortical spike trains (CSTs) of two macaque monkeys. First, we\nfound that a large RSNN model outperformed existing feedforward spiking neural\nnetworks (SNNs) and artificial neural networks (ANNs) in terms of their\ndecoding accuracy. We next developed a tiny RSNN with a smaller memory\nfootprint, low firing rates, and sparse connectivity. Despite its reduced\ncomputational requirements, the resulting model performed substantially better\nthan existing SNN and ANN decoders. Our results thus demonstrate that RSNNs\noffer competitive CST decoding performance under tight resource constraints and\nare promising candidates for fully implanted ultra-low-power BMIs with the\npotential to revolutionize patient care.\n", "date": "2024-09-03", "categories": []}
{"id": "2408.13112", "title": "An Introduction to Cognidynamics", "abstract": "  This paper gives an introduction to \\textit{Cognidynamics}, that is to the\ndynamics of cognitive systems driven by optimal objectives imposed over time\nwhen they interact either with a defined virtual or with a real-world\nenvironment. The proposed theory is developed in the general framework of\ndynamic programming which leads to think of computational laws dictated by\nclassic Hamiltonian equations. Those equations lead to the formulation of a\nneural propagation scheme in cognitive agents modeled by dynamic neural\nnetworks which exhibits locality in both space and time, thus contributing the\nlongstanding debate on biological plausibility of learning algorithms like\nBackpropagation. We interpret the learning process in terms of energy exchange\nwith the environment and show the crucial role of energy dissipation and its\nlinks with focus of attention mechanisms and conscious behavior.\n", "date": "2024-08-18", "categories": []}
{"id": "2408.11948", "title": "Topological Representational Similarity Analysis in Brains and Beyond", "abstract": "  Understanding how the brain represents and processes information is crucial\nfor advancing neuroscience and artificial intelligence. Representational\nsimilarity analysis (RSA) has been instrumental in characterizing neural\nrepresentations, but traditional RSA relies solely on geometric properties,\noverlooking crucial topological information. This thesis introduces Topological\nRSA (tRSA), a novel framework combining geometric and topological properties of\nneural representations.\n  tRSA applies nonlinear monotonic transforms to representational\ndissimilarities, emphasizing local topology while retaining intermediate-scale\ngeometry. The resulting geo-topological matrices enable model comparisons\nrobust to noise and individual idiosyncrasies. This thesis introduces several\nkey methodological advances: (1) Topological RSA (tRSA) for identifying\ncomputational signatures and testing topological hypotheses; (2) Adaptive\nGeo-Topological Dependence Measure (AGTDM) for detecting complex multivariate\nrelationships; (3) Procrustes-aligned Multidimensional Scaling (pMDS) for\nrevealing neural computation stages; (4) Temporal Topological Data Analysis\n(tTDA) for uncovering developmental trajectories; and (5) Single-cell\nTopological Simplicial Analysis (scTSA) for characterizing cell population\ncomplexity.\n  Through analyses of neural recordings, biological data, and neural network\nsimulations, this thesis demonstrates the power and versatility of these\nmethods in understanding brains, computational models, and complex biological\nsystems. They not only offer robust approaches for adjudicating among competing\nmodels but also reveal novel theoretical insights into the nature of neural\ncomputation. This work lays the foundation for future investigations at the\nintersection of topology, neuroscience, and time series analysis, paving the\nway for more nuanced understanding of brain function and dysfunction.\n", "date": "2024-08-21", "categories": []}
{"id": "2408.11019", "title": "An Overlooked Role of Context-Sensitive Dendrites", "abstract": "  To date, most dendritic studies have predominantly focused on the apical zone\nof pyramidal two-point neurons (TPNs) receiving only feedback (FB) connections\nfrom higher perceptual layers and using them for learning. Recent cellular\nneurophysiology and computational neuroscience studies suggests that the apical\ninput (context), coming from feedback and lateral connections, is multifaceted\nand far more diverse, with greater implications for ongoing learning and\nprocessing in the brain than previously realized. In addition to the FB, the\napical tuft receives signals from neighboring cells of the same network as\nproximal (P) context, other parts of the brain as distal (D) context, and\noverall coherent information across the network as universal (U) context. The\nintegrated context (C) amplifies and suppresses the transmission of coherent\nand conflicting feedforward (FF) signals, respectively. Specifically, we show\nthat complex context-sensitive (CS)-TPNs flexibly integrate C moment-by-moment\nwith the FF somatic current at the soma such that the somatic current is\namplified when both feedforward (FF) and C are coherent; otherwise, it is\nattenuated. This generates the event only when the FF and C currents are\ncoherent, which is then translated into a singlet or a burst based on the FB\ninformation. Spiking simulation results show that this flexible integration of\nsomatic and contextual currents enables the propagation of more coherent\nsignals (bursts), making learning faster with fewer neurons. Similar behavior\nis observed when this functioning is used in conventional artificial networks,\nwhere orders of magnitude fewer neurons are required to process vast amounts of\nheterogeneous real-world audio-visual (AV) data trained using backpropagation\n(BP). The computational findings presented here demonstrate the universality of\nCS-TPNs, suggesting a dendritic narrative that was previously overlooked.\n", "date": "2024-08-20", "categories": []}
{"id": "2312.09430", "title": "Deep Representation Learning for Open Vocabulary\n  Electroencephalography-to-Text Decoding", "abstract": "  Previous research has demonstrated the potential of using pre-trained\nlanguage models for decoding open vocabulary Electroencephalography (EEG)\nsignals captured through a non-invasive Brain-Computer Interface (BCI).\nHowever, the impact of embedding EEG signals in the context of language models\nand the effect of subjectivity, remain unexplored, leading to uncertainty about\nthe best approach to enhance decoding performance. Additionally, current\nevaluation metrics used to assess decoding effectiveness are predominantly\nsyntactic and do not provide insights into the comprehensibility of the decoded\noutput for human understanding. We present an end-to-end deep learning\nframework for non-invasive brain recordings that brings modern representational\nlearning approaches to neuroscience. Our proposal introduces the following\ninnovations: 1) an end-to-end deep learning architecture for open vocabulary\nEEG decoding, incorporating a subject-dependent representation learning module\nfor raw EEG encoding, a BART language model, and a GPT-4 sentence refinement\nmodule; 2) a more comprehensive sentence-level evaluation metric based on the\nBERTScore; 3) an ablation study that analyses the contributions of each module\nwithin our proposal, providing valuable insights for future research. We\nevaluate our approach on two publicly available datasets, ZuCo v1.0 and v2.0,\ncomprising EEG recordings of 30 subjects engaged in natural reading tasks. Our\nmodel achieves a BLEU-1 score of 42.75%, a ROUGE-1-F of 33.28%, and a\nBERTScore-F of 53.86%, outperforming the previous state-of-the-art methods by\n3.38%, 8.43%, and 6.31%, respectively.\n", "date": "2023-11-15", "categories": []}
{"id": "2408.03062", "title": "Analysis of Argument Structure Constructions in a Deep Recurrent\n  Language Model", "abstract": "  Understanding how language and linguistic constructions are processed in the\nbrain is a fundamental question in cognitive computational neuroscience. In\nthis study, we explore the representation and processing of Argument Structure\nConstructions (ASCs) in a recurrent neural language model. We trained a Long\nShort-Term Memory (LSTM) network on a custom-made dataset consisting of 2000\nsentences, generated using GPT-4, representing four distinct ASCs: transitive,\nditransitive, caused-motion, and resultative constructions.\n  We analyzed the internal activations of the LSTM model's hidden layers using\nMultidimensional Scaling (MDS) and t-Distributed Stochastic Neighbor Embedding\n(t-SNE) to visualize the sentence representations. The Generalized\nDiscrimination Value (GDV) was calculated to quantify the degree of clustering\nwithin these representations. Our results show that sentence representations\nform distinct clusters corresponding to the four ASCs across all hidden layers,\nwith the most pronounced clustering observed in the last hidden layer before\nthe output layer. This indicates that even a relatively simple,\nbrain-constrained recurrent neural network can effectively differentiate\nbetween various construction types.\n  These findings are consistent with previous studies demonstrating the\nemergence of word class and syntax rule representations in recurrent language\nmodels trained on next word prediction tasks. In future work, we aim to\nvalidate these results using larger language models and compare them with\nneuroimaging data obtained during continuous speech perception. This study\nhighlights the potential of recurrent neural language models to mirror\nlinguistic processing in the human brain, providing valuable insights into the\ncomputational and neural mechanisms underlying language understanding.\n", "date": "2024-08-06", "categories": []}
{"id": "2408.01528", "title": "Can multivariate Granger causality detect directed connectivity of a\n  multistable and dynamic biological decision network model?", "abstract": "  Extracting causal connections can advance interpretable AI and machine\nlearning. Granger causality (GC) is a robust statistical method for estimating\ndirected influences (DC) between signals. While GC has been widely applied to\nanalysing neuronal signals in biological neural networks and other domains, its\napplication to complex, nonlinear, and multistable neural networks is less\nexplored. In this study, we applied time-domain multi-variate Granger causality\n(MVGC) to the time series neural activity of all nodes in a trained multistable\nbiologically based decision neural network model with real-time decision\nuncertainty monitoring. Our analysis demonstrated that challenging two-choice\ndecisions, where input signals could be closely matched, and the appropriate\napplication of fine-grained sliding time windows, could readily reveal the\noriginal model's DC. Furthermore, the identified DC varied based on whether the\nnetwork had correct or error decisions. Integrating the identified DC from\ndifferent decision outcomes recovered most of the original model's\narchitecture, despite some spurious and missing connectivity. This approach\ncould be used as an initial exploration to enhance the interpretability and\ntransparency of dynamic multistable and nonlinear biological or AI systems by\nrevealing causal connections throughout different phases of neural network\ndynamics and outcomes.\n", "date": "2024-08-02", "categories": []}
{"id": "2408.01316", "title": "Synergistic pathways of modulation enable robust task packing within\n  neural dynamics", "abstract": "  Understanding how brain networks learn and manage multiple tasks\nsimultaneously is of interest in both neuroscience and artificial intelligence.\nIn this regard, a recent research thread in theoretical neuroscience has\nfocused on how recurrent neural network models and their internal dynamics\nenact multi-task learning. To manage different tasks requires a mechanism to\nconvey information about task identity or context into the model, which from a\nbiological perspective may involve mechanisms of neuromodulation. In this\nstudy, we use recurrent network models to probe the distinctions between two\nforms of contextual modulation of neural dynamics, at the level of neuronal\nexcitability and at the level of synaptic strength. We characterize these\nmechanisms in terms of their functional outcomes, focusing on their robustness\nto context ambiguity and, relatedly, their efficiency with respect to packing\nmultiple tasks into finite size networks. We also demonstrate distinction\nbetween these mechanisms at the level of the neuronal dynamics they induce.\nTogether, these characterizations indicate complementarity and synergy in how\nthese mechanisms act, potentially over multiple time-scales, toward enhancing\nrobustness of multi-task learning.\n", "date": "2024-08-02", "categories": []}
{"id": "2408.00525", "title": "Identifying the Hierarchical Emotional Areas in the Human Brain Through\n  Information Fusion", "abstract": "  The brain basis of emotion has consistently received widespread attention,\nattracting a large number of studies to explore this cutting-edge topic.\nHowever, the methods employed in these studies typically only model the\npairwise relationship between two brain regions, while neglecting the\ninteractions and information fusion among multiple brain\nregions$\\unicode{x2014}$one of the key ideas of the psychological\nconstructionist hypothesis. To overcome the limitations of traditional methods,\nthis study provides an in-depth theoretical analysis of how to maximize\ninteractions and information fusion among brain regions. Building on the\nresults of this analysis, we propose to identify the hierarchical emotional\nareas in the human brain through multi-source information fusion and graph\nmachine learning methods. Comprehensive experiments reveal that the identified\nhierarchical emotional areas, from lower to higher levels, primarily facilitate\nthe fundamental process of emotion perception, the construction of basic\npsychological operations, and the coordination and integration of these\noperations. Overall, our findings provide unique insights into the brain\nmechanisms underlying specific emotions based on the psychological\nconstructionist hypothesis.\n", "date": "2024-08-01", "categories": []}
{"id": "2406.19884", "title": "Investigating the Timescales of Language Processing with EEG and\n  Language Models", "abstract": "  This study explores the temporal dynamics of language processing by examining\nthe alignment between word representations from a pre-trained transformer-based\nlanguage model, and EEG data. Using a Temporal Response Function (TRF) model,\nwe investigate how neural activity corresponds to model representations across\ndifferent layers, revealing insights into the interaction between artificial\nlanguage models and brain responses during language comprehension. Our analysis\nreveals patterns in TRFs from distinct layers, highlighting varying\ncontributions to lexical and compositional processing. Additionally, we used\nlinear discriminant analysis (LDA) to isolate part-of-speech (POS)\nrepresentations, offering insights into their influence on neural responses and\nthe underlying mechanisms of syntactic processing. These findings underscore\nEEG's utility for probing language processing dynamics with high temporal\nresolution. By bridging artificial language models and neural activity, this\nstudy advances our understanding of their interaction at fine timescales.\n", "date": "2024-06-28", "categories": []}
{"id": "2407.21195", "title": "Diffusion-Based Generation of Neural Activity from Disentangled Latent\n  Codes", "abstract": "  Recent advances in recording technology have allowed neuroscientists to\nmonitor activity from thousands of neurons simultaneously. Latent variable\nmodels are increasingly valuable for distilling these recordings into compact\nand interpretable representations. Here we propose a new approach to neural\ndata analysis that leverages advances in conditional generative modeling to\nenable the unsupervised inference of disentangled behavioral variables from\nrecorded neural activity. Our approach builds on InfoDiffusion, which augments\ndiffusion models with a set of latent variables that capture important factors\nof variation in the data. We apply our model, called Generating Neural\nObservations Conditioned on Codes with High Information (GNOCCHI), to time\nseries neural data and test its application to synthetic and biological\nrecordings of neural activity during reaching. In comparison to a VAE-based\nsequential autoencoder, GNOCCHI learns higher-quality latent spaces that are\nmore clearly structured and more disentangled with respect to key behavioral\nvariables. These properties enable accurate generation of novel samples (unseen\nbehavioral conditions) through simple linear traversal of the latent spaces\nproduced by GNOCCHI. Our work demonstrates the potential of unsupervised,\ninformation-based models for the discovery of interpretable latent spaces from\nneural data, enabling researchers to generate high-quality samples from unseen\nconditions.\n", "date": "2024-07-30", "categories": []}
{"id": "2407.20622", "title": "Decoding Linguistic Representations of Human Brain", "abstract": "  Language, as an information medium created by advanced organisms, has always\nbeen a concern of neuroscience regarding how it is represented in the brain.\nDecoding linguistic representations in the evoked brain has shown\ngroundbreaking achievements, thanks to the rapid improvement of neuroimaging,\nmedical technology, life sciences and artificial intelligence. In this work, we\npresent a taxonomy of brain-to-language decoding of both textual and speech\nformats. This work integrates two types of research: neuroscience focusing on\nlanguage understanding and deep learning-based brain decoding. Generating\ndiscernible language information from brain activity could not only help those\nwith limited articulation, especially amyotrophic lateral sclerosis (ALS)\npatients but also open up a new way for the next generation's brain-computer\ninterface (BCI). This article will help brain scientists and deep-learning\nresearchers to gain a bird's eye view of fine-grained language perception, and\nthus facilitate their further investigation and research of neural process and\nlanguage decoding.\n", "date": "2024-07-30", "categories": []}
{"id": "2407.18413", "title": "Simulation of Neural Responses to Classical Music Using Organoid\n  Intelligence Methods", "abstract": "  Music is a complex auditory stimulus capable of eliciting significant changes\nin brain activity, influencing cognitive processes such as memory, attention,\nand emotional regulation. However, the underlying mechanisms of music-induced\ncognitive processes remain largely unknown. Organoid intelligence and deep\nlearning models show promise for simulating and analyzing these neural\nresponses to classical music, an area significantly unexplored in computational\nneuroscience. Hence, we present the PyOrganoid library, an innovative tool that\nfacilitates the simulation of organoid learning models, integrating\nsophisticated machine learning techniques with biologically inspired organoid\nsimulations. Our study features the development of the Pianoid model, a \"deep\norganoid learning\" model that utilizes a Bidirectional LSTM network to predict\nEEG responses based on audio features from classical music recordings. This\nmodel demonstrates the feasibility of using computational methods to replicate\ncomplex neural processes, providing valuable insights into music perception and\ncognition. Likewise, our findings emphasize the utility of synthetic models in\nneuroscience research and highlight the PyOrganoid library's potential as a\nversatile tool for advancing studies in neuroscience and artificial\nintelligence.\n", "date": "2024-07-25", "categories": []}
{"id": "2402.14213", "title": "Contrastive Learning of Shared Spatiotemporal EEG Representations Across\n  Individuals for Naturalistic Neuroscience", "abstract": "  Neural representations induced by naturalistic stimuli offer insights into\nhow humans respond to stimuli in daily life. Understanding neural mechanisms\nunderlying naturalistic stimuli processing hinges on the precise identification\nand extraction of the shared neural patterns that are consistently present\nacross individuals. Targeting the Electroencephalogram (EEG) technique, known\nfor its rich spatial and temporal information, this study presents a framework\nfor Contrastive Learning of Shared SpatioTemporal EEG Representations across\nindividuals (CL-SSTER). CL-SSTER utilizes contrastive learning to maximize the\nsimilarity of EEG representations across individuals for identical stimuli,\ncontrasting with those for varied stimuli. The network employed spatial and\ntemporal convolutions to simultaneously learn the spatial and temporal patterns\ninherent in EEG. The versatility of CL-SSTER was demonstrated on three EEG\ndatasets, including a synthetic dataset, a natural speech comprehension EEG\ndataset, and an emotional video watching EEG dataset. CL-SSTER attained the\nhighest inter-subject correlation (ISC) values compared to the state-of-the-art\nISC methods. The latent representations generated by CL-SSTER exhibited\nreliable spatiotemporal EEG patterns, which can be explained by properties of\nthe naturalistic stimuli. CL-SSTER serves as an interpretable and scalable\nframework for the identification of inter-subject shared neural representations\nin naturalistic neuroscience.\n", "date": "2024-02-22", "categories": []}
{"id": "2403.04884", "title": "Optimizing Retinal Prosthetic Stimuli with Conditional Invertible Neural\n  Networks", "abstract": "  Implantable retinal prostheses offer a promising solution to restore partial\nvision by circumventing damaged photoreceptor cells in the retina and directly\nstimulating the remaining functional retinal cells. However, the information\ntransmission between the camera and retinal cells is often limited by the low\nresolution of the electrode array and the lack of specificity for different\nganglion cell types, resulting in suboptimal stimulations. In this work, we\npropose to utilize normalizing flow-based conditional invertible neural\nnetworks to optimize retinal implant stimulation in an unsupervised manner. The\ninvertibility of these networks allows us to use them as a surrogate for the\ncomputational model of the visual system, while also encoding input camera\nsignals into optimized electrical stimuli on the electrode array. Compared to\nother methods, such as trivial downsampling, linear models, and feed-forward\nconvolutional neural networks, the flow-based invertible neural network and its\nconditional extension yield better visual reconstruction qualities w.r.t.\nvarious metrics using a physiologically validated simulation tool.\n", "date": "2024-03-07", "categories": []}
{"id": "2407.09508", "title": "Focused State Recognition Using EEG with Eye Movement-Assisted\n  Annotation", "abstract": "  With the rapid advancement in machine learning, the recognition and analysis\nof brain activity based on EEG and eye movement signals have attained a high\nlevel of sophistication. Utilizing deep learning models for learning EEG and\neye movement features proves effective in classifying brain activities. A\nfocused state indicates intense concentration on a task or thought.\nDistinguishing focused and unfocused states can be achieved through eye\nmovement behaviors, reflecting variations in brain activities. By calculating\nbinocular focusing point disparity in eye movement signals and integrating\nrelevant EEG features, we propose an annotation method for focused states. The\nresulting comprehensive dataset, derived from raw data processed through a\nbio-acquisition device, includes both EEG features and focused labels annotated\nby eye movements. Extensive training and testing on several deep learning\nmodels, particularly the Transformer, yielded a 90.16% accuracy on the\nsubject-dependent experiments. The validity of this approach was demonstrated,\nwith cross-subject experiments, key frequency band and brain region analyses\nconfirming its generalizability and providing physiological explanations.\n", "date": "2024-06-15", "categories": []}
{"id": "2407.10414", "title": "Teaching CORnet Human fMRI Representations for Enhanced Model-Brain\n  Alignment", "abstract": "  Deep convolutional neural networks (DCNNs) have demonstrated excellent\nperformance in object recognition and have been found to share some\nsimilarities with brain visual processing. However, the substantial gap between\nDCNNs and human visual perception still exists. Functional magnetic resonance\nimaging (fMRI) as a widely used technique in cognitive neuroscience can record\nneural activation in the human visual cortex during the process of visual\nperception. Can we teach DCNNs human fMRI signals to achieve a more brain-like\nmodel? To answer this question, this study proposed ReAlnet-fMRI, a model based\non the SOTA vision model CORnet but optimized using human fMRI data through a\nmulti-layer encoding-based alignment framework. This framework has been shown\nto effectively enable the model to learn human brain representations. The\nfMRI-optimized ReAlnet-fMRI exhibited higher similarity to the human brain than\nboth CORnet and the control model in within-and across-subject as well as\nwithin- and across-modality model-brain (fMRI and EEG) alignment evaluations.\nAdditionally, we conducted an in-depth analyses to investigate how the internal\nrepresentations of ReAlnet-fMRI differ from CORnet in encoding various object\ndimensions. These findings provide the possibility of enhancing the\nbrain-likeness of visual models by integrating human neural data, helping to\nbridge the gap between computer vision and visual neuroscience.\n", "date": "2024-07-15", "categories": []}
{"id": "2407.10737", "title": "Aligning Neuronal Coding of Dynamic Visual Scenes with Foundation Vision\n  Models", "abstract": "  Our brains represent the ever-changing environment with neurons in a highly\ndynamic fashion. The temporal features of visual pixels in dynamic natural\nscenes are entrapped in the neuronal responses of the retina. It is crucial to\nestablish the intrinsic temporal relationship between visual pixels and\nneuronal responses. Recent foundation vision models have paved an advanced way\nof understanding image pixels. Yet, neuronal coding in the brain largely lacks\na deep understanding of its alignment with pixels. Most previous studies employ\nstatic images or artificial videos derived from static images for emulating\nmore real and complicated stimuli. Despite these simple scenarios effectively\nhelp to separate key factors influencing visual coding, complex temporal\nrelationships receive no consideration. To decompose the temporal features of\nvisual coding in natural scenes, here we propose Vi-ST, a spatiotemporal\nconvolutional neural network fed with a self-supervised Vision Transformer\n(ViT) prior, aimed at unraveling the temporal-based encoding patterns of\nretinal neuronal populations. The model demonstrates robust predictive\nperformance in generalization tests. Furthermore, through detailed ablation\nexperiments, we demonstrate the significance of each temporal module.\nFurthermore, we introduce a visual coding evaluation metric designed to\nintegrate temporal considerations and compare the impact of different numbers\nof neuronal populations on complementary coding. In conclusion, our proposed\nVi-ST demonstrates a novel modeling framework for neuronal coding of dynamic\nvisual scenes in the brain, effectively aligning our brain representation of\nvideo with neuronal activity. The code is available at\nhttps://github.com/wurining/Vi-ST.\n", "date": "2024-07-15", "categories": []}
{"id": "2403.04325", "title": "Measuring Meaning Composition in the Human Brain with Composition Scores\n  from Large Language Models", "abstract": "  The process of meaning composition, wherein smaller units like morphemes or\nwords combine to form the meaning of phrases and sentences, is essential for\nhuman sentence comprehension. Despite extensive neurolinguistic research into\nthe brain regions involved in meaning composition, a computational metric to\nquantify the extent of composition is still lacking. Drawing on the key-value\nmemory interpretation of transformer feed-forward network blocks, we introduce\nthe Composition Score, a novel model-based metric designed to quantify the\ndegree of meaning composition during sentence comprehension. Experimental\nfindings show that this metric correlates with brain clusters associated with\nword frequency, structural processing, and general sensitivity to words,\nsuggesting the multifaceted nature of meaning composition during human sentence\ncomprehension.\n", "date": "2024-03-07", "categories": []}
{"id": "2402.14982", "title": "Human Brain Exhibits Distinct Patterns When Listening to Fake Versus\n  Real Audio: Preliminary Evidence", "abstract": "  In this paper we study the variations in human brain activity when listening\nto real and fake audio. Our preliminary results suggest that the\nrepresentations learned by a state-of-the-art deepfake audio detection\nalgorithm, do not exhibit clear distinct patterns between real and fake audio.\nIn contrast, human brain activity, as measured by EEG, displays distinct\npatterns when individuals are exposed to fake versus real audio. This\npreliminary evidence enables future research directions in areas such as\ndeepfake audio detection.\n", "date": "2024-02-22", "categories": []}
{"id": "2209.11737", "title": "Visual representations in the human brain are aligned with large\n  language models", "abstract": "  The human brain extracts complex information from visual inputs, including\nobjects, their spatial and semantic interrelations, and their interactions with\nthe environment. However, a quantitative approach for studying this information\nremains elusive. Here, we test whether the contextual information encoded in\nlarge language models (LLMs) is beneficial for modelling the complex visual\ninformation extracted by the brain from natural scenes. We show that LLM\nembeddings of scene captions successfully characterise brain activity evoked by\nviewing the natural scenes. This mapping captures selectivities of different\nbrain areas, and is sufficiently robust that accurate scene captions can be\nreconstructed from brain activity. Using carefully controlled model\ncomparisons, we then proceed to show that the accuracy with which LLM\nrepresentations match brain representations derives from the ability of LLMs to\nintegrate complex information contained in scene captions beyond that conveyed\nby individual words. Finally, we train deep neural network models to transform\nimage inputs into LLM representations. Remarkably, these networks learn\nrepresentations that are better aligned with brain representations than a large\nnumber of state-of-the-art alternative models, despite being trained on\norders-of-magnitude less data. Overall, our results suggest that LLM embeddings\nof scene captions provide a representational format that accounts for complex\ninformation extracted by the brain from visual inputs.\n", "date": "2022-09-23", "categories": []}
{"id": "2407.04816", "title": "Object recognition in primates: What can early visual areas contribute?", "abstract": "  If neuroscientists were asked which brain area is responsible for object\nrecognition in primates, most would probably answer infero-temporal (IT)\ncortex. While IT is likely responsible for fine discriminations, and it is\naccordingly dominated by foveal visual inputs, there is more to object\nrecognition than fine discrimination. Importantly, foveation of an object of\ninterest usually requires recognizing, with reasonable confidence, its presence\nin the periphery. Arguably, IT plays a secondary role in such peripheral\nrecognition, and other visual areas might instead be more critical. To\ninvestigate how signals carried by early visual processing areas (such as LGN\nand V1) could be used for object recognition in the periphery, we focused here\non the task of distinguishing faces from non-faces. We tested how sensitive\nvarious models were to nuisance parameters, such as changes in scale and\norientation of the image, and the type of image background. We found that a\nmodel of V1 simple or complex cells could provide quite reliable information,\nresulting in performance better than 80% in realistic scenarios. An LGN model\nperformed considerably worse. Because peripheral recognition is both crucial to\nenable fine recognition (by bringing an object of interest on the fovea), and\nprobably sufficient to account for a considerable fraction of our daily\nrecognition-guided behavior, we think that the current focus on area IT and\nfoveal processing is too narrow. We propose that rather than a hierarchical\nsystem with IT-like properties as its primary aim, object recognition should be\nseen as a parallel process, with high-accuracy foveal modules operating in\nparallel with lower-accuracy and faster modules that can operate across the\nvisual field.\n", "date": "2024-07-05", "categories": []}
{"id": "2407.01884", "title": "EIT-1M: One Million EEG-Image-Text Pairs for Human Visual-textual\n  Recognition and More", "abstract": "  Recently, electroencephalography (EEG) signals have been actively\nincorporated to decode brain activity to visual or textual stimuli and achieve\nobject recognition in multi-modal AI. Accordingly, endeavors have been focused\non building EEG-based datasets from visual or textual single-modal stimuli.\nHowever, these datasets offer limited EEG epochs per category, and the complex\nsemantics of stimuli presented to participants compromise their quality and\nfidelity in capturing precise brain activity. The study in neuroscience unveils\nthat the relationship between visual and textual stimulus in EEG recordings\nprovides valuable insights into the brain's ability to process and integrate\nmulti-modal information simultaneously. Inspired by this, we propose a novel\nlarge-scale multi-modal dataset, named EIT-1M, with over 1 million\nEEG-image-text pairs. Our dataset is superior in its capacity of reflecting\nbrain activities in simultaneously processing multi-modal information. To\nachieve this, we collected data pairs while participants viewed alternating\nsequences of visual-textual stimuli from 60K natural images and\ncategory-specific texts. Common semantic categories are also included to elicit\nbetter reactions from participants' brains. Meanwhile, response-based stimulus\ntiming and repetition across blocks and sessions are included to ensure data\ndiversity. To verify the effectiveness of EIT-1M, we provide an in-depth\nanalysis of EEG data captured from multi-modal stimuli across different\ncategories and participants, along with data quality scores for transparency.\nWe demonstrate its validity on two tasks: 1) EEG recognition from visual or\ntextual stimuli or both and 2) EEG-to-visual generation.\n", "date": "2024-07-02", "categories": []}
{"id": "2407.00824", "title": "A data-driven approach to modeling brain activity using differential\n  equations", "abstract": "  This research focuses on an innovative task of extracting equations from\nincomplete data, moving away from traditional methods used for complete\nsolutions. The study addresses the challenge of extracting equations from data,\nparticularly in the study of brain activity using electrophysiological data,\nwhich is often limited by insufficient information. The study provides a brief\nreview of existing open-source equation derivation approaches in the context of\nmodeling brain activity. The section below introduces a novel algorithm that\nemploys incomplete data and prior domain knowledge to recover differential\nequations. The algorithm's practicality in real-world scenarios is demonstrated\nthrough its application on both synthetic and real datasets.\n", "date": "2024-04-14", "categories": []}
{"id": "2406.16910", "title": "Mind's Eye: Image Recognition by EEG via Multimodal Similarity-Keeping\n  Contrastive Learning", "abstract": "  Decoding images from non-invasive electroencephalographic (EEG) signals has\nbeen a grand challenge in understanding how the human brain process visual\ninformation in real-world scenarios. To cope with the issues of signal-to-noise\nratio and nonstationarity, this paper introduces a MUltimodal\nSimilarity-keeping contrastivE learning (MUSE) framework for zero-shot\nEEG-based image classification. We develop a series of multivariate time-series\nencoders tailored for EEG signals and assess the efficacy of regularized\ncontrastive EEG-Image pretraining using an extensive visual EEG dataset. Our\nmethod achieves state-of-the-art performance, with a top-1 accuracy of 19.3%\nand a top-5 accuracy of 48.8% in 200-way zero-shot image classification.\nFurthermore, we visualize neural patterns via model interpretation, shedding\nlight on the visual processing dynamics in the human brain. The code repository\nfor this work is available at: https://github.com/ChiShengChen/MUSE_EEG.\n", "date": "2024-06-05", "categories": []}
{"id": "2406.15488", "title": "Orangutan: A Multiscale Brain Emulation-Based Artificial Intelligence\n  Framework for Dynamic Environments", "abstract": "  Achieving General Artificial Intelligence (AGI) has long been a grand\nchallenge in the field of AI, and brain-inspired computing is widely\nacknowledged as one of the most promising approaches to realize this goal. This\npaper introduces a novel brain-inspired AI framework, Orangutan. It simulates\nthe structure and computational mechanisms of biological brains on multiple\nscales, encompassing multi-compartment neuron architectures, diverse synaptic\nconnection modalities, neural microcircuits, cortical columns, and brain\nregions, as well as biochemical processes including facilitation, feedforward\ninhibition, short-term potentiation, and short-term depression, all grounded in\nsolid neuroscience. Building upon these highly integrated brain-like\nmechanisms, I have developed a sensorimotor model that simulates human saccadic\neye movements during object observation. The model's algorithmic efficacy was\nvalidated through testing with the observation of handwritten digit images.\n", "date": "2024-06-18", "categories": []}
{"id": "2406.01538", "title": "What Are Large Language Models Mapping to in the Brain? A Case Against\n  Over-Reliance on Brain Scores", "abstract": "  Given the remarkable capabilities of large language models (LLMs), there has\nbeen a growing interest in evaluating their similarity to the human brain. One\napproach towards quantifying this similarity is by measuring how well a model\npredicts neural signals, also called \"brain score\". Internal representations\nfrom LLMs achieve state-of-the-art brain scores, leading to speculation that\nthey share computational principles with human language processing. This\ninference is only valid if the subset of neural activity predicted by LLMs\nreflects core elements of language processing. Here, we question this\nassumption by analyzing three neural datasets used in an impactful study on\nLLM-to-brain mappings, with a particular focus on an fMRI dataset where\nparticipants read short passages. We first find that when using shuffled\ntrain-test splits, as done in previous studies with these datasets, a trivial\nfeature that encodes temporal autocorrelation not only outperforms LLMs but\nalso accounts for the majority of neural variance that LLMs explain. We\ntherefore use contiguous splits moving forward. Second, we explain the\nsurprisingly high brain scores of untrained LLMs by showing they do not account\nfor additional neural variance beyond two simple features: sentence length and\nsentence position. This undermines evidence used to claim that the transformer\narchitecture biases computations to be more brain-like. Third, we find that\nbrain scores of trained LLMs on this dataset can largely be explained by\nsentence length, position, and pronoun-dereferenced static word embeddings; a\nsmall, additional amount is explained by sense-specific embeddings and\ncontextual representations of sentence structure. We conclude that\nover-reliance on brain scores can lead to over-interpretations of similarity\nbetween LLMs and brains, and emphasize the importance of deconstructing what\nLLMs are mapping to in neural signals.\n", "date": "2024-06-03", "categories": []}
{"id": "2406.14358", "title": "The neural correlates of logical-mathematical symbol systems processing\n  resemble that of spatial cognition more than natural language processing", "abstract": "  The ability to manipulate logical-mathematical symbols (LMS), encompassing\ntasks such as calculation, reasoning, and programming, is a cognitive skill\narguably unique to humans. Considering the relatively recent emergence of this\nability in human evolutionary history, it has been suggested that LMS\nprocessing may build upon more fundamental cognitive systems, possibly through\nneuronal recycling. Previous studies have pinpointed two primary candidates,\nnatural language processing and spatial cognition. Existing comparisons between\nthese domains largely relied on task-level comparison, which may be confounded\nby task idiosyncrasy. The present study instead compared the neural correlates\nat the domain level with both automated meta-analysis and synthesized maps\nbased on three representative LMS tasks, reasoning, calculation, and mental\nprogramming. Our results revealed a more substantial cortical overlap between\nLMS processing and spatial cognition, in contrast to language processing.\nFurthermore, in regions activated by both spatial and language processing, the\nmultivariate activation pattern for LMS processing exhibited greater\nmultivariate similarity to spatial cognition than to language processing. A\nhierarchical clustering analysis further indicated that typical LMS tasks were\nindistinguishable from spatial cognition tasks at the neural level, suggesting\nan inherent connection between these two cognitive processes. Taken together,\nour findings support the hypothesis that spatial cognition is likely the basis\nof LMS processing, which may shed light on the limitations of large language\nmodels in logical reasoning, particularly those trained exclusively on textual\ndata without explicit emphasis on spatial content.\n", "date": "2024-06-20", "categories": []}
{"id": "2406.14481", "title": "Revealing Vision-Language Integration in the Brain with Multimodal\n  Networks", "abstract": "  We use (multi)modal deep neural networks (DNNs) to probe for sites of\nmultimodal integration in the human brain by predicting stereoencephalography\n(SEEG) recordings taken while human subjects watched movies. We operationalize\nsites of multimodal integration as regions where a multimodal vision-language\nmodel predicts recordings better than unimodal language, unimodal vision, or\nlinearly-integrated language-vision models. Our target DNN models span\ndifferent architectures (e.g., convolutional networks and transformers) and\nmultimodal training techniques (e.g., cross-attention and contrastive\nlearning). As a key enabling step, we first demonstrate that trained vision and\nlanguage models systematically outperform their randomly initialized\ncounterparts in their ability to predict SEEG signals. We then compare unimodal\nand multimodal models against one another. Because our target DNN models often\nhave different architectures, number of parameters, and training sets (possibly\nobscuring those differences attributable to integration), we carry out a\ncontrolled comparison of two models (SLIP and SimCLR), which keep all of these\nattributes the same aside from input modality. Using this approach, we identify\na sizable number of neural sites (on average 141 out of 1090 total sites or\n12.94%) and brain regions where multimodal integration seems to occur.\nAdditionally, we find that among the variants of multimodal training techniques\nwe assess, CLIP-style training is the best suited for downstream prediction of\nthe neural activity in these sites.\n", "date": "2024-06-20", "categories": []}
{"id": "2310.10531", "title": "Learning optimal integration of spatial and temporal information in\n  noisy chemotaxis", "abstract": "  We investigate the boundary between chemotaxis driven by spatial estimation\nof gradients and chemotaxis driven by temporal estimation. While it is well\nknown that spatial chemotaxis becomes disadvantageous for small organisms at\nhigh noise levels, it is unclear whether there is a discontinuous switch of\noptimal strategies or a continuous transition exists. Here, we employ deep\nreinforcement learning to study the possible integration of spatial and\ntemporal information in an a priori unconstrained manner. We parameterize such\na combined chemotactic policy by a recurrent neural network and evaluate it\nusing a minimal theoretical model of a chemotactic cell. By comparing with\nconstrained variants of the policy, we show that it converges to purely\ntemporal and spatial strategies at small and large cell sizes, respectively. We\nfind that the transition between the regimes is continuous, with the combined\nstrategy outperforming in the transition region both the constrained variants\nas well as models that explicitly integrate spatial and temporal information.\nFinally, by utilizing the attribution method of integrated gradients, we show\nthat the policy relies on a non-trivial combination of spatially and temporally\nderived gradient information in a ratio that varies dynamically during the\nchemotactic trajectories.\n", "date": "2023-10-16", "categories": []}
{"id": "2311.04664", "title": "Speech language models lack important brain-relevant semantics", "abstract": "  Despite known differences between reading and listening in the brain, recent\nwork has shown that text-based language models predict both text-evoked and\nspeech-evoked brain activity to an impressive degree. This poses the question\nof what types of information language models truly predict in the brain. We\ninvestigate this question via a direct approach, in which we systematically\nremove specific low-level stimulus features (textual, speech, and visual) from\nlanguage model representations to assess their impact on alignment with fMRI\nbrain recordings during reading and listening. Comparing these findings with\nspeech-based language models reveals starkly different effects of low-level\nfeatures on brain alignment. While text-based models show reduced alignment in\nearly sensory regions post-removal, they retain significant predictive power in\nlate language regions. In contrast, speech-based models maintain strong\nalignment in early auditory regions even after feature removal but lose all\npredictive power in late language regions. These results suggest that\nspeech-based models provide insights into additional information processed by\nearly auditory regions, but caution is needed when using them to model\nprocessing in late language regions. We make our code publicly available.\n[https://github.com/subbareddy248/speech-llm-brain]\n", "date": "2023-11-08", "categories": []}
{"id": "2406.09940", "title": "Implementing engrams from a machine learning perspective: XOR as a basic\n  motif", "abstract": "  We have previously presented the idea of how complex multimodal information\ncould be represented in our brains in a compressed form, following mechanisms\nsimilar to those employed in machine learning tools, like autoencoders. In this\nshort comment note we reflect, mainly with a didactical purpose, upon the basic\nquestion for a biological implementation: what could be the mechanism working\nas a loss function, and how it could be connected to a neuronal network\nproviding the required feedback to build a simple training configuration. We\npresent our initial ideas based on a basic motif that implements an XOR switch,\nusing few excitatory and inhibitory neurons. Such motif is guided by a\nprinciple of homeostasis, and it implements a loss function that could provide\nfeedback to other neuronal structures, establishing a control system. We\nanalyse the presence of this XOR motif in the connectome of C.Elegans, and\nindicate the relationship with the well-known lateral inhibition motif. We then\nexplore how to build a basic biological neuronal structure with learning\ncapacity integrating this XOR motif. Guided by the computational analogy, we\nshow an initial example that indicates the feasibility of this approach,\napplied to learning binary sequences, like it is the case for simple melodies.\nIn summary, we provide didactical examples exploring the parallelism between\nbiological and computational learning mechanisms, identifying basic motifs and\ntraining procedures, and how an engram encoding a melody could be built using a\nsimple recurrent network involving both excitatory and inhibitory neurons.\n", "date": "2024-06-14", "categories": []}
{"id": "2405.19277", "title": "Deep Latent Variable Modeling of Physiological Signals", "abstract": "  A deep latent variable model is a powerful method for capturing complex\ndistributions. These models assume that underlying structures, but unobserved,\nare present within the data. In this dissertation, we explore high-dimensional\nproblems related to physiological monitoring using latent variable models.\nFirst, we present a novel deep state-space model to generate electrical\nwaveforms of the heart using optically obtained signals as inputs. This can\nbring about clinical diagnoses of heart disease via simple assessment through\nwearable devices. Second, we present a brain signal modeling scheme that\ncombines the strengths of probabilistic graphical models and deep adversarial\nlearning. The structured representations can provide interpretability and\nencode inductive biases to reduce the data complexity of neural oscillations.\nThe efficacy of the learned representations is further studied in epilepsy\nseizure detection formulated as an unsupervised learning problem. Third, we\npropose a framework for the joint modeling of physiological measures and\nbehavior. Existing methods to combine multiple sources of brain data provided\nare limited. Direct analysis of the relationship between different types of\nphysiological measures usually does not involve behavioral data. Our method can\nidentify the unique and shared contributions of brain regions to behavior and\ncan be used to discover new functions of brain regions. The success of these\ninnovative computational methods would allow the translation of biomarker\nfindings across species and provide insight into neurocognitive analysis in\nnumerous biological studies and clinical diagnoses, as well as emerging\nconsumer applications.\n", "date": "2024-05-29", "categories": []}
{"id": "2406.08471", "title": "Surprise! Using Physiological Stress for Allostatic Regulation Under the\n  Active Inference Framework [Pre-Print]", "abstract": "  Allostasis proposes that long-term viability of a living system is achieved\nthrough anticipatory adjustments of its physiology and behaviour: emphasising\nphysiological and affective stress as an adaptive state of adaptation that\nminimizes long-term prediction errors. More recently, the active inference\nframework (AIF) has also sought to explain action and long-term adaptation\nthrough the minimization of future errors (free energy), through the learning\nof statistical contingencies of the world, offering a formalism for allostatic\nregulation. We suggest that framing prediction errors through the lens of\nbiological hormonal dynamics proposed by allostasis offers a way to integrate\nthese two models together in a biologically-plausible manner. In this paper, we\ndescribe our initial work in developing a model that grounds prediction errors\n(surprisal) into the secretion of a physiological stress hormone (cortisol)\nacting as an adaptive, allostatic mediator on a homeostatically-controlled\nphysiology. We evaluate this using a computational model in simulations using\nan active inference agent endowed with an artificial physiology, regulated\nthrough homeostatic and allostatic control in a stochastic environment. Our\nresults find that allostatic functions of cortisol (stress), secreted as a\nfunction of prediction errors, provide adaptive advantages to the agent's\nlong-term physiological regulation. We argue that the coupling of\ninformation-theoretic prediction errors to low-level, biological hormonal\ndynamics of stress can provide a computationally efficient model to long-term\nregulation for embodied intelligent systems.\n", "date": "2024-06-12", "categories": []}
{"id": "2406.06626", "title": "Benchmarking Neural Decoding Backbones towards Enhanced On-edge iBCI\n  Applications", "abstract": "  Traditional invasive Brain-Computer Interfaces (iBCIs) typically depend on\nneural decoding processes conducted on workstations within laboratory settings,\nwhich prevents their everyday usage. Implementing these decoding processes on\nedge devices, such as the wearables, introduces considerable challenges related\nto computational demands, processing speed, and maintaining accuracy. This\nstudy seeks to identify an optimal neural decoding backbone that boasts robust\nperformance and swift inference capabilities suitable for edge deployment. We\nexecuted a series of neural decoding experiments involving nonhuman primates\nengaged in random reaching tasks, evaluating four prospective models, Gated\nRecurrent Unit (GRU), Transformer, Receptance Weighted Key Value (RWKV), and\nSelective State Space model (Mamba), across several metrics: single-session\ndecoding, multi-session decoding, new session fine-tuning, inference speed,\ncalibration speed, and scalability. The findings indicate that although the GRU\nmodel delivers sufficient accuracy, the RWKV and Mamba models are preferable\ndue to their superior inference and calibration speeds. Additionally, RWKV and\nMamba comply with the scaling law, demonstrating improved performance with\nlarger data sets and increased model sizes, whereas GRU shows less pronounced\nscalability, and the Transformer model requires computational resources that\nscale prohibitively. This paper presents a thorough comparative analysis of the\nfour models in various scenarios. The results are pivotal in pinpointing an\noptimal backbone that can handle increasing data volumes and is viable for edge\nimplementation. This analysis provides essential insights for ongoing research\nand practical applications in the field.\n", "date": "2024-06-08", "categories": []}
{"id": "2402.02686", "title": "Multi-Region Markovian Gaussian Process: An Efficient Method to Discover\n  Directional Communications Across Multiple Brain Regions", "abstract": "  Studying the complex interactions between different brain regions is crucial\nin neuroscience. Various statistical methods have explored the latent\ncommunication across multiple brain regions. Two main categories are the\nGaussian Process (GP) and Linear Dynamical System (LDS), each with unique\nstrengths. The GP-based approach effectively discovers latent variables with\nfrequency bands and communication directions. Conversely, the LDS-based\napproach is computationally efficient but lacks powerful expressiveness in\nlatent representation. In this study, we merge both methodologies by creating\nan LDS mirroring a multi-output GP, termed Multi-Region Markovian Gaussian\nProcess (MRM-GP). Our work establishes a connection between an LDS and a\nmulti-output GP that explicitly models frequencies and phase delays within the\nlatent space of neural recordings. Consequently, the model achieves a linear\ninference cost over time points and provides an interpretable low-dimensional\nrepresentation, revealing communication directions across brain regions and\nseparating oscillatory communications into different frequency bands.\n", "date": "2024-02-05", "categories": []}
{"id": "2401.09500", "title": "MorphGrower: A Synchronized Layer-by-layer Growing Approach for\n  Plausible Neuronal Morphology Generation", "abstract": "  Neuronal morphology is essential for studying brain functioning and\nunderstanding neurodegenerative disorders. As acquiring real-world morphology\ndata is expensive, computational approaches for morphology generation have been\nstudied. Traditional methods heavily rely on expert-set rules and parameter\ntuning, making it difficult to generalize across different types of\nmorphologies. Recently, MorphVAE was introduced as the sole learning-based\nmethod, but its generated morphologies lack plausibility, i.e., they do not\nappear realistic enough and most of the generated samples are topologically\ninvalid. To fill this gap, this paper proposes MorphGrower, which mimicks the\nneuron natural growth mechanism for generation. Specifically, MorphGrower\ngenerates morphologies layer by layer, with each subsequent layer conditioned\non the previously generated structure. During each layer generation,\nMorphGrower utilizes a pair of sibling branches as the basic generation block\nand generates branch pairs synchronously. This approach ensures topological\nvalidity and allows for fine-grained generation, thereby enhancing the realism\nof the final generated morphologies. Results on four real-world datasets\ndemonstrate that MorphGrower outperforms MorphVAE by a notable margin.\nImportantly, the electrophysiological response simulation demonstrates the\nplausibility of our generated samples from a neuroscience perspective. Our code\nis available at https://github.com/Thinklab-SJTU/MorphGrower.\n", "date": "2024-01-17", "categories": []}
{"id": "2405.15036", "title": "Input-driven circuit reconfiguration in critical recurrent neural\n  networks", "abstract": "  Changing a circuit dynamically, without actually changing the hardware\nitself, is called reconfiguration, and is of great importance due to its\nmanifold technological applications. Circuit reconfiguration appears to be a\nfeature of the cerebral cortex, and hence understanding the neuroarchitectural\nand dynamical features underlying self-reconfiguration may prove key to\nelucidate brain function. We present a very simple single-layer recurrent\nnetwork, whose signal pathways can be reconfigured \"on the fly\" using only its\ninputs, with no changes to its synaptic weights. We use the low spatio-temporal\nfrequencies of the input to landscape the ongoing activity, which in turn\npermits or denies the propagation of traveling waves. This mechanism uses the\ninherent properties of dynamically-critical systems, which we guarantee through\nunitary convolution kernels. We show this network solves the classical\nconnectedness problem, by allowing signal propagation only along the regions to\nbe evaluated for connectedness and forbidding it elsewhere.\n", "date": "2024-05-23", "categories": []}
{"id": "2405.16946", "title": "Biological Neurons Compete with Deep Reinforcement Learning in Sample\n  Efficiency in a Simulated Gameworld", "abstract": "  How do biological systems and machine learning algorithms compare in the\nnumber of samples required to show significant improvements in completing a\ntask? We compared the learning efficiency of in vitro biological neural\nnetworks to the state-of-the-art deep reinforcement learning (RL) algorithms in\na simplified simulation of the game `Pong'. Using DishBrain, a system that\nembodies in vitro neural networks with in silico computation using a\nhigh-density multi-electrode array, we contrasted the learning rate and the\nperformance of these biological systems against time-matched learning from\nthree state-of-the-art deep RL algorithms (i.e., DQN, A2C, and PPO) in the same\ngame environment. This allowed a meaningful comparison between biological\nneural systems and deep RL. We find that when samples are limited to a\nreal-world time course, even these very simple biological cultures outperformed\ndeep RL algorithms across various game performance characteristics, implying a\nhigher sample efficiency. Ultimately, even when tested across multiple types of\ninformation input to assess the impact of higher dimensional data input,\nbiological neurons showcased faster learning than all deep reinforcement\nlearning agents.\n", "date": "2024-05-27", "categories": []}
{"id": "2405.13190", "title": "Interpretable Spatio-Temporal Embedding for Brain Structural-Effective\n  Network with Ordinary Differential Equation", "abstract": "  The MRI-derived brain network serves as a pivotal instrument in elucidating\nboth the structural and functional aspects of the brain, encompassing the\nramifications of diseases and developmental processes. However, prevailing\nmethodologies, often focusing on synchronous BOLD signals from functional MRI\n(fMRI), may not capture directional influences among brain regions and rarely\ntackle temporal functional dynamics. In this study, we first construct the\nbrain-effective network via the dynamic causal model. Subsequently, we\nintroduce an interpretable graph learning framework termed Spatio-Temporal\nEmbedding ODE (STE-ODE). This framework incorporates specifically designed\ndirected node embedding layers, aiming at capturing the dynamic interplay\nbetween structural and effective networks via an ordinary differential equation\n(ODE) model, which characterizes spatial-temporal brain dynamics. Our framework\nis validated on several clinical phenotype prediction tasks using two\nindependent publicly available datasets (HCP and OASIS). The experimental\nresults clearly demonstrate the advantages of our model compared to several\nstate-of-the-art methods.\n", "date": "2024-05-21", "categories": []}
{"id": "2405.13394", "title": "A theory of neural emulators", "abstract": "  A central goal in neuroscience is to provide explanations for how animal\nnervous systems can generate actions and cognitive states such as consciousness\nwhile artificial intelligence (AI) and machine learning (ML) seek to provide\nmodels that are increasingly better at prediction. Despite many decades of\nresearch we have made limited progress on providing neuroscience explanations\nyet there is an increased use of AI and ML methods in neuroscience for\nprediction of behavior and even cognitive states. Here we propose emulator\ntheory (ET) and neural emulators as circuit- and scale-independent predictive\nmodels of biological brain activity and emulator theory (ET) as an alternative\nresearch paradigm in neuroscience. ET proposes that predictive models trained\nsolely on neural dynamics and behaviors can generate functionally\nindistinguishable systems from their sources. That is, compared to the\nbiological organisms which they model, emulators may achieve indistinguishable\nbehavior and cognitive states - including consciousness - without any\nmechanistic explanations. We posit ET via several conjectures, discuss the\nnature of endogenous and exogenous activation of neural circuits, and discuss\nneural causality of phenomenal states. ET provides the conceptual and empirical\nframework for prediction-based models of neural dynamics and behavior without\nexplicit representations of idiosyncratically evolved nervous systems.\n", "date": "2024-05-22", "categories": []}
{"id": "2405.14600", "title": "Discretization of continuous input spaces in the hippocampal autoencoder", "abstract": "  The hippocampus has been associated with both spatial cognition and episodic\nmemory formation, but integrating these functions into a unified framework\nremains challenging. Here, we demonstrate that forming discrete memories of\nvisual events in sparse autoencoder neurons can produce spatial tuning similar\nto hippocampal place cells. We then show that the resulting very\nhigh-dimensional code enables neurons to discretize and tile the underlying\nimage space with minimal overlap. Additionally, we extend our results to the\nauditory domain, showing that neurons similarly tile the frequency space in an\nexperience-dependent manner. Lastly, we show that reinforcement learning agents\ncan effectively perform various visuo-spatial cognitive tasks using these\nsparse, very high-dimensional representations.\n", "date": "2024-05-23", "categories": []}
{"id": "2309.16457", "title": "SI-SD: Sleep Interpreter through awake-guided cross-subject Semantic\n  Decoding", "abstract": "  Understanding semantic content from brain activity during sleep represents a\nmajor goal in neuroscience. While studies in rodents have shown spontaneous\nneural reactivation of memories during sleep, capturing the semantic content of\nhuman sleep poses a significant challenge due to the absence of well-annotated\nsleep datasets and the substantial differences in neural patterns between\nwakefulness and sleep. To address these challenges, we designed a novel\ncognitive neuroscience experiment and collected a comprehensive, well-annotated\nelectroencephalography (EEG) dataset from 134 subjects during both wakefulness\nand sleep. Leveraging this benchmark dataset, we developed SI-SD that enhances\nsleep semantic decoding through the position-wise alignment of neural latent\nsequence between wakefulness and sleep. In the 15-way classification task, our\nmodel achieves 24.12% and 21.39% top-1 accuracy on unseen subjects for NREM 2/3\nand REM sleep, respectively, surpassing all other baselines. With additional\nfine-tuning, decoding performance improves to 30.32% and 31.65%, respectively.\nBesides, inspired by previous neuroscientific findings, we systematically\nanalyze how the \"Slow Oscillation\" event impacts decoding performance in NREM\n2/3 sleep -- decoding performance on unseen subjects further improves to\n40.02%. Together, our findings and methodologies contribute to a promising\nneuro-AI framework for decoding brain activity during sleep.\n", "date": "2023-09-28", "categories": []}
{"id": "2405.11683", "title": "Conditionally-Conjugate Gaussian Process Factor Analysis for Spike Count\n  Data via Data Augmentation", "abstract": "  Gaussian process factor analysis (GPFA) is a latent variable modeling\ntechnique commonly used to identify smooth, low-dimensional latent trajectories\nunderlying high-dimensional neural recordings. Specifically, researchers model\nspiking rates as Gaussian observations, resulting in tractable inference.\nRecently, GPFA has been extended to model spike count data. However, due to the\nnon-conjugacy of the likelihood, the inference becomes intractable. Prior works\nrely on either black-box inference techniques, numerical integration or\npolynomial approximations of the likelihood to handle intractability. To\novercome this challenge, we propose a conditionally-conjugate Gaussian process\nfactor analysis (ccGPFA) resulting in both analytically and computationally\ntractable inference for modeling neural activity from spike count data. In\nparticular, we develop a novel data augmentation based method that renders the\nmodel conditionally conjugate. Consequently, our model enjoys the advantage of\nsimple closed-form updates using a variational EM algorithm. Furthermore, due\nto its conditional conjugacy, we show our model can be readily scaled using\nsparse Gaussian Processes and accelerated inference via natural gradients. To\nvalidate our method, we empirically demonstrate its efficacy through\nexperiments.\n", "date": "2024-05-19", "categories": []}
{"id": "2405.08237", "title": "A predictive learning model can simulate temporal dynamics and context\n  effects found in neural representations of continuous speech", "abstract": "  Speech perception involves storing and integrating sequentially presented\nitems. Recent work in cognitive neuroscience has identified temporal and\ncontextual characteristics in humans' neural encoding of speech that may\nfacilitate this temporal processing. In this study, we simulated similar\nanalyses with representations extracted from a computational model that was\ntrained on unlabelled speech with the learning objective of predicting upcoming\nacoustics. Our simulations revealed temporal dynamics similar to those in brain\nsignals, implying that these properties can arise without linguistic knowledge.\nAnother property shared between brains and the model is that the encoding\npatterns of phonemes support some degree of cross-context generalization.\nHowever, we found evidence that the effectiveness of these generalizations\ndepends on the specific contexts, which suggests that this analysis alone is\ninsufficient to support the presence of context-invariant encoding.\n", "date": "2024-05-13", "categories": []}
{"id": "2402.01590", "title": "NeuroCine: Decoding Vivid Video Sequences from Human Brain Activties", "abstract": "  In the pursuit to understand the intricacies of human brain's visual\nprocessing, reconstructing dynamic visual experiences from brain activities\nemerges as a challenging yet fascinating endeavor. While recent advancements\nhave achieved success in reconstructing static images from non-invasive brain\nrecordings, the domain of translating continuous brain activities into video\nformat remains underexplored. In this work, we introduce NeuroCine, a novel\ndual-phase framework to targeting the inherent challenges of decoding fMRI\ndata, such as noises, spatial redundancy and temporal lags. This framework\nproposes spatial masking and temporal interpolation-based augmentation for\ncontrastive learning fMRI representations and a diffusion model enhanced by\ndependent prior noise for video generation. Tested on a publicly available fMRI\ndataset, our method shows promising results, outperforming the previous\nstate-of-the-art models by a notable margin of ${20.97\\%}$, ${31.00\\%}$ and\n${12.30\\%}$ respectively on decoding the brain activities of three subjects in\nthe fMRI dataset, as measured by SSIM. Additionally, our attention analysis\nsuggests that the model aligns with existing brain structures and functions,\nindicating its biological plausibility and interpretability.\n", "date": "2024-02-02", "categories": []}
{"id": "2405.06732", "title": "A Global Data-Driven Model for The Hippocampus and Nucleus Accumbens of\n  Rat From The Local Field Potential Recordings (LFP)", "abstract": "  In brain neural networks, Local Field Potential (LFP) signals represent the\ndynamic flow of information. Analyzing LFP clinical data plays a critical role\nin improving our understanding of brain mechanisms. One way to enhance our\nunderstanding of these mechanisms is to identify a global model to predict\nbrain signals in different situations. This paper identifies a global\ndata-driven based on LFP recordings of the Nucleus Accumbens and Hippocampus\nregions in freely moving rats. The LFP is recorded from each rat in two\ndifferent situations: before and after the process of getting a reward which\ncan be either a drug (Morphine) or natural food (like popcorn or biscuit). A\ncomparison of five machine learning methods including Long Short Term Memory\n(LSTM), Echo State Network (ESN), Deep Echo State Network (DeepESN), Radial\nBasis Function (RBF), and Local Linear Model Tree (LLM) is conducted to develop\nthis model. LoLiMoT was chosen with the best performance among all methods.\nThis model can predict the future states of these regions with one pre-trained\nmodel. Identifying this model showed that Morphine and natural rewards do not\nchange the dynamic features of neurons in these regions.\n", "date": "2024-05-10", "categories": []}
{"id": "2405.05424", "title": "Latent Variable Double Gaussian Process Model for Decoding Complex\n  Neural Data", "abstract": "  Non-parametric models, such as Gaussian Processes (GP), show promising\nresults in the analysis of complex data. Their applications in neuroscience\ndata have recently gained traction. In this research, we introduce a novel\nneural decoder model built upon GP models. The core idea is that two GPs\ngenerate neural data and their associated labels using a set of low-dimensional\nlatent variables. Under this modeling assumption, the latent variables\nrepresent the underlying manifold or essential features present in the neural\ndata. When GPs are trained, the latent variable can be inferred from neural\ndata to decode the labels with a high accuracy. We demonstrate an application\nof this decoder model in a verbal memory experiment dataset and show that the\ndecoder accuracy in predicting stimulus significantly surpasses the\nstate-of-the-art decoder models. The preceding performance of this model\nhighlights the importance of utilizing non-parametric models in the analysis of\nneuroscience data.\n", "date": "2024-05-08", "categories": []}
{"id": "2310.04420", "title": "BrainSCUBA: Fine-Grained Natural Language Captions of Visual Cortex\n  Selectivity", "abstract": "  Understanding the functional organization of higher visual cortex is a\ncentral focus in neuroscience. Past studies have primarily mapped the visual\nand semantic selectivity of neural populations using hand-selected stimuli,\nwhich may potentially bias results towards pre-existing hypotheses of visual\ncortex functionality. Moving beyond conventional approaches, we introduce a\ndata-driven method that generates natural language descriptions for images\npredicted to maximally activate individual voxels of interest. Our method --\nSemantic Captioning Using Brain Alignments (\"BrainSCUBA\") -- builds upon the\nrich embedding space learned by a contrastive vision-language model and\nutilizes a pre-trained large language model to generate interpretable captions.\nWe validate our method through fine-grained voxel-level captioning across\nhigher-order visual regions. We further perform text-conditioned image\nsynthesis with the captions, and show that our images are semantically coherent\nand yield high predicted activations. Finally, to demonstrate how our method\nenables scientific discovery, we perform exploratory investigations on the\ndistribution of \"person\" representations in the brain, and discover\nfine-grained semantic selectivity in body-selective areas. Unlike earlier\nstudies that decode text, our method derives voxel-wise captions of semantic\nselectivity. Our results show that BrainSCUBA is a promising means for\nunderstanding functional preferences in the brain, and provides motivation for\nfurther hypothesis-driven investigation of visual cortex.\n", "date": "2023-10-06", "categories": []}
{"id": "2405.01012", "title": "Correcting Biased Centered Kernel Alignment Measures in Biological and\n  Artificial Neural Networks", "abstract": "  Centred Kernel Alignment (CKA) has recently emerged as a popular metric to\ncompare activations from biological and artificial neural networks (ANNs) in\norder to quantify the alignment between internal representations derived from\nstimuli sets (e.g. images, text, video) that are presented to both systems. In\nthis paper we highlight issues that the community should take into account if\nusing CKA as an alignment metric with neural data. Neural data are in the\nlow-data high-dimensionality domain, which is one of the cases where (biased)\nCKA results in high similarity scores even for pairs of random matrices. Using\nfMRI and MEG data from the THINGS project, we show that if biased CKA is\napplied to representations of different sizes in the low-data\nhigh-dimensionality domain, they are not directly comparable due to biased\nCKA's sensitivity to differing feature-sample ratios and not stimuli-driven\nresponses. This situation can arise both when comparing a pre-selected area of\ninterest (e.g. ROI) to multiple ANN layers, as well as when determining to\nwhich ANN layer multiple regions of interest (ROIs) / sensor groups of\ndifferent dimensionality are most similar. We show that biased CKA can be\nartificially driven to its maximum value when using independent random data of\ndifferent sample-feature ratios. We further show that shuffling sample-feature\npairs of real neural data does not drastically alter biased CKA similarity in\ncomparison to unshuffled data, indicating an undesirable lack of sensitivity to\nstimuli-driven neural responses. Positive alignment of true stimuli-driven\nresponses is only achieved by using debiased CKA. Lastly, we report findings\nthat suggest biased CKA is sensitive to the inherent structure of neural data,\nonly differing from shuffled data when debiased CKA detects stimuli-driven\nalignment.\n", "date": "2024-05-02", "categories": []}
{"id": "2308.01941", "title": "Digital twin brain: a bridge between biological intelligence and\n  artificial intelligence", "abstract": "  In recent years, advances in neuroscience and artificial intelligence have\npaved the way for unprecedented opportunities for understanding the complexity\nof the brain and its emulation by computational systems. Cutting-edge\nadvancements in neuroscience research have revealed the intricate relationship\nbetween brain structure and function, while the success of artificial neural\nnetworks highlights the importance of network architecture. Now is the time to\nbring them together to better unravel how intelligence emerges from the brain's\nmultiscale repositories. In this review, we propose the Digital Twin Brain\n(DTB) as a transformative platform that bridges the gap between biological and\nartificial intelligence. It consists of three core elements: the brain\nstructure that is fundamental to the twinning process, bottom-layer models to\ngenerate brain functions, and its wide spectrum of applications. Crucially,\nbrain atlases provide a vital constraint, preserving the brain's network\norganization within the DTB. Furthermore, we highlight open questions that\ninvite joint efforts from interdisciplinary fields and emphasize the\nfar-reaching implications of the DTB. The DTB can offer unprecedented insights\ninto the emergence of intelligence and neurological disorders, which holds\ntremendous promise for advancing our understanding of both biological and\nartificial intelligence, and ultimately propelling the development of\nartificial general intelligence and facilitating precision mental healthcare.\n", "date": "2023-08-03", "categories": []}
{"id": "2210.13004", "title": "Efficient Representation of Natural Image Patches", "abstract": "  Utilizing an abstract information processing model based on minimal yet\nrealistic assumptions inspired by biological systems, we study how to achieve\nthe early visual system's two ultimate objectives: efficient information\ntransmission and accurate sensor probability distribution modeling. We prove\nthat optimizing for information transmission does not guarantee optimal\nprobability distribution modeling in general. We illustrate, using a two-pixel\n(2D) system and image patches, that an efficient representation can be realized\nthrough a nonlinear population code driven by two types of biologically\nplausible loss functions that depend solely on output. After unsupervised\nlearning, our abstract information processing model bears remarkable\nresemblances to biological systems, despite not mimicking many features of real\nneurons, such as spiking activity. A preliminary comparison with a contemporary\ndeep learning model suggests that our model offers a significant efficiency\nadvantage. Our model provides novel insights into the computational theory of\nearly visual systems as well as a potential new approach to enhance the\nefficiency of deep learning models.\n", "date": "2022-10-24", "categories": []}
{"id": "2402.16770", "title": "Neural population geometry and optimal coding of tasks with shared\n  latent structure", "abstract": "  Humans and animals can recognize latent structures in their environment and\napply this information to efficiently navigate the world. However, it remains\nunclear what aspects of neural activity contribute to these computational\ncapabilities. Here, we develop an analytical theory linking the geometry of a\nneural population's activity to the generalization performance of a linear\nreadout on a set of tasks that depend on a common latent structure. We show\nthat four geometric measures of the activity determine performance across\ntasks. Using this theory, we find that experimentally observed disentangled\nrepresentations naturally emerge as an optimal solution to the multi-task\nlearning problem. When data is scarce, these optimal neural codes compress less\ninformative latent variables, and when data is abundant, they expand these\nvariables in the state space. We validate our theory using macaque ventral\nstream recordings. Our results therefore tie population geometry to multi-task\nlearning.\n", "date": "2024-02-26", "categories": []}
{"id": "2404.05501", "title": "Data Science In Olfaction", "abstract": "  Advances in neural sensing technology are making it possible to observe the\nolfactory process in great detail. In this paper, we conceptualize smell from a\nData Science and AI perspective, that relates the properties of odorants to how\nthey are sensed and analyzed in the olfactory system from the nose to the\nbrain. Drawing distinctions to color vision, we argue that smell presents\nunique measurement challenges, including the complexity of stimuli, the high\ndimensionality of the sensory apparatus, as well as what constitutes ground\ntruth. In the face of these challenges, we argue for the centrality of\nodorant-receptor interactions in developing a theory of olfaction. Such a\ntheory is likely to find widespread industrial applications, and enhance our\nunderstanding of smell, and in the longer-term, how it relates to other senses\nand language. As an initial use case of the data, we present results using\nmachine learning-based classification of neural responses to odors as they are\nrecorded in the mouse olfactory bulb with calcium imaging.\n", "date": "2024-04-08", "categories": []}
{"id": "2308.13234", "title": "Decoding Natural Images from EEG for Object Recognition", "abstract": "  Electroencephalography (EEG) signals, known for convenient non-invasive\nacquisition but low signal-to-noise ratio, have recently gained substantial\nattention due to the potential to decode natural images. This paper presents a\nself-supervised framework to demonstrate the feasibility of learning image\nrepresentations from EEG signals, particularly for object recognition. The\nframework utilizes image and EEG encoders to extract features from paired image\nstimuli and EEG responses. Contrastive learning aligns these two modalities by\nconstraining their similarity. With the framework, we attain significantly\nabove-chance results on a comprehensive EEG-image dataset, achieving a top-1\naccuracy of 15.6% and a top-5 accuracy of 42.8% in challenging 200-way\nzero-shot tasks. Moreover, we perform extensive experiments to explore the\nbiological plausibility by resolving the temporal, spatial, spectral, and\nsemantic aspects of EEG signals. Besides, we introduce attention modules to\ncapture spatial correlations, providing implicit evidence of the brain activity\nperceived from EEG data. These findings yield valuable insights for neural\ndecoding and brain-computer interfaces in real-world scenarios. The code will\nbe released on https://github.com/eeyhsong/NICE-EEG.\n", "date": "2023-08-25", "categories": []}
{"id": "2403.20022", "title": "Psychometry: An Omnifit Model for Image Reconstruction from Human Brain\n  Activity", "abstract": "  Reconstructing the viewed images from human brain activity bridges human and\ncomputer vision through the Brain-Computer Interface. The inherent variability\nin brain function between individuals leads existing literature to focus on\nacquiring separate models for each individual using their respective brain\nsignal data, ignoring commonalities between these data. In this article, we\ndevise Psychometry, an omnifit model for reconstructing images from functional\nMagnetic Resonance Imaging (fMRI) obtained from different subjects. Psychometry\nincorporates an omni mixture-of-experts (Omni MoE) module where all the experts\nwork together to capture the inter-subject commonalities, while each expert\nassociated with subject-specific parameters copes with the individual\ndifferences. Moreover, Psychometry is equipped with a retrieval-enhanced\ninference strategy, termed Ecphory, which aims to enhance the learned fMRI\nrepresentation via retrieving from prestored subject-specific memories. These\ndesigns collectively render Psychometry omnifit and efficient, enabling it to\ncapture both inter-subject commonality and individual specificity across\nsubjects. As a result, the enhanced fMRI representations serve as conditional\nsignals to guide a generation model to reconstruct high-quality and realistic\nimages, establishing Psychometry as state-of-the-art in terms of both\nhigh-level and low-level metrics.\n", "date": "2024-03-29", "categories": []}
{"id": "2403.19149", "title": "Topological Cycle Graph Attention Network for Brain Functional\n  Connectivity", "abstract": "  This study, we introduce a novel Topological Cycle Graph Attention Network\n(CycGAT), designed to delineate a functional backbone within brain functional\ngraph--key pathways essential for signal transmissio--from non-essential,\nredundant connections that form cycles around this core structure. We first\nintroduce a cycle incidence matrix that establishes an independent cycle basis\nwithin a graph, mapping its relationship with edges. We propose a cycle graph\nconvolution that leverages a cycle adjacency matrix, derived from the cycle\nincidence matrix, to specifically filter edge signals in a domain of cycles.\nAdditionally, we strengthen the representation power of the cycle graph\nconvolution by adding an attention mechanism, which is further augmented by the\nintroduction of edge positional encodings in cycles, to enhance the topological\nawareness of CycGAT. We demonstrate CycGAT's localization through simulation\nand its efficacy on an ABCD study's fMRI data (n=8765), comparing it with\nbaseline models. CycGAT outperforms these models, identifying a functional\nbackbone with significantly fewer cycles, crucial for understanding neural\ncircuits related to general intelligence. Our code will be released once\naccepted.\n", "date": "2024-03-28", "categories": []}
{"id": "2311.00118", "title": "Extracting the Multiscale Causal Backbone of Brain Dynamics", "abstract": "  The bulk of the research effort on brain connectivity revolves around\nstatistical associations among brain regions, which do not directly relate to\nthe causal mechanisms governing brain dynamics. Here we propose the multiscale\ncausal backbone (MCB) of brain dynamics, shared by a set of individuals across\nmultiple temporal scales, and devise a principled methodology to extract it.\n  Our approach leverages recent advances in multiscale causal structure\nlearning and optimizes the trade-off between the model fit and its complexity.\nEmpirical assessment on synthetic data shows the superiority of our methodology\nover a baseline based on canonical functional connectivity networks. When\napplied to resting-state fMRI data, we find sparse MCBs for both the left and\nright brain hemispheres. Thanks to its multiscale nature, our approach shows\nthat at low-frequency bands, causal dynamics are driven by brain regions\nassociated with high-level cognitive functions; at higher frequencies instead,\nnodes related to sensory processing play a crucial role. Finally, our analysis\nof individual multiscale causal structures confirms the existence of a causal\nfingerprint of brain connectivity, thus supporting the existing extensive\nresearch in brain connectivity fingerprinting from a causal perspective.\n", "date": "2023-10-31", "categories": []}
{"id": "2402.18546", "title": "Generalizability Under Sensor Failure: Tokenization + Transformers\n  Enable More Robust Latent Spaces", "abstract": "  A major goal in neuroscience is to discover neural data representations that\ngeneralize. This goal is challenged by variability along recording sessions\n(e.g. environment), subjects (e.g. varying neural structures), and sensors\n(e.g. sensor noise), among others. Recent work has begun to address\ngeneralization across sessions and subjects, but few study robustness to sensor\nfailure which is highly prevalent in neuroscience experiments. In order to\naddress these generalizability dimensions we first collect our own\nelectroencephalography dataset with numerous sessions, subjects, and sensors,\nthen study two time series models: EEGNet (Lawhern et al., 2018) and TOTEM\n(Talukder et al., 2024). EEGNet is a widely used convolutional neural network,\nwhile TOTEM is a discrete time series tokenizer and transformer model. We find\nthat TOTEM outperforms or matches EEGNet across all generalizability cases.\nFinally through analysis of TOTEM's latent codebook we observe that\ntokenization enables generalization.\n", "date": "2024-02-28", "categories": []}
{"id": "2403.13368", "title": "Computational Models to Study Language Processing in the Human Brain: A\n  Survey", "abstract": "  Despite differing from the human language processing mechanism in\nimplementation and algorithms, current language models demonstrate remarkable\nhuman-like or surpassing language capabilities. Should computational language\nmodels be employed in studying the brain, and if so, when and how? To delve\ninto this topic, this paper reviews efforts in using computational models for\nbrain research, highlighting emerging trends. To ensure a fair comparison, the\npaper evaluates various computational models using consistent metrics on the\nsame dataset. Our analysis reveals that no single model outperforms others on\nall datasets, underscoring the need for rich testing datasets and rigid\nexperimental control to draw robust conclusions in studies involving\ncomputational models.\n", "date": "2024-03-20", "categories": []}
{"id": "2311.00136", "title": "Neuroformer: Multimodal and Multitask Generative Pretraining for Brain\n  Data", "abstract": "  State-of-the-art systems neuroscience experiments yield large-scale\nmultimodal data, and these data sets require new tools for analysis. Inspired\nby the success of large pretrained models in vision and language domains, we\nreframe the analysis of large-scale, cellular-resolution neuronal spiking data\ninto an autoregressive spatiotemporal generation problem. Neuroformer is a\nmultimodal, multitask generative pretrained transformer (GPT) model that is\nspecifically designed to handle the intricacies of data in systems\nneuroscience. It scales linearly with feature size, can process an arbitrary\nnumber of modalities, and is adaptable to downstream tasks, such as predicting\nbehavior. We first trained Neuroformer on simulated datasets, and found that it\nboth accurately predicted simulated neuronal circuit activity, and also\nintrinsically inferred the underlying neural circuit connectivity, including\ndirection. When pretrained to decode neural responses, the model predicted the\nbehavior of a mouse with only few-shot fine-tuning, suggesting that the model\nbegins learning how to do so directly from the neural representations\nthemselves, without any explicit supervision. We used an ablation study to show\nthat joint training on neuronal responses and behavior boosted performance,\nhighlighting the model's ability to associate behavioral and neural\nrepresentations in an unsupervised manner. These findings show that Neuroformer\ncan analyze neural datasets and their emergent properties, informing the\ndevelopment of models and hypotheses associated with the brain.\n", "date": "2023-10-31", "categories": []}
{"id": "2403.11771", "title": "Modality-Agnostic fMRI Decoding of Vision and Language", "abstract": "  Previous studies have shown that it is possible to map brain activation data\nof subjects viewing images onto the feature representation space of not only\nvision models (modality-specific decoding) but also language models\n(cross-modal decoding). In this work, we introduce and use a new large-scale\nfMRI dataset (~8,500 trials per subject) of people watching both images and\ntext descriptions of such images. This novel dataset enables the development of\nmodality-agnostic decoders: a single decoder that can predict which stimulus a\nsubject is seeing, irrespective of the modality (image or text) in which the\nstimulus is presented. We train and evaluate such decoders to map brain signals\nonto stimulus representations from a large range of publicly available vision,\nlanguage and multimodal (vision+language) models. Our findings reveal that (1)\nmodality-agnostic decoders perform as well as (and sometimes even better than)\nmodality-specific decoders (2) modality-agnostic decoders mapping brain data\nonto representations from unimodal models perform as well as decoders relying\non multimodal representations (3) while language and low-level visual\n(occipital) brain regions are best at decoding text and image stimuli,\nrespectively, high-level visual (temporal) regions perform well on both\nstimulus types.\n", "date": "2024-03-18", "categories": []}
{"id": "2309.08045", "title": "Traveling Waves Encode the Recent Past and Enhance Sequence Learning", "abstract": "  Traveling waves of neural activity have been observed throughout the brain at\na diversity of regions and scales; however, their precise computational role is\nstill debated. One physically inspired hypothesis suggests that the cortical\nsheet may act like a wave-propagating system capable of invertibly storing a\nshort-term memory of sequential stimuli through induced waves traveling across\nthe cortical surface, and indeed many experimental results from neuroscience\ncorrelate wave activity with memory tasks. To date, however, the computational\nimplications of this idea have remained hypothetical due to the lack of a\nsimple recurrent neural network architecture capable of exhibiting such waves.\nIn this work, we introduce a model to fill this gap, which we denote the\nWave-RNN (wRNN), and demonstrate how such an architecture indeed efficiently\nencodes the recent past through a suite of synthetic memory tasks where wRNNs\nlearn faster and reach significantly lower error than wave-free counterparts.\nWe further explore the implications of this memory storage system on more\ncomplex sequence modeling tasks such as sequential image classification and\nfind that wave-based models not only again outperform comparable wave-free RNNs\nwhile using significantly fewer parameters, but additionally perform comparably\nto more complex gated architectures such as LSTMs and GRUs.\n", "date": "2023-09-03", "categories": []}
{"id": "2302.05451", "title": "Brain Effective Connectome based on fMRI and DTI Data: Bayesian Causal\n  Learning and Assessment", "abstract": "  Neuroscientific studies aim to find an accurate and reliable brain Effective\nConnectome (EC). Although current EC discovery methods have contributed to our\nunderstanding of brain organization, their performances are severely\nconstrained by the short sample size and poor temporal resolution of fMRI data,\nand high dimensionality of the brain connectome. By leveraging the DTI data as\nprior knowledge, we introduce two Bayesian causal discovery frameworks -- the\nBayesian GOLEM (BGOLEM) and Bayesian FGES (BFGES) methods -- that offer\nsignificantly more accurate and reliable ECs and address the shortcomings of\nthe existing causal discovery methods in discovering ECs based on only fMRI\ndata. Through a series of simulation studies on synthetic and hybrid (DTI of\nthe Human Connectome Project (HCP) subjects and synthetic fMRI) data, we\ndemonstrate the effectiveness of the proposed methods in discovering EC. To\nnumerically assess the improvement in the accuracy of ECs with our method on\nempirical data, we first introduce the Pseudo False Discovery Rate (PFDR) as a\nnew computational accuracy metric for causal discovery in the brain. We show\nthat our Bayesian methods achieve higher accuracy than traditional methods on\nHCP data. Additionally, we measure the reliability of discovered ECs using the\nRogers-Tanimoto index for test-retest data and show that our Bayesian methods\nprovide significantly more reproducible ECs than traditional methods. Overall,\nour study's numerical and graphical results highlight the potential for these\nframeworks to advance our understanding of brain function and organization\nsignificantly.\n", "date": "2023-02-10", "categories": []}
{"id": "2306.06138", "title": "Extraction and Recovery of Spatio-Temporal Structure in Latent Dynamics\n  Alignment with Diffusion Models", "abstract": "  In the field of behavior-related brain computation, it is necessary to align\nraw neural signals against the drastic domain shift among them. A foundational\nframework within neuroscience research posits that trial-based neural\npopulation activities rely on low-dimensional latent dynamics, thus focusing on\nthe latter greatly facilitates the alignment procedure. Despite this field's\nprogress, existing methods ignore the intrinsic spatio-temporal structure\nduring the alignment phase. Hence, their solutions usually lead to poor quality\nin latent dynamics structures and overall performance. To tackle this problem,\nwe propose an alignment method ERDiff, which leverages the expressivity of the\ndiffusion model to preserve the spatio-temporal structure of latent dynamics.\nSpecifically, the latent dynamics structures of the source domain are first\nextracted by a diffusion model. Then, under the guidance of this diffusion\nmodel, such structures are well-recovered through a maximum likelihood\nalignment procedure in the target domain. We first demonstrate the\neffectiveness of our proposed method on a synthetic dataset. Then, when applied\nto neural recordings from the non-human primate motor cortex, under both\ncross-day and inter-subject settings, our method consistently manifests its\ncapability of preserving the spatiotemporal structure of latent dynamics and\noutperforms existing approaches in alignment goodness-of-fit and neural\ndecoding performance.\n", "date": "2023-06-09", "categories": []}
{"id": "2309.07080", "title": "Discovering Dynamic Effective Connectome of Brain with Bayesian Dynamic\n  DAG Learning", "abstract": "  Understanding the complex mechanisms of the brain can be unraveled by\nextracting the Dynamic Effective Connectome (DEC). Recently, score-based\nDirected Acyclic Graph (DAG) discovery methods have shown significant\nimprovements in extracting the causal structure and inferring effective\nconnectivity. However, learning DEC through these methods still faces two main\nchallenges: one with the fundamental impotence of high-dimensional dynamic DAG\ndiscovery methods and the other with the low quality of fMRI data. In this\npaper, we introduce Bayesian Dynamic DAG learning with M-matrices Acyclicity\ncharacterization (BDyMA) method to address the challenges in discovering DEC.\nThe presented dynamic causal model enables us to discover direct feedback loop\nedges as well. Leveraging an unconstrained framework in the BDyMA method leads\nto more accurate results in detecting high-dimensional networks, achieving\nsparser outcomes, making it particularly suitable for extracting DEC.\nAdditionally, the score function of the BDyMA method allows the incorporation\nof prior knowledge into the process of dynamic causal discovery which further\nenhances the accuracy of results. Comprehensive simulations on synthetic data\nand experiments on Human Connectome Project (HCP) data demonstrate that our\nmethod can handle both of the two main challenges, yielding more accurate and\nreliable DEC compared to state-of-the-art and traditional methods.\nAdditionally, we investigate the trustworthiness of DTI data as prior knowledge\nfor DEC discovery and show the improvements in DEC discovery when the DTI data\nis incorporated into the process.\n", "date": "2023-09-07", "categories": []}
{"id": "2305.19394", "title": "Synaptic Weight Distributions Depend on the Geometry of Plasticity", "abstract": "  A growing literature in computational neuroscience leverages gradient descent\nand learning algorithms that approximate it to study synaptic plasticity in the\nbrain. However, the vast majority of this work ignores a critical underlying\nassumption: the choice of distance for synaptic changes - i.e. the geometry of\nsynaptic plasticity. Gradient descent assumes that the distance is Euclidean,\nbut many other distances are possible, and there is no reason that biology\nnecessarily uses Euclidean geometry. Here, using the theoretical tools provided\nby mirror descent, we show that the distribution of synaptic weights will\ndepend on the geometry of synaptic plasticity. We use these results to show\nthat experimentally-observed log-normal weight distributions found in several\nbrain areas are not consistent with standard gradient descent (i.e. a Euclidean\ngeometry), but rather with non-Euclidean distances. Finally, we show that it\nshould be possible to experimentally test for different synaptic geometries by\ncomparing synaptic weight distributions before and after learning. Overall, our\nwork shows that the current paradigm in theoretical work on synaptic plasticity\nthat assumes Euclidean synaptic geometry may be misguided and that it should be\npossible to experimentally determine the true geometry of synaptic plasticity\nin the brain.\n", "date": "2023-05-30", "categories": []}
{"id": "2402.14102", "title": "Learning dynamic representations of the functional connectome in\n  neurobiological networks", "abstract": "  The static synaptic connectivity of neuronal circuits stands in direct\ncontrast to the dynamics of their function. As in changing community\ninteractions, different neurons can participate actively in various\ncombinations to effect behaviors at different times. We introduce an\nunsupervised approach to learn the dynamic affinities between neurons in live,\nbehaving animals, and to reveal which communities form among neurons at\ndifferent times. The inference occurs in two major steps. First, pairwise\nnon-linear affinities between neuronal traces from brain-wide calcium activity\nare organized by non-negative tensor factorization (NTF). Each factor specifies\nwhich groups of neurons are most likely interacting for an inferred interval in\ntime, and for which animals. Finally, a generative model that allows for\nweighted community detection is applied to the functional motifs produced by\nNTF to reveal a dynamic functional connectome. Since time codes the different\nexperimental variables (e.g., application of chemical stimuli), this provides\nan atlas of neural motifs active during separate stages of an experiment (e.g.,\nstimulus application or spontaneous behaviors). Results from our analysis are\nexperimentally validated, confirming that our method is able to robustly\npredict causal interactions between neurons to generate behavior. Code is\navailable at https://github.com/dyballa/dynamic-connectomes.\n", "date": "2024-02-21", "categories": []}
{"id": "2311.05106", "title": "A differentiable brain simulator bridging brain simulation and\n  brain-inspired computing", "abstract": "  Brain simulation builds dynamical models to mimic the structure and functions\nof the brain, while brain-inspired computing (BIC) develops intelligent systems\nby learning from the structure and functions of the brain. The two fields are\nintertwined and should share a common programming framework to facilitate each\nother's development. However, none of the existing software in the fields can\nachieve this goal, because traditional brain simulators lack differentiability\nfor training, while existing deep learning (DL) frameworks fail to capture the\nbiophysical realism and complexity of brain dynamics. In this paper, we\nintroduce BrainPy, a differentiable brain simulator developed using JAX and\nXLA, with the aim of bridging the gap between brain simulation and BIC. BrainPy\nexpands upon the functionalities of JAX, a powerful AI framework, by\nintroducing complete capabilities for flexible, efficient, and scalable brain\nsimulation. It offers a range of sparse and event-driven operators for\nefficient and scalable brain simulation, an abstraction for managing the\nintricacies of synaptic computations, a modular and flexible interface for\nconstructing multi-scale brain models, and an object-oriented just-in-time\ncompilation approach to handle the memory-intensive nature of brain dynamics.\nWe showcase the efficiency and scalability of BrainPy on benchmark tasks,\nhighlight its differentiable simulation for biologically plausible spiking\nmodels, and discuss its potential to support research at the intersection of\nbrain simulation and BIC.\n", "date": "2023-11-09", "categories": []}
{"id": "2402.13275", "title": "Implementation of a Model of the Cortex Basal Ganglia Loop", "abstract": "  This article presents a simple model of the cortex-basal ganglia-thalamus\nloop, which is thought to serve for action selection and executions, and\nreports the results of its implementation. The model is based on the hypothesis\nthat the cerebral cortex predicts actions, while the basal ganglia use\nreinforcement learning to decide whether to perform the actions predicted by\nthe cortex. The implementation is intended to be used as a component of models\nof the brain consisting of cortical regions or brain-inspired cognitive\narchitectures.\n", "date": "2024-02-17", "categories": []}
{"id": "2305.08459", "title": "Introduction to dynamical mean-field theory of randomly connected neural\n  networks with bidirectionally correlated couplings", "abstract": "  Dynamical mean-field theory is a powerful physics tool used to analyze the\ntypical behavior of neural networks, where neurons can be recurrently\nconnected, or multiple layers of neurons can be stacked. However, it is not\neasy for beginners to access the essence of this tool and the underlying\nphysics. Here, we give a pedagogical introduction of this method in a\nparticular example of random neural networks, where neurons are randomly and\nfully connected by correlated synapses and therefore the network exhibits rich\nemergent collective dynamics. We also review related past and recent important\nworks applying this tool. In addition, a physically transparent and alternative\nmethod, namely the dynamical cavity method, is also introduced to derive\nexactly the same results. The numerical implementation of solving the\nintegro-differential mean-field equations is also detailed, with an\nillustration of exploring the fluctuation dissipation theorem.\n", "date": "2023-05-15", "categories": []}
{"id": "2310.08513", "title": "How connectivity structure shapes rich and lazy learning in neural\n  circuits", "abstract": "  In theoretical neuroscience, recent work leverages deep learning tools to\nexplore how some network attributes critically influence its learning dynamics.\nNotably, initial weight distributions with small (resp. large) variance may\nyield a rich (resp. lazy) regime, where significant (resp. minor) changes to\nnetwork states and representation are observed over the course of learning.\nHowever, in biology, neural circuit connectivity could exhibit a low-rank\nstructure and therefore differs markedly from the random initializations\ngenerally used for these studies. As such, here we investigate how the\nstructure of the initial weights -- in particular their effective rank --\ninfluences the network learning regime. Through both empirical and theoretical\nanalyses, we discover that high-rank initializations typically yield smaller\nnetwork changes indicative of lazier learning, a finding we also confirm with\nexperimentally-driven initial connectivity in recurrent neural networks.\nConversely, low-rank initialization biases learning towards richer learning.\nImportantly, however, as an exception to this rule, we find lazier learning can\nstill occur with a low-rank initialization that aligns with task and data\nstatistics. Our research highlights the pivotal role of initial weight\nstructures in shaping learning regimes, with implications for metabolic costs\nof plasticity and risks of catastrophic forgetting.\n", "date": "2023-10-12", "categories": []}
{"id": "2310.19192", "title": "Emergence of Grid-like Representations by Training Recurrent Networks\n  with Conformal Normalization", "abstract": "  Grid cells in the entorhinal cortex of mammalian brains exhibit striking\nhexagon grid firing patterns in their response maps as the animal (e.g., a rat)\nnavigates in a 2D open environment. In this paper, we study the emergence of\nthe hexagon grid patterns of grid cells based on a general recurrent neural\nnetwork (RNN) model that captures the navigation process. The responses of grid\ncells collectively form a high dimensional vector, representing the 2D\nself-position of the agent. As the agent moves, the vector is transformed by an\nRNN that takes the velocity of the agent as input. We propose a simple yet\ngeneral conformal normalization of the input velocity of the RNN, so that the\nlocal displacement of the position vector in the high-dimensional neural space\nis proportional to the local displacement of the agent in the 2D physical\nspace, regardless of the direction of the input velocity. We apply this\nmechanism to both a linear RNN and nonlinear RNNs. Theoretically, we provide an\nunderstanding that explains the connection between conformal normalization and\nthe emergence of hexagon grid patterns. Empirically, we conduct extensive\nexperiments to verify that conformal normalization is crucial for the emergence\nof hexagon grid patterns, across various types of RNNs. The learned patterns\nshare similar profiles to biological grid cells, and the topological properties\nof the patterns also align with our theoretical understanding.\n", "date": "2023-10-29", "categories": []}
{"id": "2401.09491", "title": "Memory, Space, and Planning: Multiscale Predictive Representations", "abstract": "  Memory is inherently entangled with prediction and planning. Flexible\nbehavior in biological and artificial agents depends on the interplay of\nlearning from the past and predicting the future in ever-changing environments.\nThis chapter reviews computational, behavioral, and neural evidence suggesting\nthese processes rely on learning the relational structure of experiences, known\nas cognitive maps, and draws two key takeaways. First, that these memory\nstructures are organized as multiscale, compact predictive representations in\nhippocampal and prefrontal cortex, or PFC, hierarchies. Second, we argue that\nsuch predictive memory structures are crucial to the complementary functions of\nthe hippocampus and PFC, both for enabling a recall of detailed and coherent\npast episodes as well as generalizing experiences at varying scales for\nefficient prediction and planning. These insights advance our understanding of\nmemory and planning mechanisms in the brain and hold significant implications\nfor advancing artificial intelligence systems.\n", "date": "2024-01-16", "categories": []}
{"id": "2402.09464", "title": "Different Algorithms (Might) Uncover Different Patterns: A Brain-Age\n  Prediction Case Study", "abstract": "  Machine learning is a rapidly evolving field with a wide range of\napplications, including biological signal analysis, where novel algorithms\noften improve the state-of-the-art. However, robustness to algorithmic\nvariability - measured by different algorithms, consistently uncovering similar\nfindings - is seldom explored. In this paper we investigate whether established\nhypotheses in brain-age prediction from EEG research validate across\nalgorithms. First, we surveyed literature and identified various features known\nto be informative for brain-age prediction. We employed diverse feature\nextraction techniques, processing steps, and models, and utilized the\ninterpretative power of SHapley Additive exPlanations (SHAP) values to align\nour findings with the existing research in the field. Few of our models\nachieved state-of-the-art performance on the specific data-set we utilized.\nMoreover, analysis demonstrated that while most models do uncover similar\npatterns in the EEG signals, some variability could still be observed. Finally,\na few prominent findings could only be validated using specific models. We\nconclude by suggesting remedies to the potential implications of this lack of\nrobustness to model variability.\n", "date": "2024-02-08", "categories": []}
{"id": "2402.08811", "title": "Deep and shallow data science for multi-scale optical neuroscience", "abstract": "  Optical imaging of the brain has expanded dramatically in the past two\ndecades. New optics, indicators, and experimental paradigms are now enabling\nin-vivo imaging from the synaptic to the cortex-wide scales. To match the\nresulting flood of data across scales, computational methods are continuously\nbeing developed to meet the need of extracting biologically relevant\ninformation. In this pursuit, challenges arise in some domains (e.g., SNR and\nresolution limits in micron-scale data) that require specialized algorithms.\nThese algorithms can, for example, make use of state-of-the-art machine\nlearning to maximally learn the details of a given scale to optimize the\nprocessing pipeline. In contrast, other methods, however, such as graph signal\nprocessing, seek to abstract away from some of the details that are\nscale-specific to provide solutions to specific sub-problems common across\nscales of neuroimaging. Here we discuss limitations and tradeoffs in\nalgorithmic design with the goal of identifying how data quality and\nvariability can hamper algorithm use and dissemination.\n", "date": "2024-02-13", "categories": []}
{"id": "2402.04274", "title": "FPGA Deployment of LFADS for Real-time Neuroscience Experiments", "abstract": "  Large-scale recordings of neural activity are providing new opportunities to\nstudy neural population dynamics. A powerful method for analyzing such\nhigh-dimensional measurements is to deploy an algorithm to learn the\nlow-dimensional latent dynamics. LFADS (Latent Factor Analysis via Dynamical\nSystems) is a deep learning method for inferring latent dynamics from\nhigh-dimensional neural spiking data recorded simultaneously in single trials.\nThis method has shown a remarkable performance in modeling complex brain\nsignals with an average inference latency in milliseconds. As our capacity of\nsimultaneously recording many neurons is increasing exponentially, it is\nbecoming crucial to build capacity for deploying low-latency inference of the\ncomputing algorithms. To improve the real-time processing ability of LFADS, we\nintroduce an efficient implementation of the LFADS models onto Field\nProgrammable Gate Arrays (FPGA). Our implementation shows an inference latency\nof 41.97 $\\mu$s for processing the data in a single trial on a Xilinx U55C.\n", "date": "2024-02-02", "categories": []}
{"id": "2402.00588", "title": "BrainSLAM: SLAM on Neural Population Activity Data", "abstract": "  Simultaneous localisation and mapping (SLAM) algorithms are commonly used in\nrobotic systems for learning maps of novel environments. Brains also appear to\nlearn maps, but the mechanisms are not known and it is unclear how to infer\nthese maps from neural activity data. We present BrainSLAM; a method for\nperforming SLAM using only population activity (local field potential, LFP)\ndata simultaneously recorded from three brain regions in rats: hippocampus,\nprefrontal cortex, and parietal cortex. This system uses a convolutional neural\nnetwork (CNN) to decode velocity and familiarity information from wavelet\nscalograms of neural local field potential data recorded from rats as they\nnavigate a 2D maze. The CNN's output drives a RatSLAM-inspired architecture,\npowering an attractor network which performs path integration plus a separate\nsystem which performs `loop closure' (detecting previously visited locations\nand correcting map aliasing errors). Together, these three components can\nconstruct faithful representations of the environment while simultaneously\ntracking the animal's location. This is the first demonstration of inference of\na spatial map from brain recordings. Our findings expand SLAM to a new\nmodality, enabling a new method of mapping environments and facilitating a\nbetter understanding of the role of cognitive maps in navigation and decision\nmaking.\n", "date": "2024-02-01", "categories": []}
{"id": "2310.04645", "title": "Do self-supervised speech and language models extract similar\n  representations as human brain?", "abstract": "  Speech and language models trained through self-supervised learning (SSL)\ndemonstrate strong alignment with brain activity during speech and language\nperception. However, given their distinct training modalities, it remains\nunclear whether they correlate with the same neural aspects. We directly\naddress this question by evaluating the brain prediction performance of two\nrepresentative SSL models, Wav2Vec2.0 and GPT-2, designed for speech and\nlanguage tasks. Our findings reveal that both models accurately predict speech\nresponses in the auditory cortex, with a significant correlation between their\nbrain predictions. Notably, shared speech contextual information between\nWav2Vec2.0 and GPT-2 accounts for the majority of explained variance in brain\nactivity, surpassing static semantic and lower-level acoustic-phonetic\ninformation. These results underscore the convergence of speech contextual\nrepresentations in SSL models and their alignment with the neural network\nunderlying speech perception, offering valuable insights into both SSL models\nand the neural basis of speech and language processing.\n", "date": "2023-10-07", "categories": []}
{"id": "2401.17711", "title": "Prediction of multitasking performance post-longitudinal tDCS via\n  EEG-based functional connectivity and machine learning methods", "abstract": "  Predicting and understanding the changes in cognitive performance, especially\nafter a longitudinal intervention, is a fundamental goal in neuroscience.\nLongitudinal brain stimulation-based interventions like transcranial direct\ncurrent stimulation (tDCS) induce short-term changes in the resting membrane\npotential and influence cognitive processes. However, very little research has\nbeen conducted on predicting these changes in cognitive performance\npost-intervention. In this research, we intend to address this gap in the\nliterature by employing different EEG-based functional connectivity analyses\nand machine learning algorithms to predict changes in cognitive performance in\na complex multitasking task. Forty subjects were divided into experimental and\nactive-control conditions. On Day 1, all subjects executed a multitasking task\nwith simultaneous 32-channel EEG being acquired. From Day 2 to Day 7, subjects\nin the experimental condition undertook 15 minutes of 2mA anodal tDCS\nstimulation during task training. Subjects in the active-control condition\nundertook 15 minutes of sham stimulation during task training. On Day 10, all\nsubjects again executed the multitasking task with EEG acquisition.\nSource-level functional connectivity metrics, namely phase lag index and\ndirected transfer function, were extracted from the EEG data on Day 1 and Day\n10. Various machine learning models were employed to predict changes in\ncognitive performance. Results revealed that the multi-layer perceptron and\ndirected transfer function recorded a cross-validation training RMSE of 5.11%\nand a test RMSE of 4.97%. We discuss the implications of our results in\ndeveloping real-time cognitive state assessors for accurately predicting\ncognitive performance in dynamic and complex tasks post-tDCS intervention\n", "date": "2024-01-31", "categories": []}
{"id": "2401.16742", "title": "Generative AI-based closed-loop fMRI system", "abstract": "  While generative AI is now widespread and useful in society, there are\npotential risks of misuse, e.g., unconsciously influencing cognitive processes\nor decision-making. Although this causes a security problem in the cognitive\ndomain, there has been no research about neural and computational mechanisms\ncounteracting the impact of malicious generative AI in humans. We propose\nDecNefGAN, a novel framework that combines a generative adversarial system and\na neural reinforcement model. More specifically, DecNefGAN bridges human and\ngenerative AI in a closed-loop system, with the AI creating stimuli that induce\nspecific mental states, thus exerting external control over neural activity.\nThe objective of the human is the opposite, to compete and reach an orthogonal\nmental state. This framework can contribute to elucidating how the human brain\nresponds to and counteracts the potential influence of generative AI.\n", "date": "2024-01-30", "categories": []}
{"id": "2305.18417", "title": "Determinantal Point Process Attention Over Grid Cell Code Supports Out\n  of Distribution Generalization", "abstract": "  Deep neural networks have made tremendous gains in emulating human-like\nintelligence, and have been used increasingly as ways of understanding how the\nbrain may solve the complex computational problems on which this relies.\nHowever, these still fall short of, and therefore fail to provide insight into\nhow the brain supports strong forms of generalization of which humans are\ncapable. One such case is out-of-distribution (OOD) generalization-successful\nperformance on test examples that lie outside the distribution of the training\nset. Here, we identify properties of processing in the brain that may\ncontribute to this ability. We describe a two-part algorithm that draws on\nspecific features of neural computation to achieve OOD generalization, and\nprovide a proof of concept by evaluating performance on two challenging\ncognitive tasks. First we draw on the fact that the mammalian brain represents\nmetric spaces using grid cell code (e.g., in the entorhinal cortex): abstract\nrepresentations of relational structure, organized in recurring motifs that\ncover the representational space. Second, we propose an attentional mechanism\nthat operates over the grid cell code using Determinantal Point Process (DPP),\nthat we call DPP attention (DPP-A) -- a transformation that ensures maximum\nsparseness in the coverage of that space. We show that a loss function that\ncombines standard task-optimized error with DPP-A can exploit the recurring\nmotifs in the grid cell code, and can be integrated with common architectures\nto achieve strong OOD generalization performance on analogy and arithmetic\ntasks. This provides both an interpretation of how the grid cell code in the\nmammalian brain may contribute to generalization performance, and at the same\ntime a potential means for improving such capabilities in artificial neural\nnetworks.\n", "date": "2023-05-28", "categories": []}
{"id": "2401.11008", "title": "Helmholtz-Decomposition and Optical Flow: A new method to characterize\n  GCamP recordings", "abstract": "  During deep sleep and under anaesthesia spontaneous patterns of cortical\nactivation frequently take the form of slow travelling waves. Slow wave sleep\nis an important cognitive state especially because of its relevance for memory\nconsolidation. However, despite extensive research the exact mechanisms are\nstill ill-understood. Novel methods such as high speed widefield imaging of\nGCamP activity offer new potentials. Here we show how data recorded from\ntransgenic mice under anesthesia can be processed to analyze sources, sinks and\npatterns of flow. To make the best possible use of the data novel means of data\nprocessing are necessary. Therefore, we (1) give a an brief account on\nprocesses that play a role in generating slow waves and demonstrate (2) a novel\napproach to characterize its patterns in GCamP recordings. While slow waves are\nhighly variable, it shows that some are surprisingly similar. To enable\nquantitative means of analysis and examine the structure of such prototypical\nevents we propose a novel approach for the characterization of slow waves: The\nHelmholtz-Decomposition of gradient-based Dense Optical Flow of the pixeldense\nGCamP contrast (df/f). It allows to detect the sources and sinks of activation\nand discern them from global patterns of neural flow. Aggregated features can\nbe analyzed with variational autoencoders. The results unravel regularities\nbetween slow waves and shows how they relate to the experimental conditions.\nThe approach reveals a complex topology of different features in latent slow\nwave space and identifies prototypical examples for each stage.\n", "date": "2024-01-19", "categories": []}
{"id": "2401.10285", "title": "Analyzing Brain Activity During Learning Tasks with EEG and Machine\n  Learning", "abstract": "  This study aimed to analyze brain activity during various STEM activities,\nexploring the feasibility of classifying between different tasks. EEG brain\ndata from twenty subjects engaged in five cognitive tasks were collected and\nsegmented into 4-second clips. Power spectral densities of brain frequency\nwaves were then analyzed. Testing different k-intervals with XGBoost, Random\nForest, and Bagging Classifier revealed that Random Forest performed best,\nachieving a testing accuracy of 91.07% at an interval size of two. When\nutilizing all four EEG channels, cognitive flexibility was most recognizable.\nTask-specific classification accuracy showed the right frontal lobe excelled in\nmathematical processing and planning, the left frontal lobe in cognitive\nflexibility and mental flexibility, and the left temporoparietal lobe in\nconnections. Notably, numerous connections between frontal and temporoparietal\nlobes were observed during STEM activities. This study contributes to a deeper\nunderstanding of implementing machine learning in analyzing brain activity and\nsheds light on the brain's mechanisms.\n", "date": "2024-01-15", "categories": []}
{"id": "2401.05446", "title": "Self-supervised Learning for Electroencephalogram: A Systematic Survey", "abstract": "  Electroencephalogram (EEG) is a non-invasive technique to record\nbioelectrical signals. Integrating supervised deep learning techniques with EEG\nsignals has recently facilitated automatic analysis across diverse EEG-based\ntasks. However, the label issues of EEG signals have constrained the\ndevelopment of EEG-based deep models. Obtaining EEG annotations is difficult\nthat requires domain experts to guide collection and labeling, and the\nvariability of EEG signals among different subjects causes significant label\nshifts. To solve the above challenges, self-supervised learning (SSL) has been\nproposed to extract representations from unlabeled samples through\nwell-designed pretext tasks. This paper concentrates on integrating SSL\nframeworks with temporal EEG signals to achieve efficient representation and\nproposes a systematic review of the SSL for EEG signals. In this paper, 1) we\nintroduce the concept and theory of self-supervised learning and typical SSL\nframeworks. 2) We provide a comprehensive review of SSL for EEG analysis,\nincluding taxonomy, methodology, and technique details of the existing\nEEG-based SSL frameworks, and discuss the difference between these methods. 3)\nWe investigate the adaptation of the SSL approach to various downstream tasks,\nincluding the task description and related benchmark datasets. 4) Finally, we\ndiscuss the potential directions for future SSL-EEG research.\n", "date": "2024-01-09", "categories": []}
{"id": "2401.06005", "title": "How does the primate brain combine generative and discriminative\n  computations in vision?", "abstract": "  Vision is widely understood as an inference problem. However, two contrasting\nconceptions of the inference process have each been influential in research on\nbiological vision as well as the engineering of machine vision. The first\nemphasizes bottom-up signal flow, describing vision as a largely feedforward,\ndiscriminative inference process that filters and transforms the visual\ninformation to remove irrelevant variation and represent behaviorally relevant\ninformation in a format suitable for downstream functions of cognition and\nbehavioral control. In this conception, vision is driven by the sensory data,\nand perception is direct because the processing proceeds from the data to the\nlatent variables of interest. The notion of \"inference\" in this conception is\nthat of the engineering literature on neural networks, where feedforward\nconvolutional neural networks processing images are said to perform inference.\nThe alternative conception is that of vision as an inference process in\nHelmholtz's sense, where the sensory evidence is evaluated in the context of a\ngenerative model of the causal processes giving rise to it. In this conception,\nvision inverts a generative model through an interrogation of the evidence in a\nprocess often thought to involve top-down predictions of sensory data to\nevaluate the likelihood of alternative hypotheses. The authors include\nscientists rooted in roughly equal numbers in each of the conceptions and\nmotivated to overcome what might be a false dichotomy between them and engage\nthe other perspective in the realm of theory and experiment. The primate brain\nemploys an unknown algorithm that may combine the advantages of both\nconceptions. We explain and clarify the terminology, review the key empirical\nevidence, and propose an empirical research program that transcends the\ndichotomy and sets the stage for revealing the mysterious hybrid algorithm of\nprimate vision.\n", "date": "2024-01-11", "categories": []}
{"id": "2401.03639", "title": "Deep Learning for Visual Neuroprosthesis", "abstract": "  The visual pathway involves complex networks of cells and regions which\ncontribute to the encoding and processing of visual information. While some\naspects of visual perception are understood, there are still many unanswered\nquestions regarding the exact mechanisms of visual encoding and the\norganization of visual information along the pathway. This chapter discusses\nthe importance of visual perception and the challenges associated with\nunderstanding how visual information is encoded and represented in the brain.\nFurthermore, this chapter introduces the concept of neuroprostheses: devices\ndesigned to enhance or replace bodily functions, and highlights the importance\nof constructing computational models of the visual pathway in the\nimplementation of such devices. A number of such models, employing the use of\ndeep learning models, are outlined, and their value to understanding visual\ncoding and natural vision is discussed.\n", "date": "2024-01-08", "categories": []}
{"id": "2401.03851", "title": "Aligned with LLM: a new multi-modal training paradigm for encoding fMRI\n  activity in visual cortex", "abstract": "  Recently, there has been a surge in the popularity of pre trained large\nlanguage models (LLMs) (such as GPT-4), sweeping across the entire Natural\nLanguage Processing (NLP) and Computer Vision (CV) communities. These LLMs have\ndemonstrated advanced multi-modal understanding capabilities and showcased\nstrong performance across various benchmarks. The LLM has started to embody\ntraits of artificial general intelligence, which holds vital guidance for\nenhancing brain-like characteristics within visual encoding models. Hence, This\npaper proposes a new multi-modal training paradigm, aligning with LLM, for\nencoding fMRI activity in visual cortex. Based on this paradigm, we trained an\nencoding model in fMRI data named the LLM-Visual Encoding Model (LLM-VEM).\nSpecifically, we utilize LLM (miniGPT4) to generate descriptive text for all\nstimulus images, forming a high-quality textual description set. Moreover, we\nuse the pre-trained text encoder (CLIP) to process these detailed descriptions,\nobtaining the text embedding features. Next, we use the contrast loss function\nto minimize the distance between the image embedding features and the text\nembedding features to complete the alignment operation of the stimulus image\nand text information. With the assistance of the pre-trained LLM, this\nalignment process facilitates better learning of the visual encoding model,\nresulting in higher precision. The final experimental results indicate that our\ntraining paradigm has significantly aided in enhancing the performance of the\nvisual encoding model.\n", "date": "2024-01-08", "categories": []}
{"id": "2401.01489", "title": "The Neuron as a Direct Data-Driven Controller", "abstract": "  In the quest to model neuronal function amidst gaps in physiological data, a\npromising strategy is to develop a normative theory that interprets neuronal\nphysiology as optimizing a computational objective. This study extends the\ncurrent normative models, which primarily optimize prediction, by\nconceptualizing neurons as optimal feedback controllers. We posit that neurons,\nespecially those beyond early sensory areas, act as controllers, steering their\nenvironment towards a specific desired state through their output. This\nenvironment comprises both synaptically interlinked neurons and external motor\nsensory feedback loops, enabling neurons to evaluate the effectiveness of their\ncontrol via synaptic feedback. Utilizing the novel Direct Data-Driven Control\n(DD-DC) framework, we model neurons as biologically feasible controllers which\nimplicitly identify loop dynamics, infer latent states and optimize control.\nOur DD-DC neuron model explains various neurophysiological phenomena: the shift\nfrom potentiation to depression in Spike-Timing-Dependent Plasticity (STDP)\nwith its asymmetry, the duration and adaptive nature of feedforward and\nfeedback neuronal filters, the imprecision in spike generation under constant\nstimulation, and the characteristic operational variability and noise in the\nbrain. Our model presents a significant departure from the traditional,\nfeedforward, instant-response McCulloch-Pitts-Rosenblatt neuron, offering a\nnovel and biologically-informed fundamental unit for constructing neural\nnetworks.\n", "date": "2024-01-03", "categories": []}
{"id": "2401.01813", "title": "Signal Processing in the Retina: Interpretable Graph Classifier to\n  Predict Ganglion Cell Responses", "abstract": "  It is a popular hypothesis in neuroscience that ganglion cells in the retina\nare activated by selectively detecting visual features in an observed scene.\nWhile ganglion cell firings can be predicted via data-trained deep neural nets,\nthe networks remain indecipherable, thus providing little understanding of the\ncells' underlying operations. To extract knowledge from the cell firings, in\nthis paper we learn an interpretable graph-based classifier from data to\npredict the firings of ganglion cells in response to visual stimuli.\nSpecifically, we learn a positive semi-definite (PSD) metric matrix $\\mathbf{M}\n\\succeq 0$ that defines Mahalanobis distances between graph nodes (visual\nevents) endowed with pre-computed feature vectors; the computed inter-node\ndistances lead to edge weights and a combinatorial graph that is amenable to\nbinary classification. Mathematically, we define the objective of metric matrix\n$\\mathbf{M}$ optimization using a graph adaptation of large margin nearest\nneighbor (LMNN), which is rewritten as a semi-definite programming (SDP)\nproblem. We solve it efficiently via a fast approximation called Gershgorin\ndisc perfect alignment (GDPA) linearization. The learned metric matrix\n$\\mathbf{M}$ provides interpretability: important features are identified along\n$\\mathbf{M}$'s diagonal, and their mutual relationships are inferred from\noff-diagonal terms. Our fast metric learning framework can be applied to other\nbiological systems with pre-chosen features that require interpretation.\n", "date": "2024-01-03", "categories": []}
{"id": "2309.02332", "title": "Information Processing by Neuron Populations in the Central Nervous\n  System: Mathematical Structure of Data and Operations", "abstract": "  In the intricate architecture of the mammalian central nervous system,\nneurons form populations. Axonal bundles communicate between these clusters\nusing spike trains. However, these neuron populations' precise encoding and\noperations have yet to be discovered. In our analysis, the starting point is a\nstate-of-the-art mechanistic model of a generic neuron endowed with plasticity.\nFrom this simple framework emerges a subtle mathematical construct: The\nrepresentation and manipulation of information can be precisely characterized\nby an algebra of convex cones. Furthermore, these neuron populations are not\nmerely passive transmitters. They act as operators within this algebraic\nstructure, mirroring the functionality of a low-level programming language.\nWhen these populations interconnect, they embody succinct yet potent algebraic\nexpressions. These networks allow them to implement many operations, such as\nspecialization, generalization, novelty detection, dimensionality reduction,\ninverse modeling, prediction, and associative memory. In broader terms, this\nwork illuminates the potential of matrix embeddings in advancing our\nunderstanding in fields like cognitive science and AI. These embeddings enhance\nthe capacity for concept processing and hierarchical description over their\nvector counterparts.\n", "date": "2023-09-05", "categories": []}
{"id": "2306.12045", "title": "Temporal Conditioning Spiking Latent Variable Models of the Neural\n  Response to Natural Visual Scenes", "abstract": "  Developing computational models of neural response is crucial for\nunderstanding sensory processing and neural computations. Current\nstate-of-the-art neural network methods use temporal filters to handle temporal\ndependencies, resulting in an unrealistic and inflexible processing paradigm.\nMeanwhile, these methods target trial-averaged firing rates and fail to capture\nimportant features in spike trains. This work presents the temporal\nconditioning spiking latent variable models (TeCoS-LVM) to simulate the neural\nresponse to natural visual stimuli. We use spiking neurons to produce spike\noutputs that directly match the recorded trains. This approach helps to avoid\nlosing information embedded in the original spike trains. We exclude the\ntemporal dimension from the model parameter space and introduce a temporal\nconditioning operation to allow the model to adaptively explore and exploit\ntemporal dependencies in stimuli sequences in a {\\it natural paradigm}. We show\nthat TeCoS-LVM models can produce more realistic spike activities and\naccurately fit spike statistics than powerful alternatives. Additionally,\nlearned TeCoS-LVM models can generalize well to longer time scales. Overall,\nwhile remaining computationally tractable, our model effectively captures key\nfeatures of neural coding systems. It thus provides a useful tool for building\naccurate predictive computational accounts for various sensory perception\ncircuits.\n", "date": "2023-06-21", "categories": []}
{"id": "2312.07547", "title": "Active Inference and Intentional Behaviour", "abstract": "  Recent advances in theoretical biology suggest that basal cognition and\nsentient behaviour are emergent properties of in vitro cell cultures and\nneuronal networks, respectively. Such neuronal networks spontaneously learn\nstructured behaviours in the absence of reward or reinforcement. In this paper,\nwe characterise this kind of self-organisation through the lens of the free\nenergy principle, i.e., as self-evidencing. We do this by first discussing the\ndefinitions of reactive and sentient behaviour in the setting of active\ninference, which describes the behaviour of agents that model the consequences\nof their actions. We then introduce a formal account of intentional behaviour,\nthat describes agents as driven by a preferred endpoint or goal in latent\nstate-spaces. We then investigate these forms of (reactive, sentient, and\nintentional) behaviour using simulations. First, we simulate the aforementioned\nin vitro experiments, in which neuronal cultures spontaneously learn to play\nPong, by implementing nested, free energy minimising processes. The simulations\nare then used to deconstruct the ensuing predictive behaviour, leading to the\ndistinction between merely reactive, sentient, and intentional behaviour, with\nthe latter formalised in terms of inductive planning. This distinction is\nfurther studied using simple machine learning benchmarks (navigation in a grid\nworld and the Tower of Hanoi problem), that show how quickly and efficiently\nadaptive behaviour emerges under an inductive form of active inference.\n", "date": "2023-12-06", "categories": []}
{"id": "2312.08809", "title": "Performance evaluation of matrix factorization for fMRI data", "abstract": "  In the study of the brain, there is a hypothesis that sparse coding is\nrealized in information representation of external stimuli, which is\nexperimentally confirmed for visual stimulus recently. However, unlike the\nspecific functional region in the brain, sparse coding in information\nprocessing in the whole brain has not been clarified sufficiently. In this\nstudy, we investigate the validity of sparse coding in the whole human brain by\napplying various matrix factorization methods to functional magnetic resonance\nimaging data of neural activities in the whole human brain. The result suggests\nsparse coding hypothesis in information representation in the whole human\nbrain, because extracted features from sparse MF method, SparsePCA or MOD under\nhigh sparsity setting, or approximate sparse MF method, FastICA, can classify\nexternal visual stimuli more accurately than non-sparse MF method or sparse MF\nmethod under low sparsity setting.\n", "date": "2023-12-14", "categories": []}
{"id": "2310.19614", "title": "Dis-inhibitory neuronal circuits can control the sign of synaptic\n  plasticity", "abstract": "  How neuronal circuits achieve credit assignment remains a central unsolved\nquestion in systems neuroscience. Various studies have suggested plausible\nsolutions for back-propagating error signals through multi-layer networks.\nThese purely functionally motivated models assume distinct neuronal\ncompartments to represent local error signals that determine the sign of\nsynaptic plasticity. However, this explicit error modulation is inconsistent\nwith phenomenological plasticity models in which the sign depends primarily on\npostsynaptic activity. Here we show how a plausible microcircuit model and\nHebbian learning rule derived within an adaptive control theory framework can\nresolve this discrepancy. Assuming errors are encoded in top-down\ndis-inhibitory synaptic afferents, we show that error-modulated learning\nemerges naturally at the circuit level when recurrent inhibition explicitly\ninfluences Hebbian plasticity. The same learning rule accounts for\nexperimentally observed plasticity in the absence of inhibition and performs\ncomparably to back-propagation of error (BP) on several non-linearly separable\nbenchmarks. Our findings bridge the gap between functional and experimentally\nobserved plasticity rules and make concrete predictions on inhibitory\nmodulation of excitatory plasticity.\n", "date": "2023-10-30", "categories": []}
{"id": "2306.03088", "title": "DeepGraphDMD: Interpretable Spatio-Temporal Decomposition of Non-linear\n  Functional Brain Network Dynamics", "abstract": "  Functional brain dynamics is supported by parallel and overlapping functional\nnetwork modes that are associated with specific neural circuits. Decomposing\nthese network modes from fMRI data and finding their temporal characteristics\nis challenging due to their time-varying nature and the non-linearity of the\nfunctional dynamics. Dynamic Mode Decomposition (DMD) algorithms have been\nquite popular for solving this decomposition problem in recent years. In this\nwork, we apply GraphDMD -- an extension of the DMD for network data -- to\nextract the dynamic network modes and their temporal characteristics from the\nfMRI time series in an interpretable manner. GraphDMD, however, regards the\nunderlying system as a linear dynamical system that is sub-optimal for\nextracting the network modes from non-linear functional data. In this work, we\ndevelop a generalized version of the GraphDMD algorithm -- DeepGraphDMD --\napplicable to arbitrary non-linear graph dynamical systems. DeepGraphDMD is an\nautoencoder-based deep learning model that learns Koopman eigenfunctions for\ngraph data and embeds the non-linear graph dynamics into a latent linear space.\nWe show the effectiveness of our method in both simulated data and the HCP\nresting-state fMRI data. In the HCP data, DeepGraphDMD provides novel insights\ninto cognitive brain functions by discovering two major network modes related\nto fluid and crystallized intelligence.\n", "date": "2023-06-05", "categories": []}
{"id": "2305.13388", "title": "The neural dynamics of auditory word recognition and integration", "abstract": "  Listeners recognize and integrate words in rapid and noisy everyday speech by\ncombining expectations about upcoming content with incremental sensory\nevidence. We present a computational model of word recognition which formalizes\nthis perceptual process in Bayesian decision theory. We fit this model to\nexplain scalp EEG signals recorded as subjects passively listened to a\nfictional story, revealing both the dynamics of the online auditory word\nrecognition process and the neural correlates of the recognition and\nintegration of words.\n  The model reveals distinct neural processing of words depending on whether or\nnot they can be quickly recognized. While all words trigger a neural response\ncharacteristic of probabilistic integration -- voltage modulations predicted by\na word's surprisal in context -- these modulations are amplified for words\nwhich require more than roughly 150 ms of input to be recognized. We observe no\ndifference in the latency of these neural responses according to words'\nrecognition times. Our results are consistent with a two-part model of speech\ncomprehension, combining an eager and rapid process of word recognition with a\ntemporally independent process of word integration. However, we also developed\nalternative models of the scalp EEG signal not incorporating word recognition\ndynamics which showed similar performance improvements. We discuss potential\nfuture modeling steps which may help to separate these hypotheses.\n", "date": "2023-05-22", "categories": []}
{"id": "2311.15682", "title": "Information theoretic study of the neural geometry induced by category\n  learning", "abstract": "  Categorization is an important topic both for biological and artificial\nneural networks. Here, we take an information theoretic approach to assess the\nefficiency of the representations induced by category learning. We show that\none can decompose the relevant Bayesian cost into two components, one for the\ncoding part and one for the decoding part. Minimizing the coding cost implies\nmaximizing the mutual information between the set of categories and the neural\nactivities. We analytically show that this mutual information can be written as\nthe sum of two terms that can be interpreted as (i) finding an appropriate\nrepresentation space, and, (ii) building a representation with the appropriate\nmetrics, based on the neural Fisher information on this space. One main\nconsequence is that category learning induces an expansion of neural space near\ndecision boundaries. Finally, we provide numerical illustrations that show how\nFisher information of the coding neural population aligns with the boundaries\nbetween categories.\n", "date": "2023-11-27", "categories": []}
{"id": "2309.04153", "title": "Mapping EEG Signals to Visual Stimuli: A Deep Learning Approach to Match\n  vs. Mismatch Classification", "abstract": "  Existing approaches to modeling associations between visual stimuli and brain\nresponses are facing difficulties in handling between-subject variance and\nmodel generalization. Inspired by the recent progress in modeling speech-brain\nresponse, we propose in this work a \"match-vs-mismatch\" deep learning model to\nclassify whether a video clip induces excitatory responses in recorded EEG\nsignals and learn associations between the visual content and corresponding\nneural recordings. Using an exclusive experimental dataset, we demonstrate that\nthe proposed model is able to achieve the highest accuracy on unseen subjects\nas compared to other baseline models. Furthermore, we analyze the inter-subject\nnoise using a subject-level silhouette score in the embedding space and show\nthat the developed model is able to mitigate inter-subject noise and\nsignificantly reduce the silhouette score. Moreover, we examine the Grad-CAM\nactivation score and show that the brain regions associated with language\nprocessing contribute most to the model predictions, followed by regions\nassociated with visual processing. These results have the potential to\nfacilitate the development of neural recording-based video reconstruction and\nits related applications.\n", "date": "2023-09-08", "categories": []}
{"id": "2310.13849", "title": "A Dual-Stream Neural Network Explains the Functional Segregation of\n  Dorsal and Ventral Visual Pathways in Human Brains", "abstract": "  The human visual system uses two parallel pathways for spatial processing and\nobject recognition. In contrast, computer vision systems tend to use a single\nfeedforward pathway, rendering them less robust, adaptive, or efficient than\nhuman vision. To bridge this gap, we developed a dual-stream vision model\ninspired by the human eyes and brain. At the input level, the model samples two\ncomplementary visual patterns to mimic how the human eyes use magnocellular and\nparvocellular retinal ganglion cells to separate retinal inputs to the brain.\nAt the backend, the model processes the separate input patterns through two\nbranches of convolutional neural networks (CNN) to mimic how the human brain\nuses the dorsal and ventral cortical pathways for parallel visual processing.\nThe first branch (WhereCNN) samples a global view to learn spatial attention\nand control eye movements. The second branch (WhatCNN) samples a local view to\nrepresent the object around the fixation. Over time, the two branches interact\nrecurrently to build a scene representation from moving fixations. We compared\nthis model with the human brains processing the same movie and evaluated their\nfunctional alignment by linear transformation. The WhereCNN and WhatCNN\nbranches were found to differentially match the dorsal and ventral pathways of\nthe visual cortex, respectively, primarily due to their different learning\nobjectives. These model-based results lead us to speculate that the distinct\nresponses and representations of the ventral and dorsal streams are more\ninfluenced by their distinct goals in visual attention and object recognition\nthan by their specific bias or selectivity in retinal inputs. This dual-stream\nmodel takes a further step in brain-inspired computer vision, enabling parallel\nneural networks to actively explore and understand the visual surroundings.\n", "date": "2023-10-20", "categories": []}
{"id": "2311.11390", "title": "Addressing the speed-accuracy simulation trade-off for adaptive spiking\n  neurons", "abstract": "  The adaptive leaky integrate-and-fire (ALIF) model is fundamental within\ncomputational neuroscience and has been instrumental in studying our brains\n$\\textit{in silico}$. Due to the sequential nature of simulating these neural\nmodels, a commonly faced issue is the speed-accuracy trade-off: either\naccurately simulate a neuron using a small discretisation time-step (DT), which\nis slow, or more quickly simulate a neuron using a larger DT and incur a loss\nin simulation accuracy. Here we provide a solution to this dilemma, by\nalgorithmically reinterpreting the ALIF model, reducing the sequential\nsimulation complexity and permitting a more efficient parallelisation on GPUs.\nWe computationally validate our implementation to obtain over a $50\\times$\ntraining speedup using small DTs on synthetic benchmarks. We also obtained a\ncomparable performance to the standard ALIF implementation on different\nsupervised classification tasks - yet in a fraction of the training time.\nLastly, we showcase how our model makes it possible to quickly and accurately\nfit real electrophysiological recordings of cortical neurons, where very fine\nsub-millisecond DTs are crucial for capturing exact spike timing.\n", "date": "2023-11-19", "categories": []}
{"id": "2308.11809", "title": "Expressive probabilistic sampling in recurrent neural networks", "abstract": "  In sampling-based Bayesian models of brain function, neural activities are\nassumed to be samples from probability distributions that the brain uses for\nprobabilistic computation. However, a comprehensive understanding of how\nmechanistic models of neural dynamics can sample from arbitrary distributions\nis still lacking. We use tools from functional analysis and stochastic\ndifferential equations to explore the minimum architectural requirements for\n$\\textit{recurrent}$ neural circuits to sample from complex distributions. We\nfirst consider the traditional sampling model consisting of a network of\nneurons whose outputs directly represent the samples (sampler-only network). We\nargue that synaptic current and firing-rate dynamics in the traditional model\nhave limited capacity to sample from a complex probability distribution. We\nshow that the firing rate dynamics of a recurrent neural circuit with a\nseparate set of output units can sample from an arbitrary probability\ndistribution. We call such circuits reservoir-sampler networks (RSNs). We\npropose an efficient training procedure based on denoising score matching that\nfinds recurrent and output weights such that the RSN implements Langevin\nsampling. We empirically demonstrate our model's ability to sample from several\ncomplex data distributions using the proposed neural dynamics and discuss its\napplicability to developing the next generation of sampling-based brain models.\n", "date": "2023-08-22", "categories": []}
{"id": "2311.08113", "title": "Understanding learning from EEG data: Combining machine learning and\n  feature engineering based on hidden Markov models and mixed models", "abstract": "  Theta oscillations, ranging from 4-8 Hz, play a significant role in spatial\nlearning and memory functions during navigation tasks. Frontal theta\noscillations are thought to play an important role in spatial navigation and\nmemory. Electroencephalography (EEG) datasets are very complex, making any\nchanges in the neural signal related to behaviour difficult to interpret.\nHowever, multiple analytical methods are available to examine complex data\nstructure, especially machine learning based techniques. These methods have\nshown high classification performance and the combination with feature\nengineering enhances the capability of these methods. This paper proposes using\nhidden Markov and linear mixed effects models to extract features from EEG\ndata. Based on the engineered features obtained from frontal theta EEG data\nduring a spatial navigation task in two key trials (first, last) and between\ntwo conditions (learner and non-learner), we analysed the performance of six\nmachine learning methods (Polynomial Support Vector Machines, Non-linear\nSupport Vector Machines, Random Forests, K-Nearest Neighbours, Ridge, and Deep\nNeural Networks) on classifying learner and non-learner participants. We also\nanalysed how different standardisation methods used to pre-process the EEG data\ncontribute to classification performance. We compared the classification\nperformance of each trial with data gathered from the same subjects, including\nsolely coordinate-based features, such as idle time and average speed. We found\nthat more machine learning methods perform better classification using\ncoordinate-based data. However, only deep neural networks achieved an area\nunder the ROC curve higher than 80% using the theta EEG data alone. Our\nfindings suggest that standardising the theta EEG data and using deep neural\nnetworks enhances the classification of learner and non-learner subjects in a\nspatial learning task.\n", "date": "2023-11-14", "categories": []}
{"id": "2311.06417", "title": "Resolving uncertainty on the fly: Modeling adaptive driving behavior as\n  active inference", "abstract": "  Understanding adaptive human driving behavior, in particular how drivers\nmanage uncertainty, is of key importance for developing simulated human driver\nmodels that can be used in the evaluation and development of autonomous\nvehicles. However, existing traffic psychology models of adaptive driving\nbehavior either lack computational rigor or only address specific scenarios\nand/or behavioral phenomena. While models developed in the fields of machine\nlearning and robotics can effectively learn adaptive driving behavior from\ndata, due to their black box nature, they offer little or no explanation of the\nmechanisms underlying the adaptive behavior. Thus, a generalizable,\ninterpretable, computational model of adaptive human driving behavior is still\nlacking. This paper proposes such a model based on active inference, a\nbehavioral modeling framework originating in computational neuroscience. The\nmodel offers a principled solution to how humans trade progress against caution\nthrough policy selection based on the single mandate to minimize expected free\nenergy. This casts goal-seeking and information-seeking (uncertainty-resolving)\nbehavior under a single objective function, allowing the model to seamlessly\nresolve uncertainty as a means to obtain its goals. We apply the model in two\napparently disparate driving scenarios that require managing uncertainty, (1)\ndriving past an occluding object and (2) visual time sharing between driving\nand a secondary task, and show how human-like adaptive driving behavior emerges\nfrom the single principle of expected free energy minimization.\n", "date": "2023-11-10", "categories": []}
{"id": "2311.07563", "title": "Learning Control Policies of Hodgkin-Huxley Neuronal Dynamics", "abstract": "  We present a neural network approach for closed-loop deep brain stimulation\n(DBS). We cast the problem of finding an optimal neurostimulation strategy as a\ncontrol problem. In this setting, control policies aim to optimize therapeutic\noutcomes by tailoring the parameters of a DBS system, typically via electrical\nstimulation, in real time based on the patient's ongoing neuronal activity. We\napproximate the value function offline using a neural network to enable\ngenerating controls (stimuli) in real time via the feedback form. The neuronal\nactivity is characterized by a nonlinear, stiff system of differential\nequations as dictated by the Hodgkin-Huxley model. Our training process\nleverages the relationship between Pontryagin's maximum principle and\nHamilton-Jacobi-Bellman equations to update the value function estimates\nsimultaneously. Our numerical experiments illustrate the accuracy of our\napproach for out-of-distribution samples and the robustness to moderate shocks\nand disturbances in the system.\n", "date": "2023-11-13", "categories": []}
{"id": "2305.09156", "title": "Modelling Human Visual Motion Processing with Trainable Motion Energy\n  Sensing and a Self-attention Network", "abstract": "  Visual motion processing is essential for humans to perceive and interact\nwith dynamic environments. Despite extensive research in cognitive\nneuroscience, image-computable models that can extract informative motion flow\nfrom natural scenes in a manner consistent with human visual processing have\nyet to be established. Meanwhile, recent advancements in computer vision (CV),\npropelled by deep learning, have led to significant progress in optical flow\nestimation, a task closely related to motion perception. Here we propose an\nimage-computable model of human motion perception by bridging the gap between\nbiological and CV models. Specifically, we introduce a novel two-stages\napproach that combines trainable motion energy sensing with a recurrent\nself-attention network for adaptive motion integration and segregation. This\nmodel architecture aims to capture the computations in V1-MT, the core\nstructure for motion perception in the biological visual system, while\nproviding the ability to derive informative motion flow for a wide range of\nstimuli, including complex natural scenes. In silico neurophysiology reveals\nthat our model's unit responses are similar to mammalian neural recordings\nregarding motion pooling and speed tuning. The proposed model can also\nreplicate human responses to a range of stimuli examined in past psychophysical\nstudies. The experimental results on the Sintel benchmark demonstrate that our\nmodel predicts human responses better than the ground truth, whereas the\nstate-of-the-art CV models show the opposite. Our study provides a\ncomputational architecture consistent with human visual motion processing,\nalthough the physiological correspondence may not be exact.\n", "date": "2023-05-16", "categories": []}
{"id": "2311.03508", "title": "Astrocytes as a mechanism for meta-plasticity and contextually-guided\n  network function", "abstract": "  Astrocytes are a ubiquitous and enigmatic type of non-neuronal cell and are\nfound in the brain of all vertebrates. While traditionally viewed as being\nsupportive of neurons, it is increasingly recognized that astrocytes may play a\nmore direct and active role in brain function and neural computation. On\naccount of their sensitivity to a host of physiological covariates and ability\nto modulate neuronal activity and connectivity on slower time scales,\nastrocytes may be particularly well poised to modulate the dynamics of neural\ncircuits in functionally salient ways. In the current paper, we seek to capture\nthese features via actionable abstractions within computational models of\nneuron-astrocyte interaction. Specifically, we engage how nested feedback loops\nof neuron-astrocyte interaction, acting over separated time-scales may endow\nastrocytes with the capability to enable learning in context-dependent\nsettings, where fluctuations in task parameters may occur much more slowly than\nwithin-task requirements. We pose a general model of neuron-synapse-astrocyte\ninteraction and use formal analysis to characterize how astrocytic modulation\nmay constitute a form of meta-plasticity, altering the ways in which synapses\nand neurons adapt as a function of time. We then embed this model in a\nbandit-based reinforcement learning task environment, and show how the presence\nof time-scale separated astrocytic modulation enables learning over multiple\nfluctuating contexts. Indeed, these networks learn far more reliably versus\ndynamically homogeneous networks and conventional non-network-based bandit\nalgorithms. Our results indicate how the presence of neuron-astrocyte\ninteraction in the brain may benefit learning over different time-scales and\nthe conveyance of task-relevant contextual information onto circuit dynamics.\n", "date": "2023-11-06", "categories": []}
{"id": "2306.11582", "title": "Computing a human-like reaction time metric from stable recurrent vision\n  models", "abstract": "  The meteoric rise in the adoption of deep neural networks as computational\nmodels of vision has inspired efforts to \"align\" these models with humans. One\ndimension of interest for alignment includes behavioral choices, but moving\nbeyond characterizing choice patterns to capturing temporal aspects of visual\ndecision-making has been challenging. Here, we sketch a general-purpose\nmethodology to construct computational accounts of reaction times from a\nstimulus-computable, task-optimized model. Specifically, we introduce a novel\nmetric leveraging insights from subjective logic theory summarizing evidence\naccumulation in recurrent vision models. We demonstrate that our metric aligns\nwith patterns of human reaction times for stimulus manipulations across four\ndisparate visual decision-making tasks spanning perceptual grouping, mental\nsimulation, and scene categorization. This work paves the way for exploring the\ntemporal alignment of model and human visual strategies in the context of\nvarious other cognitive tasks toward generating testable hypotheses for\nneuroscience. Links to the code and data can be found on the project page:\nhttps://serre-lab.github.io/rnn_rts_site.\n", "date": "2023-06-20", "categories": []}
{"id": "2311.02258", "title": "Learning Time-Invariant Representations for Individual Neurons from\n  Population Dynamics", "abstract": "  Neurons can display highly variable dynamics. While such variability\npresumably supports the wide range of behaviors generated by the organism,\ntheir gene expressions are relatively stable in the adult brain. This suggests\nthat neuronal activity is a combination of its time-invariant identity and the\ninputs the neuron receives from the rest of the circuit. Here, we propose a\nself-supervised learning based method to assign time-invariant representations\nto individual neurons based on permutation-, and population size-invariant\nsummary of population recordings. We fit dynamical models to neuronal activity\nto learn a representation by considering the activity of both the individual\nand the neighboring population. Our self-supervised approach and use of\nimplicit representations enable robust inference against imperfections such as\npartial overlap of neurons across sessions, trial-to-trial variability, and\nlimited availability of molecular (transcriptomic) labels for downstream\nsupervised tasks. We demonstrate our method on a public multimodal dataset of\nmouse cortical neuronal activity and transcriptomic labels. We report > 35%\nimprovement in predicting the transcriptomic subclass identity and > 20%\nimprovement in predicting class identity with respect to the state-of-the-art.\n", "date": "2023-11-03", "categories": []}
{"id": "2311.02466", "title": "Multi-State Brain Network Discovery", "abstract": "  Brain network discovery aims to find nodes and edges from the spatio-temporal\nsignals obtained by neuroimaging data, such as fMRI scans of human brains.\nExisting methods tend to derive representative or average brain networks,\nassuming observed signals are generated by only a single brain activity state.\nHowever, the human brain usually involves multiple activity states, which\njointly determine the brain activities. The brain regions and their\nconnectivity usually exhibit intricate patterns that are difficult to capture\nwith only a single-state network. Recent studies find that brain parcellation\nand connectivity change according to the brain activity state. We refer to such\nbrain networks as multi-state, and this mixture can help us understand human\nbehavior. Thus, compared to a single-state network, a multi-state network can\nprevent us from losing crucial information of cognitive brain network. To\nachieve this, we propose a new model called MNGL (Multi-state Network Graphical\nLasso), which successfully models multi-state brain networks by combining CGL\n(coherent graphical lasso) with GMM (Gaussian Mixture Model). Using both\nsynthetic and real world ADHD 200 fMRI datasets, we demonstrate that MNGL\noutperforms recent state-of-the-art alternatives by discovering more\nexplanatory and realistic results.\n", "date": "2023-11-04", "categories": []}
{"id": "2311.02704", "title": "A Goal-Driven Approach to Systems Neuroscience", "abstract": "  Humans and animals exhibit a range of interesting behaviors in dynamic\nenvironments, and it is unclear how our brains actively reformat this dense\nsensory information to enable these behaviors. Experimental neuroscience is\nundergoing a revolution in its ability to record and manipulate hundreds to\nthousands of neurons while an animal is performing a complex behavior. As these\nparadigms enable unprecedented access to the brain, a natural question that\narises is how to distill these data into interpretable insights about how\nneural circuits give rise to intelligent behaviors. The classical approach in\nsystems neuroscience has been to ascribe well-defined operations to individual\nneurons and provide a description of how these operations combine to produce a\ncircuit-level theory of neural computations. While this approach has had some\nsuccess for small-scale recordings with simple stimuli, designed to probe a\nparticular circuit computation, often times these ultimately lead to disparate\ndescriptions of the same system across stimuli. Perhaps more strikingly, many\nresponse profiles of neurons are difficult to succinctly describe in words,\nsuggesting that new approaches are needed in light of these experimental\nobservations. In this thesis, we offer a different definition of\ninterpretability that we show has promise in yielding unified structural and\nfunctional models of neural circuits, and describes the evolutionary\nconstraints that give rise to the response properties of the neural population,\nincluding those that have previously been difficult to describe individually.\nWe demonstrate the utility of this framework across multiple brain areas and\nspecies to study the roles of recurrent processing in the primate ventral\nvisual pathway; mouse visual processing; heterogeneity in rodent medial\nentorhinal cortex; and facilitating biological learning.\n", "date": "2023-11-05", "categories": []}
{"id": "2310.18377", "title": "Large-scale Foundation Models and Generative AI for BigData Neuroscience", "abstract": "  Recent advances in machine learning have made revolutionary breakthroughs in\ncomputer games, image and natural language understanding, and scientific\ndiscovery. Foundation models and large-scale language models (LLMs) have\nrecently achieved human-like intelligence thanks to BigData. With the help of\nself-supervised learning (SSL) and transfer learning, these models may\npotentially reshape the landscapes of neuroscience research and make a\nsignificant impact on the future. Here we present a mini-review on recent\nadvances in foundation models and generative AI models as well as their\napplications in neuroscience, including natural language and speech, semantic\nmemory, brain-machine interfaces (BMIs), and data augmentation. We argue that\nthis paradigm-shift framework will open new avenues for many neuroscience\nresearch directions and discuss the accompanying challenges and opportunities.\n", "date": "2023-10-27", "categories": []}
{"id": "2305.18370", "title": "Explainable Brain Age Prediction using coVariance Neural Networks", "abstract": "  In computational neuroscience, there has been an increased interest in\ndeveloping machine learning algorithms that leverage brain imaging data to\nprovide estimates of \"brain age\" for an individual. Importantly, the\ndiscordance between brain age and chronological age (referred to as \"brain age\ngap\") can capture accelerated aging due to adverse health conditions and\ntherefore, can reflect increased vulnerability towards neurological disease or\ncognitive impairments. However, widespread adoption of brain age for clinical\ndecision support has been hindered due to lack of transparency and\nmethodological justifications in most existing brain age prediction algorithms.\nIn this paper, we leverage coVariance neural networks (VNN) to propose an\nexplanation-driven and anatomically interpretable framework for brain age\nprediction using cortical thickness features. Specifically, our brain age\nprediction framework extends beyond the coarse metric of brain age gap in\nAlzheimer's disease (AD) and we make two important observations: (i) VNNs can\nassign anatomical interpretability to elevated brain age gap in AD by\nidentifying contributing brain regions, (ii) the interpretability offered by\nVNNs is contingent on their ability to exploit specific eigenvectors of the\nanatomical covariance matrix. Together, these observations facilitate an\nexplainable and anatomically interpretable perspective to the task of brain age\nprediction.\n", "date": "2023-05-27", "categories": []}
{"id": "2308.13633", "title": "Adaptive whitening with fast gain modulation and slow synaptic\n  plasticity", "abstract": "  Neurons in early sensory areas rapidly adapt to changing sensory statistics,\nboth by normalizing the variance of their individual responses and by reducing\ncorrelations between their responses. Together, these transformations may be\nviewed as an adaptive form of statistical whitening. Existing mechanistic\nmodels of adaptive whitening exclusively use either synaptic plasticity or gain\nmodulation as the biological substrate for adaptation; however, on their own,\neach of these models has significant limitations. In this work, we unify these\napproaches in a normative multi-timescale mechanistic model that adaptively\nwhitens its responses with complementary computational roles for synaptic\nplasticity and gain modulation. Gains are modified on a fast timescale to adapt\nto the current statistical context, whereas synapses are modified on a slow\ntimescale to match structural properties of the input statistics that are\ninvariant across contexts. Our model is derived from a novel multi-timescale\nwhitening objective that factorizes the inverse whitening matrix into basis\nvectors, which correspond to synaptic weights, and a diagonal matrix, which\ncorresponds to neuronal gains. We test our model on synthetic and natural\ndatasets and find that the synapses learn optimal configurations over long\ntimescales that enable adaptive whitening on short timescales using gain\nmodulation.\n", "date": "2023-08-25", "categories": []}
{"id": "2305.11772", "title": "Neural Foundations of Mental Simulation: Future Prediction of Latent\n  Representations on Dynamic Scenes", "abstract": "  Humans and animals have a rich and flexible understanding of the physical\nworld, which enables them to infer the underlying dynamical trajectories of\nobjects and events, plausible future states, and use that to plan and\nanticipate the consequences of actions. However, the neural mechanisms\nunderlying these computations are unclear. We combine a goal-driven modeling\napproach with dense neurophysiological data and high-throughput human\nbehavioral readouts to directly impinge on this question. Specifically, we\nconstruct and evaluate several classes of sensory-cognitive networks to predict\nthe future state of rich, ethologically-relevant environments, ranging from\nself-supervised end-to-end models with pixel-wise or object-centric objectives,\nto models that future predict in the latent space of purely static image-based\nor dynamic video-based pretrained foundation models. We find strong\ndifferentiation across these model classes in their ability to predict neural\nand behavioral data both within and across diverse environments. In particular,\nwe find that neural responses are currently best predicted by models trained to\npredict the future state of their environment in the latent space of pretrained\nfoundation models optimized for dynamic scenes in a self-supervised manner.\nNotably, models that future predict in the latent space of video foundation\nmodels that are optimized to support a diverse range of sensorimotor tasks,\nreasonably match both human behavioral error patterns and neural dynamics\nacross all environmental scenarios that we were able to test. Overall, these\nfindings suggest that the neural mechanisms and behaviors of primate mental\nsimulation are thus far most consistent with being optimized to future predict\non dynamic, reusable visual representations that are useful for Embodied AI\nmore generally.\n", "date": "2023-05-19", "categories": []}
{"id": "2308.09312", "title": "Path Signatures for Seizure Forecasting", "abstract": "  Predicting future system behaviour from past observed behaviour (time series)\nis fundamental to science and engineering. In computational neuroscience, the\nprediction of future epileptic seizures from brain activity measurements, using\nEEG data, remains largely unresolved despite much dedicated research effort.\nBased on a longitudinal and state-of-the-art data set using intercranial EEG\nmeasurements from people with epilepsy, we consider the automated discovery of\npredictive features (or biomarkers) to forecast seizures in a patient-specific\nway. To this end, we use the path signature, a recent development in the\nanalysis of data streams, to map from measured time series to seizure\nprediction. The predictor is based on linear classification, here augmented\nwith sparsity constraints, to discern time series with and without an impending\nseizure. This approach may be seen as a step towards a generic pattern\nrecognition pipeline where the main advantages are simplicity and ease of\ncustomisation, while maintaining forecasting performance on par with modern\nmachine learning. Nevertheless, it turns out that although the path signature\nmethod has some powerful theoretical guarantees, appropriate time series\nstatistics can achieve essentially the same results in our context of seizure\nprediction. This suggests that, due to their inherent complexity and\nnon-stationarity, the brain's dynamics are not identifiable from the available\nEEG measurement data, and, more concretely, epileptic episode prediction is not\nreliably achieved using EEG measurement data alone.\n", "date": "2023-08-18", "categories": []}
{"id": "2310.15263", "title": "One-hot Generalized Linear Model for Switching Brain State Discovery", "abstract": "  Exposing meaningful and interpretable neural interactions is critical to\nunderstanding neural circuits. Inferred neural interactions from neural signals\nprimarily reflect functional interactions. In a long experiment, subject\nanimals may experience different stages defined by the experiment, stimuli, or\nbehavioral states, and hence functional interactions can change over time. To\nmodel dynamically changing functional interactions, prior work employs\nstate-switching generalized linear models with hidden Markov models (i.e.,\nHMM-GLMs). However, we argue they lack biological plausibility, as functional\ninteractions are shaped and confined by the underlying anatomical connectome.\nHere, we propose a novel prior-informed state-switching GLM. We introduce both\na Gaussian prior and a one-hot prior over the GLM in each state. The priors are\nlearnable. We will show that the learned prior should capture the\nstate-constant interaction, shedding light on the underlying anatomical\nconnectome and revealing more likely physical neuron interactions. The\nstate-dependent interaction modeled by each GLM offers traceability to capture\nfunctional variations across multiple brain states. Our methods effectively\nrecover true interaction structures in simulated data, achieve the highest\npredictive likelihood with real neural datasets, and render interaction\nstructures and hidden states more interpretable when applied to real neural\ndata.\n", "date": "2023-10-23", "categories": []}
{"id": "2310.16046", "title": "A Unified, Scalable Framework for Neural Population Decoding", "abstract": "  Our ability to use deep learning approaches to decipher neural activity would\nlikely benefit from greater scale, in terms of both model size and datasets.\nHowever, the integration of many neural recordings into one unified model is\nchallenging, as each recording contains the activity of different neurons from\ndifferent individual animals. In this paper, we introduce a training framework\nand architecture designed to model the population dynamics of neural activity\nacross diverse, large-scale neural recordings. Our method first tokenizes\nindividual spikes within the dataset to build an efficient representation of\nneural events that captures the fine temporal structure of neural activity. We\nthen employ cross-attention and a PerceiverIO backbone to further construct a\nlatent tokenization of neural population activities. Utilizing this\narchitecture and training framework, we construct a large-scale multi-session\nmodel trained on large datasets from seven nonhuman primates, spanning over 158\ndifferent sessions of recording from over 27,373 neural units and over 100\nhours of recordings. In a number of different tasks, we demonstrate that our\npretrained model can be rapidly adapted to new, unseen sessions with\nunspecified neuron correspondence, enabling few-shot performance with minimal\nlabels. This work presents a powerful new approach for building deep learning\ntools to analyze neural data and stakes out a clear path to training at scale.\n", "date": "2023-10-24", "categories": []}
{"id": "2304.06040", "title": "Inferring Population Dynamics in Macaque Cortex", "abstract": "  The proliferation of multi-unit cortical recordings over the last two\ndecades, especially in macaques and during motor-control tasks, has generated\ninterest in neural \"population dynamics\": the time evolution of neural activity\nacross a group of neurons working together. A good model of these dynamics\nshould be able to infer the activity of unobserved neurons within the same\npopulation and of the observed neurons at future times. Accordingly,\nPandarinath and colleagues have introduced a benchmark to evaluate models on\nthese two (and related) criteria: four data sets, each consisting of firing\nrates from a population of neurons, recorded from macaque cortex during\nmovement-related tasks. Here we show that simple, general-purpose architectures\nbased on recurrent neural networks (RNNs) outperform more \"bespoke\" models, and\nindeed outperform all published models on all four data sets in the benchmark.\nPerformance can be improved further still with a novel, hybrid architecture\nthat augments the RNN with self-attention, as in transformer networks. But pure\ntransformer models fail to achieve this level of performance, either in our\nwork or that of other groups. We argue that the autoregressive bias imposed by\nRNNs is critical for achieving the highest levels of performance. We conclude,\nhowever, by proposing that the benchmark be augmented with an alternative\nevaluation of latent dynamics that favors generative over discriminative models\nlike the ones we propose in this report.\n", "date": "2023-04-05", "categories": []}
{"id": "2310.11431", "title": "Identifying Interpretable Visual Features in Artificial and Biological\n  Neural Systems", "abstract": "  Single neurons in neural networks are often interpretable in that they\nrepresent individual, intuitively meaningful features. However, many neurons\nexhibit $\\textit{mixed selectivity}$, i.e., they represent multiple unrelated\nfeatures. A recent hypothesis proposes that features in deep networks may be\nrepresented in $\\textit{superposition}$, i.e., on non-orthogonal axes by\nmultiple neurons, since the number of possible interpretable features in\nnatural data is generally larger than the number of neurons in a given network.\nAccordingly, we should be able to find meaningful directions in activation\nspace that are not aligned with individual neurons. Here, we propose (1) an\nautomated method for quantifying visual interpretability that is validated\nagainst a large database of human psychophysics judgments of neuron\ninterpretability, and (2) an approach for finding meaningful directions in\nnetwork activation space. We leverage these methods to discover directions in\nconvolutional neural networks that are more intuitively meaningful than\nindividual neurons, as we confirm and investigate in a series of analyses.\nMoreover, we apply the same method to three recent datasets of visual neural\nresponses in the brain and find that our conclusions largely transfer to real\nneural data, suggesting that superposition might be deployed by the brain. This\nalso provides a link with disentanglement and raises fundamental questions\nabout robust, efficient and factorized representations in both artificial and\nbiological neural systems.\n", "date": "2023-10-17", "categories": []}
{"id": "2306.03812", "title": "Computation with Sequences in a Model of the Brain", "abstract": "  Even as machine learning exceeds human-level performance on many\napplications, the generality, robustness, and rapidity of the brain's learning\ncapabilities remain unmatched. How cognition arises from neural activity is a\ncentral open question in neuroscience, inextricable from the study of\nintelligence itself. A simple formal model of neural activity was proposed in\nPapadimitriou [2020] and has been subsequently shown, through both mathematical\nproofs and simulations, to be capable of implementing certain simple cognitive\noperations via the creation and manipulation of assemblies of neurons. However,\nmany intelligent behaviors rely on the ability to recognize, store, and\nmanipulate temporal sequences of stimuli (planning, language, navigation, to\nlist a few). Here we show that, in the same model, time can be captured\nnaturally as precedence through synaptic weights and plasticity, and, as a\nresult, a range of computations on sequences of assemblies can be carried out.\nIn particular, repeated presentation of a sequence of stimuli leads to the\nmemorization of the sequence through corresponding neural assemblies: upon\nfuture presentation of any stimulus in the sequence, the corresponding assembly\nand its subsequent ones will be activated, one after the other, until the end\nof the sequence. Finally, we show that any finite state machine can be learned\nin a similar way, through the presentation of appropriate patterns of\nsequences. Through an extension of this mechanism, the model can be shown to be\ncapable of universal computation. We support our analysis with a number of\nexperiments to probe the limits of learning in this model in key ways. Taken\ntogether, these results provide a concrete hypothesis for the basis of the\nbrain's remarkable abilities to compute and learn, with sequences playing a\nvital role.\n", "date": "2023-06-06", "categories": []}
{"id": "2310.03186", "title": "Inferring Inference", "abstract": "  Patterns of microcircuitry suggest that the brain has an array of repeated\ncanonical computational units. Yet neural representations are distributed, so\nthe relevant computations may only be related indirectly to single-neuron\ntransformations. It thus remains an open challenge how to define canonical\ndistributed computations. We integrate normative and algorithmic theories of\nneural computation into a mathematical framework for inferring canonical\ndistributed computations from large-scale neural activity patterns. At the\nnormative level, we hypothesize that the brain creates a structured internal\nmodel of its environment, positing latent causes that explain its sensory\ninputs, and uses those sensory inputs to infer the latent causes. At the\nalgorithmic level, we propose that this inference process is a nonlinear\nmessage-passing algorithm on a graph-structured model of the world. Given a\ntime series of neural activity during a perceptual inference task, our\nframework finds (i) the neural representation of relevant latent variables,\n(ii) interactions between these variables that define the brain's internal\nmodel of the world, and (iii) message-functions specifying the inference\nalgorithm. These targeted computational properties are then statistically\ndistinguishable due to the symmetries inherent in any canonical computation, up\nto a global transformation. As a demonstration, we simulate recordings for a\nmodel brain that implicitly implements an approximate inference algorithm on a\nprobabilistic graphical model. Given its external inputs and noisy neural\nactivity, we recover the latent variables, their neural representation and\ndynamics, and canonical message-functions. We highlight features of\nexperimental design needed to successfully extract canonical computations from\nneural data. Overall, this framework provides a new tool for discovering\ninterpretable structure in neural recordings.\n", "date": "2023-10-04", "categories": []}
{"id": "2310.07106", "title": "The Temporal Structure of Language Processing in the Human Brain\n  Corresponds to The Layered Hierarchy of Deep Language Models", "abstract": "  Deep Language Models (DLMs) provide a novel computational paradigm for\nunderstanding the mechanisms of natural language processing in the human brain.\nUnlike traditional psycholinguistic models, DLMs use layered sequences of\ncontinuous numerical vectors to represent words and context, allowing a\nplethora of emerging applications such as human-like text generation. In this\npaper we show evidence that the layered hierarchy of DLMs may be used to model\nthe temporal dynamics of language comprehension in the brain by demonstrating a\nstrong correlation between DLM layer depth and the time at which layers are\nmost predictive of the human brain. Our ability to temporally resolve\nindividual layers benefits from our use of electrocorticography (ECoG) data,\nwhich has a much higher temporal resolution than noninvasive methods like fMRI.\nUsing ECoG, we record neural activity from participants listening to a\n30-minute narrative while also feeding the same narrative to a high-performing\nDLM (GPT2-XL). We then extract contextual embeddings from the different layers\nof the DLM and use linear encoding models to predict neural activity. We first\nfocus on the Inferior Frontal Gyrus (IFG, or Broca's area) and then extend our\nmodel to track the increasing temporal receptive window along the linguistic\nprocessing hierarchy from auditory to syntactic and semantic areas. Our results\nreveal a connection between human language processing and DLMs, with the DLM's\nlayer-by-layer accumulation of contextual information mirroring the timing of\nneural activity in high-order language areas.\n", "date": "2023-10-11", "categories": []}
{"id": "2310.06578", "title": "Energy-Efficient Visual Search by Eye Movement and Low-Latency Spiking\n  Neural Network", "abstract": "  Human vision incorporates non-uniform resolution retina, efficient eye\nmovement strategy, and spiking neural network (SNN) to balance the requirements\nin visual field size, visual resolution, energy cost, and inference latency.\nThese properties have inspired interest in developing human-like computer\nvision. However, existing models haven't fully incorporated the three features\nof human vision, and their learned eye movement strategies haven't been\ncompared with human's strategy, making the models' behavior difficult to\ninterpret. Here, we carry out experiments to examine human visual search\nbehaviors and establish the first SNN-based visual search model. The model\ncombines an artificial retina with spiking feature extraction, memory, and\nsaccade decision modules, and it employs population coding for fast and\nefficient saccade decisions. The model can learn either a human-like or a\nnear-optimal fixation strategy, outperform humans in search speed and accuracy,\nand achieve high energy efficiency through short saccade decision latency and\nsparse activation. It also suggests that the human search strategy is\nsuboptimal in terms of search speed. Our work connects modeling of vision in\nneuroscience and machine learning and sheds light on developing more\nenergy-efficient computer vision algorithms.\n", "date": "2023-10-10", "categories": []}
{"id": "2305.18274", "title": "Reconstructing the Mind's Eye: fMRI-to-Image with Contrastive Learning\n  and Diffusion Priors", "abstract": "  We present MindEye, a novel fMRI-to-image approach to retrieve and\nreconstruct viewed images from brain activity. Our model comprises two parallel\nsubmodules that are specialized for retrieval (using contrastive learning) and\nreconstruction (using a diffusion prior). MindEye can map fMRI brain activity\nto any high dimensional multimodal latent space, like CLIP image space,\nenabling image reconstruction using generative models that accept embeddings\nfrom this latent space. We comprehensively compare our approach with other\nexisting methods, using both qualitative side-by-side comparisons and\nquantitative evaluations, and show that MindEye achieves state-of-the-art\nperformance in both reconstruction and retrieval tasks. In particular, MindEye\ncan retrieve the exact original image even among highly similar candidates\nindicating that its brain embeddings retain fine-grained image-specific\ninformation. This allows us to accurately retrieve images even from large-scale\ndatabases like LAION-5B. We demonstrate through ablations that MindEye's\nperformance improvements over previous methods result from specialized\nsubmodules for retrieval and reconstruction, improved training techniques, and\ntraining models with orders of magnitude more parameters. Furthermore, we show\nthat MindEye can better preserve low-level image features in the\nreconstructions by using img2img, with outputs from a separate autoencoder. All\ncode is available on GitHub.\n", "date": "2023-05-29", "categories": []}
{"id": "2310.04460", "title": "Tuning In to Neural Encoding: Linking Human Brain and Artificial\n  Supervised Representations of Language", "abstract": "  To understand the algorithm that supports the human brain's language\nrepresentation, previous research has attempted to predict neural responses to\nlinguistic stimuli using embeddings generated by artificial neural networks\n(ANNs), a process known as neural encoding. However, most of these studies have\nfocused on probing neural representations of Germanic languages, such as\nEnglish, with unsupervised ANNs. In this paper, we propose to bridge the gap\nbetween human brain and supervised ANN representations of the Chinese language.\nSpecifically, we investigate how task tuning influences a pretained Transformer\nfor neural encoding and which tasks lead to the best encoding performances. We\ngenerate supervised representations on eight Natural Language Understanding\n(NLU) tasks using prompt-tuning, a technique that is seldom explored in neural\nencoding for language. We demonstrate that prompt-tuning yields representations\nthat better predict neural responses to Chinese stimuli than traditional\nfine-tuning on four tasks. Furthermore, we discover that tasks that require a\nfine-grained processing of concepts and entities lead to representations that\nare most predictive of brain activation patterns. Additionally, we reveal that\nthe proportion of tuned parameters highly influences the neural encoding\nperformance of fine-tuned models. Overall, our experimental findings could help\nus better understand the relationship between supervised artificial and brain\nlanguage representations.\n", "date": "2023-10-05", "categories": []}
{"id": "2309.16687", "title": "Duality Principle and Biologically Plausible Learning: Connecting the\n  Representer Theorem and Hebbian Learning", "abstract": "  A normative approach called Similarity Matching was recently introduced for\nderiving and understanding the algorithmic basis of neural computation focused\non unsupervised problems. It involves deriving algorithms from computational\nobjectives and evaluating their compatibility with anatomical and physiological\nobservations. In particular, it introduces neural architectures by considering\ndual alternatives instead of primal formulations of popular models such as PCA.\nHowever, its connection to the Representer theorem remains unexplored. In this\nwork, we propose to use teachings from this approach to explore supervised\nlearning algorithms and clarify the notion of Hebbian learning. We examine\nregularized supervised learning and elucidate the emergence of neural\narchitecture and additive versus multiplicative update rules. In this work, we\nfocus not on developing new algorithms but on showing that the Representer\ntheorem offers the perfect lens to study biologically plausible learning\nalgorithms. We argue that many past and current advancements in the field rely\non some form of dual formulation to introduce biological plausibility. In\nshort, as long as a dual formulation exists, it is possible to derive\nbiologically plausible algorithms. Our work sheds light on the pivotal role of\nthe Representer theorem in advancing our comprehension of neural computation.\n", "date": "2023-08-02", "categories": []}
{"id": "2309.16704", "title": "Memories in the Making: Predicting Video Memorability with Encoding\n  Phase EEG", "abstract": "  In a world of ephemeral moments, our brain diligently sieves through a\ncascade of experiences, like a skilled gold prospector searching for precious\nnuggets amidst the river's relentless flow. This study delves into the elusive\n\"moment of memorability\" -- a fleeting, yet vital instant where experiences are\nprioritised for consolidation in our memory. By transforming subjects' encoding\nphase electroencephalography (EEG) signals into the visual domain using\nscaleograms and leveraging deep learning techniques, we investigate the neural\nsignatures that underpin this moment, with the aim of predicting\nsubject-specific recognition of video. Our findings not only support the\ninvolvement of theta band (4-8Hz) oscillations over the right temporal lobe in\nthe encoding of declarative memory, but also support the existence of a\ndistinct moment of memorability, akin to the gold nuggets that define our\npersonal river of experiences.\n", "date": "2023-08-16", "categories": []}
{"id": "2309.15018", "title": "Unidirectional brain-computer interface: Artificial neural network\n  encoding natural images to fMRI response in the visual cortex", "abstract": "  While significant advancements in artificial intelligence (AI) have catalyzed\nprogress across various domains, its full potential in understanding visual\nperception remains underexplored. We propose an artificial neural network\ndubbed VISION, an acronym for \"Visual Interface System for Imaging Output of\nNeural activity,\" to mimic the human brain and show how it can foster\nneuroscientific inquiries. Using visual and contextual inputs, this multimodal\nmodel predicts the brain's functional magnetic resonance imaging (fMRI) scan\nresponse to natural images. VISION successfully predicts human hemodynamic\nresponses as fMRI voxel values to visual inputs with an accuracy exceeding\nstate-of-the-art performance by 45%. We further probe the trained networks to\nreveal representational biases in different visual areas, generate\nexperimentally testable hypotheses, and formulate an interpretable metric to\nassociate these hypotheses with cortical functions. With both a model and\nevaluation metric, the cost and time burdens associated with designing and\nimplementing functional analysis on the visual cortex could be reduced. Our\nwork suggests that the evolution of computational models may shed light on our\nfundamental understanding of the visual cortex and provide a viable approach\ntoward reliable brain-machine interfaces.\n", "date": "2023-09-26", "categories": []}
{"id": "2309.15090", "title": "Single Biological Neurons as Temporally Precise Spatio-Temporal Pattern\n  Recognizers", "abstract": "  This PhD thesis is focused on the central idea that single neurons in the\nbrain should be regarded as temporally precise and highly complex\nspatio-temporal pattern recognizers. This is opposed to the prevalent view of\nbiological neurons as simple and mainly spatial pattern recognizers by most\nneuroscientists today. In this thesis, I will attempt to demonstrate that this\nis an important distinction, predominantly because the above-mentioned\ncomputational properties of single neurons have far-reaching implications with\nrespect to the various brain circuits that neurons compose, and on how\ninformation is encoded by neuronal activity in the brain. Namely, that these\nparticular \"low-level\" details at the single neuron level have substantial\nsystem-wide ramifications. In the introduction we will highlight the main\ncomponents that comprise a neural microcircuit that can perform useful\ncomputations and illustrate the inter-dependence of these components from a\nsystem perspective. In chapter 1 we discuss the great complexity of the\nspatio-temporal input-output relationship of cortical neurons that are the\nresult of morphological structure and biophysical properties of the neuron. In\nchapter 2 we demonstrate that single neurons can generate temporally precise\noutput patterns in response to specific spatio-temporal input patterns with a\nvery simple biologically plausible learning rule. In chapter 3, we use the\ndifferentiable deep network analog of a realistic cortical neuron as a tool to\napproximate the gradient of the output of the neuron with respect to its input\nand use this capability in an attempt to teach the neuron to perform nonlinear\nXOR operation. In chapter 4 we expand chapter 3 to describe extension of our\nideas to neuronal networks composed of many realistic biological spiking\nneurons that represent either small microcircuits or entire brain regions.\n", "date": "2023-09-26", "categories": []}
{"id": "2308.13546", "title": "Functional Graph Contrastive Learning of Hyperscanning EEG Reveals\n  Emotional Contagion Evoked by Stereotype-Based Stressors", "abstract": "  This study delves into the intricacies of emotional contagion and its impact\non performance within dyadic interactions. Specifically, it focuses on the\ncontext of stereotype-based stress (SBS) during collaborative problem-solving\ntasks among female pairs. Through an exploration of emotional contagion, this\nstudy seeks to unveil its underlying mechanisms and effects. Leveraging\nEEG-based hyperscanning technology, we introduced an innovative approach known\nas the functional Graph Contrastive Learning (fGCL), which extracts\nsubject-invariant representations of neural activity patterns from feedback\ntrials. These representations are further subjected to analysis using the\nDynamic Graph Classification (DGC) model, aimed at dissecting the process of\nemotional contagion along three independent temporal stages. The results\nunderscore the substantial role of emotional contagion in shaping the\ntrajectories of participants' performance during collaborative tasks in the\npresence of SBS conditions. Overall, our research contributes invaluable\ninsights into the neural underpinnings of emotional contagion, thereby\nenriching our comprehension of the complexities underlying social interactions\nand emotional dynamics.\n", "date": "2023-08-22", "categories": []}
{"id": "2208.08478", "title": "\"Task-relevant autoencoding\" enhances machine learning for human\n  neuroscience", "abstract": "  In human neuroscience, machine learning can help reveal lower-dimensional\nneural representations relevant to subjects' behavior. However,\nstate-of-the-art models typically require large datasets to train, so are prone\nto overfitting on human neuroimaging data that often possess few samples but\nmany input dimensions. Here, we capitalized on the fact that the features we\nseek in human neuroscience are precisely those relevant to subjects' behavior.\nWe thus developed a Task-Relevant Autoencoder via Classifier Enhancement\n(TRACE), and tested its ability to extract behaviorally-relevant, separable\nrepresentations compared to a standard autoencoder, a variational autoencoder,\nand principal component analysis for two severely truncated machine learning\ndatasets. We then evaluated all models on fMRI data from 59 subjects who\nobserved animals and objects. TRACE outperformed all models nearly\nunilaterally, showing up to 12% increased classification accuracy and up to 56%\nimprovement in discovering \"cleaner\", task-relevant representations. These\nresults showcase TRACE's potential for a wide variety of data related to human\nbehavior.\n", "date": "2022-08-17", "categories": []}
{"id": "2309.09983", "title": "Exploration and Comparison of Deep Learning Architectures to Predict\n  Brain Response to Realistic Pictures", "abstract": "  We present an exploration of machine learning architectures for predicting\nbrain responses to realistic images on occasion of the Algonauts Challenge\n2023. Our research involved extensive experimentation with various pretrained\nmodels. Initially, we employed simpler models to predict brain activity but\ngradually introduced more complex architectures utilizing available data and\nembeddings generated by large-scale pre-trained models. We encountered typical\ndifficulties related to machine learning problems, e.g. regularization and\noverfitting, as well as issues specific to the challenge, such as difficulty in\ncombining multiple input encodings, as well as the high dimensionality, unclear\nstructure, and noisy nature of the output. To overcome these issues we tested\nsingle edge 3D position-based, multi-region of interest (ROI) and hemisphere\npredictor models, but we found that employing multiple simple models, each\ndedicated to a ROI in each hemisphere of the brain of each subject, yielded the\nbest results - a single fully connected linear layer with image embeddings\ngenerated by CLIP as input. While we surpassed the challenge baseline, our\nresults fell short of establishing a robust association with the data.\n", "date": "2023-09-11", "categories": []}
{"id": "2309.07149", "title": "Decoding visual brain representations from electroencephalography\n  through Knowledge Distillation and latent diffusion models", "abstract": "  Decoding visual representations from human brain activity has emerged as a\nthriving research domain, particularly in the context of brain-computer\ninterfaces. Our study presents an innovative method that employs to classify\nand reconstruct images from the ImageNet dataset using electroencephalography\n(EEG) data from subjects that had viewed the images themselves (i.e. \"brain\ndecoding\"). We analyzed EEG recordings from 6 participants, each exposed to 50\nimages spanning 40 unique semantic categories. These EEG readings were\nconverted into spectrograms, which were then used to train a convolutional\nneural network (CNN), integrated with a knowledge distillation procedure based\non a pre-trained Contrastive Language-Image Pre-Training (CLIP)-based image\nclassification teacher network. This strategy allowed our model to attain a\ntop-5 accuracy of 80%, significantly outperforming a standard CNN and various\nRNN-based benchmarks. Additionally, we incorporated an image reconstruction\nmechanism based on pre-trained latent diffusion models, which allowed us to\ngenerate an estimate of the images which had elicited EEG activity. Therefore,\nour architecture not only decodes images from neural activity but also offers a\ncredible image reconstruction from EEG only, paving the way for e.g. swift,\nindividualized feedback experiments. Our research represents a significant step\nforward in connecting neural signals with visual cognition.\n", "date": "2023-09-08", "categories": []}
{"id": "2309.05809", "title": "Divergences in Color Perception between Deep Neural Networks and Humans", "abstract": "  Deep neural networks (DNNs) are increasingly proposed as models of human\nvision, bolstered by their impressive performance on image classification and\nobject recognition tasks. Yet, the extent to which DNNs capture fundamental\naspects of human vision such as color perception remains unclear. Here, we\ndevelop novel experiments for evaluating the perceptual coherence of color\nembeddings in DNNs, and we assess how well these algorithms predict human color\nsimilarity judgments collected via an online survey. We find that\nstate-of-the-art DNN architectures $-$ including convolutional neural networks\nand vision transformers $-$ provide color similarity judgments that strikingly\ndiverge from human color judgments of (i) images with controlled color\nproperties, (ii) images generated from online searches, and (iii) real-world\nimages from the canonical CIFAR-10 dataset. We compare DNN performance against\nan interpretable and cognitively plausible model of color perception based on\nwavelet decomposition, inspired by foundational theories in computational\nneuroscience. While one deep learning model $-$ a convolutional DNN trained on\na style transfer task $-$ captures some aspects of human color perception, our\nwavelet algorithm provides more coherent color embeddings that better predict\nhuman color judgments compared to all DNNs we examine. These results hold when\naltering the high-level visual task used to train similar DNN architectures\n(e.g., image classification versus image segmentation), as well as when\nexamining the color embeddings of different layers in a given DNN architecture.\nThese findings break new ground in the effort to analyze the perceptual\nrepresentations of machine learning algorithms and to improve their ability to\nserve as cognitively plausible models of human vision. Implications for machine\nlearning, human perception, and embodied cognition are discussed.\n", "date": "2023-09-11", "categories": []}
{"id": "2302.03023", "title": "V1T: large-scale mouse V1 response prediction using a Vision Transformer", "abstract": "  Accurate predictive models of the visual cortex neural response to natural\nvisual stimuli remain a challenge in computational neuroscience. In this work,\nwe introduce V1T, a novel Vision Transformer based architecture that learns a\nshared visual and behavioral representation across animals. We evaluate our\nmodel on two large datasets recorded from mouse primary visual cortex and\noutperform previous convolution-based models by more than 12.7% in prediction\nperformance. Moreover, we show that the self-attention weights learned by the\nTransformer correlate with the population receptive fields. Our model thus sets\na new benchmark for neural response prediction and can be used jointly with\nbehavioral and neural recordings to reveal meaningful characteristic features\nof the visual cortex.\n", "date": "2023-02-06", "categories": []}
{"id": "2309.01230", "title": "lfads-torch: A modular and extensible implementation of latent factor\n  analysis via dynamical systems", "abstract": "  Latent factor analysis via dynamical systems (LFADS) is an RNN-based\nvariational sequential autoencoder that achieves state-of-the-art performance\nin denoising high-dimensional neural activity for downstream applications in\nscience and engineering. Recently introduced variants and extensions continue\nto demonstrate the applicability of the architecture to a wide variety of\nproblems in neuroscience. Since the development of the original implementation\nof LFADS, new technologies have emerged that use dynamic computation graphs,\nminimize boilerplate code, compose model configuration files, and simplify\nlarge-scale training. Building on these modern Python libraries, we introduce\nlfads-torch -- a new open-source implementation of LFADS that unifies existing\nvariants and is designed to be easier to understand, configure, and extend.\nDocumentation, source code, and issue tracking are available at\nhttps://github.com/arsedler9/lfads-torch .\n", "date": "2023-09-03", "categories": []}
{"id": "2302.06677", "title": "System identification of neural systems: If we got it right, would we\n  know?", "abstract": "  Artificial neural networks are being proposed as models of parts of the\nbrain. The networks are compared to recordings of biological neurons, and good\nperformance in reproducing neural responses is considered to support the\nmodel's validity. A key question is how much this system identification\napproach tells us about brain computation. Does it validate one model\narchitecture over another? We evaluate the most commonly used comparison\ntechniques, such as a linear encoding model and centered kernel alignment, to\ncorrectly identify a model by replacing brain recordings with known ground\ntruth models. System identification performance is quite variable; it also\ndepends significantly on factors independent of the ground truth architecture,\nsuch as stimuli images. In addition, we show the limitations of using\nfunctional similarity scores in identifying higher-level architectural motifs.\n", "date": "2023-02-13", "categories": []}
{"id": "2308.13870", "title": "Brain-like representational straightening of natural movies in robust\n  feedforward neural networks", "abstract": "  Representational straightening refers to a decrease in curvature of visual\nfeature representations of a sequence of frames taken from natural movies.\nPrior work established straightening in neural representations of the primate\nprimary visual cortex (V1) and perceptual straightening in human behavior as a\nhallmark of biological vision in contrast to artificial feedforward neural\nnetworks which did not demonstrate this phenomenon as they were not explicitly\noptimized to produce temporally predictable movie representations. Here, we\nshow robustness to noise in the input image can produce representational\nstraightening in feedforward neural networks. Both adversarial training (AT)\nand base classifiers for Random Smoothing (RS) induced remarkably straightened\nfeature codes. Demonstrating their utility within the domain of natural movies,\nthese codes could be inverted to generate intervening movie frames by linear\ninterpolation in the feature space even though they were not trained on these\ntrajectories. Demonstrating their biological utility, we found that AT and RS\ntraining improved predictions of neural data in primate V1 over baseline models\nproviding a parsimonious, bio-plausible mechanism -- noise in the sensory input\nstages -- for generating representations in early visual cortex. Finally, we\ncompared the geometric properties of frame representations in these networks to\nbetter understand how they produced representations that mimicked the\nstraightening phenomenon from biology. Overall, this work elucidating emergent\nproperties of robust neural networks demonstrates that it is not necessary to\nutilize predictive objectives or train directly on natural movie statistics to\nachieve models supporting straightened movie representations similar to human\nperception that also predict V1 neural responses.\n", "date": "2023-08-26", "categories": []}
{"id": "2208.06681", "title": "Modeling biological face recognition with deep convolutional neural\n  networks", "abstract": "  Deep convolutional neural networks (DCNNs) have become the state-of-the-art\ncomputational models of biological object recognition. Their remarkable success\nhas helped vision science break new ground and recent efforts have started to\ntransfer this achievement to research on biological face recognition. In this\nregard, face detection can be investigated by comparing face-selective\nbiological neurons and brain areas to artificial neurons and model layers.\nSimilarly, face identification can be examined by comparing in vivo and in\nsilico multidimensional \"face spaces\". In this review, we summarize the first\nstudies that use DCNNs to model biological face recognition. On the basis of a\nbroad spectrum of behavioral and computational evidence, we conclude that DCNNs\nare useful models that closely resemble the general hierarchical organization\nof face recognition in the ventral visual pathway and the core face network. In\ntwo exemplary spotlights, we emphasize the unique scientific contributions of\nthese models. First, studies on face detection in DCNNs indicate that\nelementary face selectivity emerges automatically through feedforward\nprocessing even in the absence of visual experience. Second, studies on face\nidentification in DCNNs suggest that identity-specific experience and\ngenerative mechanisms facilitate this particular challenge. Taken together, as\nthis novel modeling approach enables close control of predisposition (i.e.,\narchitecture) and experience (i.e., training data), it may be suited to inform\nlong-standing debates on the substrates of biological face recognition.\n", "date": "2022-08-13", "categories": []}
{"id": "2308.09431", "title": "End-to-end topographic networks as models of cortical map formation and\n  human visual behaviour: moving beyond convolutions", "abstract": "  Computational models are an essential tool for understanding the origin and\nfunctions of the topographic organisation of the primate visual system. Yet,\nvision is most commonly modelled by convolutional neural networks that ignore\ntopography by learning identical features across space. Here, we overcome this\nlimitation by developing All-Topographic Neural Networks (All-TNNs). Trained on\nvisual input, several features of primate topography emerge in All-TNNs: smooth\norientation maps and cortical magnification in their first layer, and\ncategory-selective areas in their final layer. In addition, we introduce a\nnovel dataset of human spatial biases in object recognition, which enables us\nto directly link models to behaviour. We demonstrate that All-TNNs\nsignificantly better align with human behaviour than previous state-of-the-art\nconvolutional models due to their topographic nature. All-TNNs thereby mark an\nimportant step forward in understanding the spatial organisation of the visual\nbrain and how it mediates visual behaviour.\n", "date": "2023-08-18", "categories": []}
{"id": "2308.02510", "title": "Seeing through the Brain: Image Reconstruction of Visual Perception from\n  Human Brain Signals", "abstract": "  Seeing is believing, however, the underlying mechanism of how human visual\nperceptions are intertwined with our cognitions is still a mystery. Thanks to\nthe recent advances in both neuroscience and artificial intelligence, we have\nbeen able to record the visually evoked brain activities and mimic the visual\nperception ability through computational approaches. In this paper, we pay\nattention to visual stimuli reconstruction by reconstructing the observed\nimages based on portably accessible brain signals, i.e., electroencephalography\n(EEG) data. Since EEG signals are dynamic in the time-series format and are\nnotorious to be noisy, processing and extracting useful information requires\nmore dedicated efforts; In this paper, we propose a comprehensive pipeline,\nnamed NeuroImagen, for reconstructing visual stimuli images from EEG signals.\nSpecifically, we incorporate a novel multi-level perceptual information\ndecoding to draw multi-grained outputs from the given EEG data. A latent\ndiffusion model will then leverage the extracted information to reconstruct the\nhigh-resolution visual stimuli images. The experimental results have\nillustrated the effectiveness of image reconstruction and superior quantitative\nperformance of our proposed method.\n", "date": "2023-07-27", "categories": []}
{"id": "2306.01991", "title": "A Bio-Inspired Chaos Sensor Model Based on the Perceptron Neural\n  Network: Machine Learning Concept and Application for Computational\n  Neuro-Science", "abstract": "  The study presents a bio-inspired chaos sensor model based on the perceptron\nneural network for the estimation of entropy of spike train in neurodynamic\nsystems. After training, the sensor on perceptron, having 50 neurons in the\nhidden layer and 1 neuron at the output, approximates the fuzzy entropy of a\nshort time series with high accuracy, with a determination coefficient of R2 ~\n0.9. The Hindmarsh-Rose spike model was used to generate time series of spike\nintervals, and datasets for training and testing the perceptron. The selection\nof the hyperparameters of the perceptron model and the estimation of the sensor\naccuracy were performed using the K-block cross-validation method. Even for a\nhidden layer with one neuron, the model approximates the fuzzy entropy with\ngood results and the metric R2 ~ 0.5-0.8. In a simplified model with one neuron\nand equal weights in the first layer, the principle of approximation is based\non the linear transformation of the average value of the time series into the\nentropy value. An example of using the chaos sensor on spike train of action\npotential recordings from the L5 dorsal rootlet of rat is provided. The\nbio-inspired chaos sensor model based on an ensemble of neurons is able to\ndynamically track the chaotic behavior of a spike signal and transmit this\ninformation to other parts of the neurodynamic model for further processing.\nThe study will be useful for specialists in the field of computational\nneuroscience, and also to create humanoid and animal robots, and bio-robots\nwith limited resources.\n", "date": "2023-06-03", "categories": []}
{"id": "2308.06443", "title": "Neural Latent Aligner: Cross-trial Alignment for Learning\n  Representations of Complex, Naturalistic Neural Data", "abstract": "  Understanding the neural implementation of complex human behaviors is one of\nthe major goals in neuroscience. To this end, it is crucial to find a true\nrepresentation of the neural data, which is challenging due to the high\ncomplexity of behaviors and the low signal-to-ratio (SNR) of the signals. Here,\nwe propose a novel unsupervised learning framework, Neural Latent Aligner\n(NLA), to find well-constrained, behaviorally relevant neural representations\nof complex behaviors. The key idea is to align representations across repeated\ntrials to learn cross-trial consistent information. Furthermore, we propose a\nnovel, fully differentiable time warping model (TWM) to resolve the temporal\nmisalignment of trials. When applied to intracranial electrocorticography\n(ECoG) of natural speaking, our model learns better representations for\ndecoding behaviors than the baseline models, especially in lower dimensional\nspace. The TWM is empirically validated by measuring behavioral coherence\nbetween aligned trials. The proposed framework learns more cross-trial\nconsistent representations than the baselines, and when visualized, the\nmanifold reveals shared neural trajectories across trials.\n", "date": "2023-08-12", "categories": []}
{"id": "2307.13181", "title": "Neural Memory Decoding with EEG Data and Representation Learning", "abstract": "  We describe a method for the neural decoding of memory from EEG data. Using\nthis method, a concept being recalled can be identified from an EEG trace with\nan average top-1 accuracy of about 78.4% (chance 4%). The method employs deep\nrepresentation learning with supervised contrastive loss to map an EEG\nrecording of brain activity to a low-dimensional space. Because representation\nlearning is used, concepts can be identified even if they do not appear in the\ntraining data set. However, reference EEG data must exist for each such\nconcept. We also show an application of the method to the problem of\ninformation retrieval. In neural information retrieval, EEG data is captured\nwhile a user recalls the contents of a document, and a list of links to\npredicted documents is produced.\n", "date": "2023-07-25", "categories": []}
{"id": "2308.01175", "title": "Memory Encoding Model", "abstract": "  We explore a new class of brain encoding model by adding memory-related\ninformation as input. Memory is an essential brain mechanism that works\nalongside visual stimuli. During a vision-memory cognitive task, we found the\nnon-visual brain is largely predictable using previously seen images. Our\nMemory Encoding Model (Mem) won the Algonauts 2023 visual brain competition\neven without model ensemble (single model score 66.8, ensemble score 70.8). Our\nensemble model without memory input (61.4) can also stand a 3rd place.\nFurthermore, we observe periodic delayed brain response correlated to 6th-7th\nprior image, and hippocampus also showed correlated activity timed with this\nperiodicity. We conjuncture that the periodic replay could be related to memory\nmechanism to enhance the working memory.\n", "date": "2023-08-02", "categories": []}
{"id": "2308.00262", "title": "The Algonauts Project 2023 Challenge: UARK-UAlbany Team Solution", "abstract": "  This work presents our solutions to the Algonauts Project 2023 Challenge. The\nprimary objective of the challenge revolves around employing computational\nmodels to anticipate brain responses captured during participants' observation\nof intricate natural visual scenes. The goal is to predict brain responses\nacross the entire visual brain, as it is the region where the most reliable\nresponses to images have been observed. We constructed an image-based brain\nencoder through a two-step training process to tackle this challenge.\nInitially, we created a pretrained encoder using data from all subjects. Next,\nwe proceeded to fine-tune individual subjects. Each step employed different\ntraining strategies, such as different loss functions and objectives, to\nintroduce diversity. Ultimately, our solution constitutes an ensemble of\nmultiple unique encoders. The code is available at\nhttps://github.com/uark-cviu/Algonauts2023\n", "date": "2023-08-01", "categories": []}
{"id": "2308.00526", "title": "Visual attention information can be traced on cortical response but not\n  on the retina: evidence from electrophysiological mouse data using natural\n  images as stimuli", "abstract": "  Visual attention forms the basis of understanding the visual world. In this\nwork we follow a computational approach to investigate the biological basis of\nvisual attention. We analyze retinal and cortical electrophysiological data\nfrom mouse. Visual Stimuli are Natural Images depicting real world scenes. Our\nresults show that in primary visual cortex (V1), a subset of around $10\\%$ of\nthe neurons responds differently to salient versus non-salient visual regions.\nVisual attention information was not traced in retinal response. It appears\nthat the retina remains naive concerning visual attention; cortical response\ngets modulated to interpret visual attention information. Experimental animal\nstudies may be designed to further explore the biological basis of visual\nattention we traced in this study. In applied and translational science, our\nstudy contributes to the design of improved visual prostheses systems --\nsystems that create artificial visual percepts to visually impaired individuals\nby electronic implants placed on either the retina or the cortex.\n", "date": "2023-08-01", "categories": []}
{"id": "2308.00678", "title": "Applicability of scaling laws to vision encoding models", "abstract": "  In this paper, we investigated how to build a high-performance vision\nencoding model to predict brain activity as part of our participation in the\nAlgonauts Project 2023 Challenge. The challenge provided brain activity\nrecorded by functional MRI (fMRI) while participants viewed images. Several\nvision models with parameter sizes ranging from 86M to 4.3B were used to build\npredictive models. To build highly accurate models, we focused our analysis on\ntwo main aspects: (1) How does the sample size of the fMRI training set change\nthe prediction accuracy? (2) How does the prediction accuracy across the visual\ncortex vary with the parameter size of the vision models? The results show that\nas the sample size used during training increases, the prediction accuracy\nimproves according to the scaling law. Similarly, we found that as the\nparameter size of the vision models increases, the prediction accuracy improves\naccording to the scaling law. These results suggest that increasing the sample\nsize of the fMRI training set and the parameter size of visual models may\ncontribute to more accurate visual models of the brain and lead to a better\nunderstanding of visual neuroscience.\n", "date": "2023-08-01", "categories": []}
{"id": "2307.14021", "title": "Retinotopy Inspired Brain Encoding Model and the All-for-One Training\n  Recipe", "abstract": "  Brain encoding models aim to predict brain voxel-wise responses to stimuli\nimages, replicating brain signals captured by neuroimaging techniques. There is\na large volume of publicly available data, but training a comprehensive brain\nencoding model is challenging. The main difficulties stem from a) diversity\nwithin individual brain, with functional heterogeneous brain regions; b)\ndiversity of brains from different subjects, due to genetic and developmental\ndifferences; c) diversity of imaging modalities and processing pipelines. We\nuse this diversity to our advantage by introducing the All-for-One training\nrecipe, which divides the challenging one-big-model problem into multiple small\nmodels, with the small models aggregating the knowledge while preserving the\ndistinction between the different functional regions. Agnostic of the training\nrecipe, we use biological knowledge of the brain, specifically retinotopy, to\nintroduce inductive bias to learn a 3D brain-to-image mapping that ensures a)\neach neuron knows which image regions and semantic levels to gather\ninformation, and b) no neurons are left behind in the model.\n  We pre-trained a brain encoding model using over one million data points from\nfive public datasets spanning three imaging modalities. To the best of our\nknowledge, this is the most comprehensive brain encoding model to the date. We\ndemonstrate the effectiveness of the pre-trained model as a drop-in replacement\nfor commonly used vision backbone models. Furthermore, we demonstrate the\napplication of the model to brain decoding. Code and the model checkpoint will\nbe made available.\n", "date": "2023-07-26", "categories": []}
{"id": "2307.00932", "title": "A large calcium-imaging dataset reveals a systematic V4 organization for\n  natural scenes", "abstract": "  The visual system evolved to process natural scenes, yet most of our\nunderstanding of the topology and function of visual cortex derives from\nstudies using artificial stimuli. To gain deeper insights into visual\nprocessing of natural scenes, we utilized widefield calcium-imaging of primate\nV4 in response to many natural images, generating a large dataset of\ncolumnar-scale responses. We used this dataset to build a digital twin of V4\nvia deep learning, generating a detailed topographical map of natural image\npreferences at each cortical position. The map revealed clustered functional\ndomains for specific classes of natural image features. These ranged from\nsurface-related attributes like color and texture to shape-related features\nsuch as edges, curvature, and facial features. We validated the model-predicted\ndomains with additional widefield calcium-imaging and single-cell resolution\ntwo-photon imaging. Our study illuminates the detailed topological organization\nand neural codes in V4 that represent natural scenes.\n", "date": "2023-07-03", "categories": []}
{"id": "2307.10177", "title": "Bayesian Spike Train Inference via Non-Local Priors", "abstract": "  Advances in neuroscience have enabled researchers to measure the activities\nof large numbers of neurons simultaneously in behaving animals. We have access\nto the fluorescence of each of the neurons which provides a first-order\napproximation of the neural activity over time. Determining the exact spike of\na neuron from this fluorescence trace constitutes an active area of research\nwithin the field of computational neuroscience. We propose a novel Bayesian\napproach based on a mixture of half-non-local prior densities and point masses\nfor this task. Instead of a computationally expensive MCMC algorithm, we adopt\na stochastic search-based approach that is capable of taking advantage of\nmodern computing environments often equipped with multiple processors, to\nexplore all possible arrangements of spikes and lack thereof in an observed\nspike train. It then reports the highest posterior probability arrangement of\nspikes and posterior probability for a spike at each location of the spike\ntrain. Our proposals lead to substantial improvements over existing proposals\nbased on L1 regularization, and enjoy comparable estimation accuracy to the\nstate-of-the-art L0 proposal, in simulations, and on recent calcium imaging\ndata sets. Notably, contrary to optimization-based frequentist approaches, our\nmethodology yields automatic uncertainty quantification associated with the\nspike-train inference.\n", "date": "2023-05-27", "categories": []}
{"id": "2307.09770", "title": "Perturbing a Neural Network to Infer Effective Connectivity: Evidence\n  from Synthetic EEG Data", "abstract": "  Identifying causal relationships among distinct brain areas, known as\neffective connectivity, holds key insights into the brain's information\nprocessing and cognitive functions. Electroencephalogram (EEG) signals exhibit\nintricate dynamics and inter-areal interactions within the brain. However,\nmethods for characterizing nonlinear causal interactions among multiple brain\nregions remain relatively underdeveloped. In this study, we proposed a\ndata-driven framework to infer effective connectivity by perturbing the trained\nneural networks. Specifically, we trained neural networks (i.e., CNN, vanilla\nRNN, GRU, LSTM, and Transformer) to predict future EEG signals according to\nhistorical data and perturbed the networks' input to obtain effective\nconnectivity (EC) between the perturbed EEG channel and the rest of the\nchannels. The EC reflects the causal impact of perturbing one node on others.\nThe performance was tested on the synthetic EEG generated by a\nbiological-plausible Jansen-Rit model. CNN and Transformer obtained the best\nperformance on both 3-channel and 90-channel synthetic EEG data, outperforming\nthe classical Granger causality method. Our work demonstrated the potential of\nperturbing an artificial neural network, learned to predict future system\ndynamics, to uncover the underlying causal structure.\n", "date": "2023-07-19", "categories": []}
{"id": "2301.03198", "title": "The Algonauts Project 2023 Challenge: How the Human Brain Makes Sense of\n  Natural Scenes", "abstract": "  The sciences of biological and artificial intelligence are ever more\nintertwined. Neural computational principles inspire new intelligent machines,\nwhich are in turn used to advance theoretical understanding of the brain. To\npromote further exchange of ideas and collaboration between biological and\nartificial intelligence researchers, we introduce the 2023 installment of the\nAlgonauts Project challenge: How the Human Brain Makes Sense of Natural Scenes\n(http://algonauts.csail.mit.edu). This installment prompts the fields of\nartificial and biological intelligence to come together towards building\ncomputational models of the visual brain using the largest and richest dataset\nof fMRI responses to visual scenes, the Natural Scenes Dataset (NSD). NSD\nprovides high-quality fMRI responses to ~73,000 different naturalistic colored\nscenes, making it the ideal candidate for data-driven model building approaches\npromoted by the 2023 challenge. The challenge is open to all and makes results\ndirectly comparable and transparent through a public leaderboard automatically\nupdated after each submission, thus allowing for rapid model development. We\nbelieve that the 2023 installment will spark symbiotic collaborations between\nbiological and artificial intelligence scientists, leading to a deeper\nunderstanding of the brain through cutting-edge computational models and to\nnovel ways of engineering artificial intelligent agents through inductive\nbiases from biological systems.\n", "date": "2023-01-09", "categories": []}
{"id": "2307.05355", "title": "UniCoRN: Unified Cognitive Signal ReconstructioN bridging cognitive\n  signals and human language", "abstract": "  Decoding text stimuli from cognitive signals (e.g. fMRI) enhances our\nunderstanding of the human language system, paving the way for building\nversatile Brain-Computer Interface. However, existing studies largely focus on\ndecoding individual word-level fMRI volumes from a restricted vocabulary, which\nis far too idealized for real-world application. In this paper, we propose\nfMRI2text, the first openvocabulary task aiming to bridge fMRI time series and\nhuman language. Furthermore, to explore the potential of this new task, we\npresent a baseline solution, UniCoRN: the Unified Cognitive Signal\nReconstructioN for Brain Decoding. By reconstructing both individual time\npoints and time series, UniCoRN establishes a robust encoder for cognitive\nsignals (fMRI & EEG). Leveraging a pre-trained language model as decoder,\nUniCoRN proves its efficacy in decoding coherent text from fMRI series across\nvarious split settings. Our model achieves a 34.77% BLEU score on fMRI2text,\nand a 37.04% BLEU when generalized to EEGto-text decoding, thereby surpassing\nthe former baseline. Experimental results indicate the feasibility of decoding\nconsecutive fMRI volumes, and the effectiveness of decoding different cognitive\nsignals using a unified structure.\n", "date": "2023-07-06", "categories": []}
{"id": "2307.02398", "title": "A Versatile Hub Model For Efficient Information Propagation And Feature\n  Selection", "abstract": "  Hub structure, characterized by a few highly interconnected nodes surrounded\nby a larger number of nodes with fewer connections, is a prominent topological\nfeature of biological brains, contributing to efficient information transfer\nand cognitive processing across various species. In this paper, a mathematical\nmodel of hub structure is presented. The proposed method is versatile and can\nbe broadly applied to both computational neuroscience and Recurrent Neural\nNetworks (RNNs) research. We employ the Echo State Network (ESN) as a means to\ninvestigate the mechanistic underpinnings of hub structures. Our findings\ndemonstrate a substantial enhancement in performance upon incorporating the hub\nstructure. Through comprehensive mechanistic analyses, we show that the hub\nstructure improves model performance by facilitating efficient information\nprocessing and better feature extractions.\n", "date": "2023-07-05", "categories": []}
{"id": "2212.03771", "title": "Expressive architectures enhance interpretability of dynamics-based\n  neural population models", "abstract": "  Artificial neural networks that can recover latent dynamics from recorded\nneural activity may provide a powerful avenue for identifying and interpreting\nthe dynamical motifs underlying biological computation. Given that neural\nvariance alone does not uniquely determine a latent dynamical system,\ninterpretable architectures should prioritize accurate and low-dimensional\nlatent dynamics. In this work, we evaluated the performance of sequential\nautoencoders (SAEs) in recovering latent chaotic attractors from simulated\nneural datasets. We found that SAEs with widely-used recurrent neural network\n(RNN)-based dynamics were unable to infer accurate firing rates at the true\nlatent state dimensionality, and that larger RNNs relied upon dynamical\nfeatures not present in the data. On the other hand, SAEs with neural ordinary\ndifferential equation (NODE)-based dynamics inferred accurate rates at the true\nlatent state dimensionality, while also recovering latent trajectories and\nfixed point structure. Ablations reveal that this is mainly because NODEs (1)\nallow use of higher-capacity multi-layer perceptrons (MLPs) to model the vector\nfield and (2) predict the derivative rather than the next state. Decoupling the\ncapacity of the dynamics model from its latent dimensionality enables NODEs to\nlearn the requisite low-D dynamics where RNN cells fail. Additionally, the fact\nthat the NODE predicts derivatives imposes a useful autoregressive prior on the\nlatent states. The suboptimal interpretability of widely-used RNN-based\ndynamics may motivate substitution for alternative architectures, such as NODE,\nthat enable learning of accurate dynamics in low-dimensional latent spaces.\n", "date": "2022-12-07", "categories": []}
{"id": "2211.05634", "title": "Generalization of generative model for neuronal ensemble inference\n  method", "abstract": "  Various brain functions that are necessary to maintain life activities\nmaterialize through the interaction of countless neurons. Therefore, it is\nimportant to analyze functional neuronal network. To elucidate the mechanism of\nbrain function, many studies are being actively conducted on functional\nneuronal ensemble and hub, including all areas of neuroscience. In addition,\nrecent study suggests that the existence of functional neuronal ensembles and\nhubs contributes to the efficiency of information processing. For these\nreasons, there is a demand for methods to infer functional neuronal ensembles\nfrom neuronal activity data, and methods based on Bayesian inference have been\nproposed. However, there is a problem in modeling the activity in Bayesian\ninference. The features of each neuron's activity have non-stationarity\ndepending on physiological experimental conditions. As a result, the assumption\nof stationarity in Bayesian inference model impedes inference, which leads to\ndestabilization of inference results and degradation of inference accuracy. In\nthis study, we extend the range of the variable for expressing the neuronal\nstate, and generalize the likelihood of the model for extended variables. By\ncomparing with the previous study, our model can express the neuronal state in\nlarger space. This generalization without restriction of the binary input\nenables us to perform soft clustering and apply the method to non-stationary\nneuroactivity data. In addition, for the effectiveness of the method, we apply\nthe developed method to multiple synthetic fluorescence data generated from the\nelectrical potential data in leaky integrated-and-fire model.\n", "date": "2022-11-07", "categories": []}
{"id": "2306.15364", "title": "The Architecture of a Biologically Plausible Language Organ", "abstract": "  We present a simulated biologically plausible language organ, made up of\nstylized but realistic neurons, synapses, brain areas, plasticity, and a\nsimplified model of sensory perception. We show through experiments that this\nmodel succeeds in an important early step in language acquisition: the learning\nof nouns, verbs, and their meanings, from the grounded input of only a modest\nnumber of sentences. Learning in this system is achieved through Hebbian\nplasticity, and without backpropagation. Our model goes beyond a parser\npreviously designed in a similar environment, with the critical addition of a\nbiologically plausible account for how language can be acquired in the infant's\nbrain, not just processed by a mature brain.\n", "date": "2023-06-27", "categories": []}
{"id": "2306.13102", "title": "MBrain: A Multi-channel Self-Supervised Learning Framework for Brain\n  Signals", "abstract": "  Brain signals are important quantitative data for understanding physiological\nactivities and diseases of human brain. Most existing studies pay attention to\nsupervised learning methods, which, however, require high-cost clinical labels.\nIn addition, the huge difference in the clinical patterns of brain signals\nmeasured by invasive (e.g., SEEG) and non-invasive (e.g., EEG) methods leads to\nthe lack of a unified method. To handle the above issues, we propose to study\nthe self-supervised learning (SSL) framework for brain signals that can be\napplied to pre-train either SEEG or EEG data. Intuitively, brain signals,\ngenerated by the firing of neurons, are transmitted among different connecting\nstructures in human brain. Inspired by this, we propose MBrain to learn\nimplicit spatial and temporal correlations between different channels (i.e.,\ncontacts of the electrode, corresponding to different brain areas) as the\ncornerstone for uniformly modeling different types of brain signals.\nSpecifically, we represent the spatial correlation by a graph structure, which\nis built with proposed multi-channel CPC. We theoretically prove that\noptimizing the goal of multi-channel CPC can lead to a better predictive\nrepresentation and apply the instantaneou-time-shift prediction task based on\nit. Then we capture the temporal correlation by designing the\ndelayed-time-shift prediction task. Finally, replace-discriminative-learning\ntask is proposed to preserve the characteristics of each channel. Extensive\nexperiments of seizure detection on both EEG and SEEG large-scale real-world\ndatasets demonstrate that our model outperforms several state-of-the-art time\nseries SSL and unsupervised models, and has the ability to be deployed to\nclinical practice.\n", "date": "2023-06-15", "categories": []}
{"id": "2306.05053", "title": "Active Inference in Hebbian Learning Networks", "abstract": "  This work studies how brain-inspired neural ensembles equipped with local\nHebbian plasticity can perform active inference (AIF) in order to control\ndynamical agents. A generative model capturing the environment dynamics is\nlearned by a network composed of two distinct Hebbian ensembles: a posterior\nnetwork, which infers latent states given the observations, and a state\ntransition network, which predicts the next expected latent state given current\nstate-action pairs. Experimental studies are conducted using the Mountain Car\nenvironment from the OpenAI gym suite, to study the effect of the various\nHebbian network parameters on the task performance. It is shown that the\nproposed Hebbian AIF approach outperforms the use of Q-learning, while not\nrequiring any replay buffer, as in typical reinforcement learning systems.\nThese results motivate further investigations of Hebbian learning for the\ndesign of AIF networks that can learn environment dynamics without the need for\nrevisiting past buffered experiences.\n", "date": "2023-06-08", "categories": []}
{"id": "2302.06403", "title": "Sources of Richness and Ineffability for Phenomenally Conscious States", "abstract": "  Conscious states (states that there is something it is like to be in) seem\nboth rich or full of detail, and ineffable or hard to fully describe or recall.\nThe problem of ineffability, in particular, is a longstanding issue in\nphilosophy that partly motivates the explanatory gap: the belief that\nconsciousness cannot be reduced to underlying physical processes. Here, we\nprovide an information theoretic dynamical systems perspective on the richness\nand ineffability of consciousness. In our framework, the richness of conscious\nexperience corresponds to the amount of information in a conscious state and\nineffability corresponds to the amount of information lost at different stages\nof processing. We describe how attractor dynamics in working memory would\ninduce impoverished recollections of our original experiences, how the discrete\nsymbolic nature of language is insufficient for describing the rich and\nhigh-dimensional structure of experiences, and how similarity in the cognitive\nfunction of two individuals relates to improved communicability of their\nexperiences to each other. While our model may not settle all questions\nrelating to the explanatory gap, it makes progress toward a fully physicalist\nexplanation of the richness and ineffability of conscious experience: two\nimportant aspects that seem to be part of what makes qualitative character so\npuzzling.\n", "date": "2023-02-13", "categories": []}
{"id": "2306.05126", "title": "Mapping Brains with Language Models: A Survey", "abstract": "  Over the years, many researchers have seemingly made the same observation:\nBrain and language model activations exhibit some structural similarities,\nenabling linear partial mappings between features extracted from neural\nrecordings and computational language models. In an attempt to evaluate how\nmuch evidence has been accumulated for this observation, we survey over 30\nstudies spanning 10 datasets and 8 metrics. How much evidence has been\naccumulated, and what, if anything, is missing before we can draw conclusions?\nOur analysis of the evaluation methods used in the literature reveals that some\nof the metrics are less conservative. We also find that the accumulated\nevidence, for now, remains ambiguous, but correlations with model size and\nquality provide grounds for cautious optimism.\n", "date": "2023-06-08", "categories": []}
{"id": "2306.05286", "title": "JGAT: a joint spatio-temporal graph attention model for brain decoding", "abstract": "  The decoding of brain neural networks has been an intriguing topic in\nneuroscience for a well-rounded understanding of different types of brain\ndisorders and cognitive stimuli. Integrating different types of connectivity,\ne.g., Functional Connectivity (FC) and Structural Connectivity (SC), from\nmulti-modal imaging techniques can take their complementary information into\naccount and therefore have the potential to get better decoding capability.\nHowever, traditional approaches for integrating FC and SC overlook the\ndynamical variations, which stand a great chance to over-generalize the brain\nneural network. In this paper, we propose a Joint kernel Graph Attention\nNetwork (JGAT), which is a new multi-modal temporal graph attention network\nframework. It integrates the data from functional Magnetic Resonance Images\n(fMRI) and Diffusion Weighted Imaging (DWI) while preserving the dynamic\ninformation at the same time. We conduct brain-decoding tasks with our JGAT on\nfour independent datasets: three of 7T fMRI datasets from the Human Connectome\nProject (HCP) and one from animal neural recordings. Furthermore, with\nAttention Scores (AS) and Frame Scores (FS) computed and learned from the\nmodel, we can locate several informative temporal segments and build meaningful\ndynamical pathways along the temporal domain for the HCP datasets. The URL to\nthe code of JGAT model: https://github.com/BRAINML-GT/JGAT.\n", "date": "2023-06-03", "categories": []}
{"id": "2302.02563", "title": "Stochastic Gradient Descent-Induced Drift of Representation in a\n  Two-Layer Neural Network", "abstract": "  Representational drift refers to over-time changes in neural activation\naccompanied by a stable task performance. Despite being observed in the brain\nand in artificial networks, the mechanisms of drift and its implications are\nnot fully understood. Motivated by recent experimental findings of\nstimulus-dependent drift in the piriform cortex, we use theory and simulations\nto study this phenomenon in a two-layer linear feedforward network.\nSpecifically, in a continual online learning scenario, we study the drift\ninduced by the noise inherent in the Stochastic Gradient Descent (SGD). By\ndecomposing the learning dynamics into the normal and tangent spaces of the\nminimum-loss manifold, we show the former corresponds to a finite variance\nfluctuation, while the latter could be considered as an effective diffusion\nprocess on the manifold. We analytically compute the fluctuation and the\ndiffusion coefficients for the stimuli representations in the hidden layer as\nfunctions of network parameters and input distribution. Further, consistent\nwith experiments, we show that the drift rate is slower for a more frequently\npresented stimulus. Overall, our analysis yields a theoretical framework for\nbetter understanding of the drift phenomenon in biological and artificial\nneural networks.\n", "date": "2023-02-06", "categories": []}
{"id": "2306.03375", "title": "Identifying Shared Decodable Concepts in the Human Brain Using\n  Image-Language Foundation Models", "abstract": "  We introduce a method that takes advantage of high-quality pretrained\nmultimodal representations to explore fine-grained semantic networks in the\nhuman brain. Previous studies have documented evidence of functional\nlocalization in the brain, with different anatomical regions preferentially\nactivating for different types of sensory input. Many such localized structures\nare known, including the fusiform face area and parahippocampal place area.\nThis raises the question of whether additional brain regions (or conjunctions\nof brain regions) are also specialized for other important semantic concepts.\nTo identify such brain regions, we developed a data-driven approach to uncover\nvisual concepts that are decodable from a massive functional magnetic resonance\nimaging (fMRI) dataset. Our analysis is broadly split into three sections.\nFirst, a fully connected neural network is trained to map brain responses to\nthe outputs of an image-language foundation model, CLIP (Radford et al., 2021).\nSubsequently, a contrastive-learning dimensionality reduction method reveals\nthe brain-decodable components of CLIP space. In the final section of our\nanalysis, we localize shared decodable concepts in the brain using a\nvoxel-masking optimization method to produce a shared decodable concept (SDC)\nspace. The accuracy of our procedure is validated by comparing it to previous\nlocalization experiments that identify regions for faces, bodies, and places.\nIn addition to these concepts, whose corresponding brain regions were already\nknown, we localize novel concept representations which are shared across\nparticipants to other areas of the human brain. We also demonstrate how this\nmethod can be used to inspect fine-grained semantic networks for individual\nparticipants. We envisage that this extensible method can also be adapted to\nexplore other questions at the intersection of AI and neuroscience.\n", "date": "2023-06-06", "categories": []}
{"id": "2306.03779", "title": "Performance-optimized deep neural networks are evolving into worse\n  models of inferotemporal visual cortex", "abstract": "  One of the most impactful findings in computational neuroscience over the\npast decade is that the object recognition accuracy of deep neural networks\n(DNNs) correlates with their ability to predict neural responses to natural\nimages in the inferotemporal (IT) cortex. This discovery supported the\nlong-held theory that object recognition is a core objective of the visual\ncortex, and suggested that more accurate DNNs would serve as better models of\nIT neuron responses to images. Since then, deep learning has undergone a\nrevolution of scale: billion parameter-scale DNNs trained on billions of images\nare rivaling or outperforming humans at visual tasks including object\nrecognition. Have today's DNNs become more accurate at predicting IT neuron\nresponses to images as they have grown more accurate at object recognition?\n  Surprisingly, across three independent experiments, we find this is not the\ncase. DNNs have become progressively worse models of IT as their accuracy has\nincreased on ImageNet. To understand why DNNs experience this trade-off and\nevaluate if they are still an appropriate paradigm for modeling the visual\nsystem, we turn to recordings of IT that capture spatially resolved maps of\nneuronal activity elicited by natural images. These neuronal activity maps\nreveal that DNNs trained on ImageNet learn to rely on different visual features\nthan those encoded by IT and that this problem worsens as their accuracy\nincreases. We successfully resolved this issue with the neural harmonizer, a\nplug-and-play training routine for DNNs that aligns their learned\nrepresentations with humans. Our results suggest that harmonized DNNs break the\ntrade-off between ImageNet accuracy and neural prediction accuracy that assails\ncurrent DNNs and offer a path to more accurate models of biological vision.\n", "date": "2023-06-06", "categories": []}
{"id": "2306.01802", "title": "Linear Time GPs for Inferring Latent Trajectories from Neural Spike\n  Trains", "abstract": "  Latent Gaussian process (GP) models are widely used in neuroscience to\nuncover hidden state evolutions from sequential observations, mainly in neural\nactivity recordings. While latent GP models provide a principled and powerful\nsolution in theory, the intractable posterior in non-conjugate settings\nnecessitates approximate inference schemes, which may lack scalability. In this\nwork, we propose cvHM, a general inference framework for latent GP models\nleveraging Hida-Mat\\'ern kernels and conjugate computation variational\ninference (CVI). With cvHM, we are able to perform variational inference of\nlatent neural trajectories with linear time complexity for arbitrary\nlikelihoods. The reparameterization of stationary kernels using Hida-Mat\\'ern\nGPs helps us connect the latent variable models that encode prior assumptions\nthrough dynamical systems to those that encode trajectory assumptions through\nGPs. In contrast to previous work, we use bidirectional information filtering,\nleading to a more concise implementation. Furthermore, we employ the Whittle\napproximate likelihood to achieve highly efficient hyperparameter learning.\n", "date": "2023-06-01", "categories": []}
{"id": "2306.00528", "title": "Neuronal Cell Type Classification using Deep Learning", "abstract": "  The brain is likely the most complex organ, given the variety of functions it\ncontrols, the number of cells it comprises, and their corresponding diversity.\nStudying and identifying neurons, the brain's primary building blocks, is a\ncrucial milestone and essential for understanding brain function in health and\ndisease. Recent developments in machine learning have provided advanced\nabilities for classifying neurons. However, these methods remain black boxes\nwith no explainability and reasoning. This paper aims to provide a robust and\nexplainable deep-learning framework to classify neurons based on their\nelectrophysiological activity. Our analysis is performed on data provided by\nthe Allen Cell Types database containing a survey of biological features\nderived from single-cell recordings of mice and humans. First, we classify\nneuronal cell types of mice data to identify excitatory and inhibitory neurons.\nThen, neurons are categorized to their broad types in humans using domain\nadaptation from mice data. Lastly, neurons are classified into sub-types based\non transgenic mouse lines using deep neural networks in an explainable fashion.\nWe show state-of-the-art results in a dendrite-type classification of\nexcitatory vs. inhibitory neurons and transgenic mouse lines classification.\nThe model is also inherently interpretable, revealing the correlations between\nneuronal types and their electrophysiological properties.\n", "date": "2023-06-01", "categories": []}
{"id": "2306.00927", "title": "Second Sight: Using brain-optimized encoding models to align image\n  distributions with human brain activity", "abstract": "  Two recent developments have accelerated progress in image reconstruction\nfrom human brain activity: large datasets that offer samples of brain activity\nin response to many thousands of natural scenes, and the open-sourcing of\npowerful stochastic image-generators that accept both low- and high-level\nguidance. Most work in this space has focused on obtaining point estimates of\nthe target image, with the ultimate goal of approximating literal pixel-wise\nreconstructions of target images from the brain activity patterns they evoke.\nThis emphasis belies the fact that there is always a family of images that are\nequally compatible with any evoked brain activity pattern, and the fact that\nmany image-generators are inherently stochastic and do not by themselves offer\na method for selecting the single best reconstruction from among the samples\nthey generate. We introduce a novel reconstruction procedure (Second Sight)\nthat iteratively refines an image distribution to explicitly maximize the\nalignment between the predictions of a voxel-wise encoding model and the brain\nactivity patterns evoked by any target image. We show that our process\nconverges on a distribution of high-quality reconstructions by refining both\nsemantic content and low-level image details across iterations. Images sampled\nfrom these converged image distributions are competitive with state-of-the-art\nreconstruction algorithms. Interestingly, the time-to-convergence varies\nsystematically across visual cortex, with earlier visual areas generally taking\nlonger and converging on narrower image distributions, relative to higher-level\nbrain areas. Second Sight thus offers a succinct and novel method for exploring\nthe diversity of representations across visual brain areas.\n", "date": "2023-06-01", "categories": []}
{"id": "2305.17023", "title": "Are Deep Neural Networks Adequate Behavioural Models of Human Visual\n  Perception?", "abstract": "  Deep neural networks (DNNs) are machine learning algorithms that have\nrevolutionised computer vision due to their remarkable successes in tasks like\nobject classification and segmentation. The success of DNNs as computer vision\nalgorithms has led to the suggestion that DNNs may also be good models of human\nvisual perception. We here review evidence regarding current DNNs as adequate\nbehavioural models of human core object recognition. To this end, we argue that\nit is important to distinguish between statistical tools and computational\nmodels, and to understand model quality as a multidimensional concept where\nclarity about modelling goals is key. Reviewing a large number of\npsychophysical and computational explorations of core object recognition\nperformance in humans and DNNs, we argue that DNNs are highly valuable\nscientific tools but that as of today DNNs should only be regarded as promising\n-- but not yet adequate -- computational models of human core object\nrecognition behaviour. On the way we dispel a number of myths surrounding DNNs\nin vision science.\n", "date": "2023-05-26", "categories": []}
{"id": "2305.14369", "title": "Learning low-dimensional dynamics from whole-brain data improves task\n  capture", "abstract": "  The neural dynamics underlying brain activity are critical to understanding\ncognitive processes and mental disorders. However, current voxel-based\nwhole-brain dimensionality reduction techniques fall short of capturing these\ndynamics, producing latent timeseries that inadequately relate to behavioral\ntasks. To address this issue, we introduce a novel approach to learning\nlow-dimensional approximations of neural dynamics by using a sequential\nvariational autoencoder (SVAE) that represents the latent dynamical system via\na neural ordinary differential equation (NODE). Importantly, our method finds\nsmooth dynamics that can predict cognitive processes with accuracy higher than\nclassical methods. Our method also shows improved spatial localization to\ntask-relevant brain regions and identifies well-known structures such as the\nmotor homunculus from fMRI motor task recordings. We also find that non-linear\nprojections to the latent space enhance performance for specific tasks,\noffering a promising direction for future research. We evaluate our approach on\nvarious task-fMRI datasets, including motor, working memory, and relational\nprocessing tasks, and demonstrate that it outperforms widely used\ndimensionality reduction techniques in how well the latent timeseries relates\nto behavioral sub-tasks, such as left-hand or right-hand tapping. Additionally,\nwe replace the NODE with a recurrent neural network (RNN) and compare the two\napproaches to understand the importance of explicitly learning a dynamical\nsystem. Lastly, we analyze the robustness of the learned dynamical systems\nthemselves and find that their fixed points are robust across seeds,\nhighlighting our method's potential for the analysis of cognitive processes as\ndynamical systems.\n", "date": "2023-05-18", "categories": []}
{"id": "2305.11278", "title": "Real-Time Variational Method for Learning Neural Trajectory and its\n  Dynamics", "abstract": "  Latent variable models have become instrumental in computational neuroscience\nfor reasoning about neural computation. This has fostered the development of\npowerful offline algorithms for extracting latent neural trajectories from\nneural recordings. However, despite the potential of real time alternatives to\ngive immediate feedback to experimentalists, and enhance experimental design,\nthey have received markedly less attention. In this work, we introduce the\nexponential family variational Kalman filter (eVKF), an online recursive\nBayesian method aimed at inferring latent trajectories while simultaneously\nlearning the dynamical system generating them. eVKF works for arbitrary\nlikelihoods and utilizes the constant base measure exponential family to model\nthe latent state stochasticity. We derive a closed-form variational analogue to\nthe predict step of the Kalman filter which leads to a provably tighter bound\non the ELBO compared to another online variational method. We validate our\nmethod on synthetic and real-world data, and, notably, show that it achieves\ncompetitive performance\n", "date": "2023-05-18", "categories": []}
{"id": "2305.11107", "title": "From Data-Fitting to Discovery: Interpreting the Neural Dynamics of\n  Motor Control through Reinforcement Learning", "abstract": "  In motor neuroscience, artificial recurrent neural networks models often\ncomplement animal studies. However, most modeling efforts are limited to\ndata-fitting, and the few that examine virtual embodied agents in a\nreinforcement learning context, do not draw direct comparisons to their\nbiological counterparts. Our study addressing this gap, by uncovering\nstructured neural activity of a virtual robot performing legged locomotion that\ndirectly support experimental findings of primate walking and cycling. We find\nthat embodied agents trained to walk exhibit smooth dynamics that avoid\ntangling -- or opposing neural trajectories in neighboring neural space -- a\ncore principle in computational neuroscience. Specifically, across a wide suite\nof gaits, the agent displays neural trajectories in the recurrent layers are\nless tangled than those in the input-driven actuation layers. To better\ninterpret the neural separation of these elliptical-shaped trajectories, we\nidentify speed axes that maximizes variance of mean activity across different\nforward, lateral, and rotational speed conditions.\n", "date": "2023-05-18", "categories": []}
{"id": "2305.11137", "title": "Parallel development of social preferences in fish and machines", "abstract": "  What are the computational foundations of social grouping? Traditional\napproaches to this question have focused on verbal reasoning or simple\n(low-dimensional) quantitative models. In the real world, however, social\npreferences emerge when high-dimensional learning systems (brains and bodies)\ninteract with high-dimensional sensory inputs during an animal's embodied\ninteractions with the world. A deep understanding of social grouping will\ntherefore require embodied models that learn directly from sensory inputs using\nhigh-dimensional learning mechanisms. To this end, we built artificial neural\nnetworks (ANNs), embodied those ANNs in virtual fish bodies, and raised the\nartificial fish in virtual fish tanks that mimicked the rearing conditions of\nreal fish. We then compared the social preferences that emerged in real fish\nversus artificial fish. We found that when artificial fish had two core\nlearning mechanisms (reinforcement learning and curiosity-driven learning),\nartificial fish developed fish-like social preferences. Like real fish, the\nartificial fish spontaneously learned to prefer members of their own group over\nmembers of other groups. The artificial fish also spontaneously learned to\nself-segregate with their in-group, akin to self-segregation behavior seen in\nnature. Our results suggest that social grouping can emerge from three\ningredients: (1) reinforcement learning, (2) intrinsic motivation, and (3)\nearly social experiences with in-group members. This approach lays a foundation\nfor reverse engineering animal-like social behavior with image-computable\nmodels, bridging the divide between high-dimensional sensory inputs and social\npreferences.\n", "date": "2023-05-18", "categories": []}
{"id": "2305.08124", "title": "Theta sequences as eligibility traces: a biological solution to credit\n  assignment", "abstract": "  Credit assignment problems, for example policy evaluation in RL, often\nrequire bootstrapping prediction errors through preceding states \\textit{or}\nmaintaining temporally extended memory traces; solutions which are unfavourable\nor implausible for biological networks of neurons. We propose theta sequences\n-- chains of neural activity during theta oscillations in the hippocampus,\nthought to represent rapid playthroughs of awake behaviour -- as a solution. By\nanalysing and simulating a model for theta sequences we show they compress\nbehaviour such that existing but short $\\mathsf{O}(10)$ ms neuronal memory\ntraces are effectively extended allowing for bootstrap-free credit assignment\nwithout long memory traces, equivalent to the use of eligibility traces in\nTD($\\lambda$).\n", "date": "2023-05-14", "categories": []}
{"id": "2305.03866", "title": "Spiking neural networks with Hebbian plasticity for unsupervised\n  representation learning", "abstract": "  We introduce a novel spiking neural network model for learning distributed\ninternal representations from data in an unsupervised procedure. We achieved\nthis by transforming the non-spiking feedforward Bayesian Confidence\nPropagation Neural Network (BCPNN) model, employing an online correlation-based\nHebbian-Bayesian learning and rewiring mechanism, shown previously to perform\nrepresentation learning, into a spiking neural network with Poisson statistics\nand low firing rate comparable to in vivo cortical pyramidal neurons. We\nevaluated the representations learned by our spiking model using a linear\nclassifier and show performance close to the non-spiking BCPNN, and competitive\nwith other Hebbian-based spiking networks when trained on MNIST and F-MNIST\nmachine learning benchmarks.\n", "date": "2023-05-05", "categories": []}
{"id": "2305.00556", "title": "Reconstructing seen images from human brain activity via guided\n  stochastic search", "abstract": "  Visual reconstruction algorithms are an interpretive tool that map brain\nactivity to pixels. Past reconstruction algorithms employed brute-force search\nthrough a massive library to select candidate images that, when passed through\nan encoding model, accurately predict brain activity. Here, we use conditional\ngenerative diffusion models to extend and improve this search-based strategy.\nWe decode a semantic descriptor from human brain activity (7T fMRI) in voxels\nacross most of visual cortex, then use a diffusion model to sample a small\nlibrary of images conditioned on this descriptor. We pass each sample through\nan encoding model, select the images that best predict brain activity, and then\nuse these images to seed another library. We show that this process converges\non high-quality reconstructions by refining low-level image details while\npreserving semantic content across iterations. Interestingly, the\ntime-to-convergence differs systematically across visual cortex, suggesting a\nsuccinct new way to measure the diversity of representations across visual\nbrain areas.\n", "date": "2023-04-30", "categories": []}
{"id": "2303.12259", "title": "Brain-inspired bodily self-perception model for robot rubber hand\n  illusion", "abstract": "  At the core of bodily self-consciousness is the perception of the ownership\nof one's body. Recent efforts to gain a deeper understanding of the mechanisms\nbehind the brain's encoding of the self-body have led to various attempts to\ndevelop a unified theoretical framework to explain related behavioral and\nneurophysiological phenomena. A central question to be explained is how body\nillusions such as the rubber hand illusion actually occur. Despite the\nconceptual descriptions of the mechanisms of bodily self-consciousness and the\npossible relevant brain areas, the existing theoretical models still lack an\nexplanation of the computational mechanisms by which the brain encodes the\nperception of one's body and how our subjectively perceived body illusions can\nbe generated by neural networks. Here we integrate the biological findings of\nbodily self-consciousness to propose a Brain-inspired bodily self-perception\nmodel, by which perceptions of bodily self can be autonomously constructed\nwithout any supervision signals. We successfully validated our computational\nmodel with six rubber hand illusion experiments and a disability experiment on\nplatforms including a iCub humanoid robot and simulated environments. The\nexperimental results show that our model can not only well replicate the\nbehavioral and neural data of monkeys in biological experiments, but also\nreasonably explain the causes and results of the rubber hand illusion from the\nneuronal level due to advantages in biological interpretability, thus\ncontributing to the revealing of the computational and neural mechanisms\nunderlying the occurrence of the rubber hand illusion.\n", "date": "2023-03-22", "categories": []}
{"id": "2210.16147", "title": "Modeling structure-building in the brain with CCG parsing and large\n  language models", "abstract": "  To model behavioral and neural correlates of language comprehension in\nnaturalistic environments researchers have turned to broad-coverage tools from\nnatural-language processing and machine learning. Where syntactic structure is\nexplicitly modeled, prior work has relied predominantly on context-free\ngrammars (CFG), yet such formalisms are not sufficiently expressive for human\nlanguages. Combinatory Categorial Grammars (CCGs) are sufficiently expressive\ndirectly compositional models of grammar with flexible constituency that\naffords incremental interpretation. In this work we evaluate whether a more\nexpressive CCG provides a better model than a CFG for human neural signals\ncollected with fMRI while participants listen to an audiobook story. We further\ntest between variants of CCG that differ in how they handle optional adjuncts.\nThese evaluations are carried out against a baseline that includes estimates of\nnext-word predictability from a Transformer neural network language model. Such\na comparison reveals unique contributions of CCG structure-building\npredominantly in the left posterior temporal lobe: CCG-derived measures offer a\nsuperior fit to neural signals compared to those derived from a CFG. These\neffects are spatially distinct from bilateral superior temporal effects that\nare unique to predictability. Neural effects for structure-building are thus\nseparable from predictability during naturalistic listening, and those effects\nare best characterized by a grammar whose expressive power is motivated on\nindependent linguistic grounds.\n", "date": "2022-10-28", "categories": []}
{"id": "2304.01346", "title": "Towards an Hybrid Hodgkin-Huxley Action Potential Generation Model", "abstract": "  Mathematical models for the generation of the action potential can improve\nthe understanding of physiological mechanisms that are consequence of the\nelectrical activity in neurons. In such models, some equations involving\nempirically obtained functions of the membrane potential are usually defined.\nThe best known of these models, the Hodgkin-Huxley model, is an example of this\nparadigm since it defines the conductances of ion channels in terms of the\nopening and closing rates of each type of gate present in the channels. These\nfunctions need to be derived from laboratory measurements that are often very\nexpensive and produce little data because they involve a time-space-independent\nmeasurement of the voltage in a single channel of the cell membrane. In this\nwork, we investigate the possibility of finding the Hodgkin-Huxley model's\nparametric functions using only two simple measurements (the membrane voltage\nas a function of time and the injected current that triggered that voltage) and\napplying Deep Learning methods to estimate these functions. This would result\nin an hybrid model of the action potential generation composed by the original\nHodgkin-Huxley equations and an Artificial Neural Network that requires a small\nset of easy-to-perform measurements to be trained. Experiments were carried out\nusing data generated from the original Hodgkin-Huxley model, and results show\nthat a simple two-layer artificial neural network (ANN) architecture trained on\na minimal amount of data can learn to model some of the fundamental proprieties\nof the action potential generation by estimating the model's rate functions.\n", "date": "2023-03-15", "categories": []}
{"id": "2303.08109", "title": "Vision-based route following by an embodied insect-inspired sparse\n  neural network", "abstract": "  We compared the efficiency of the FlyHash model, an insect-inspired sparse\nneural network (Dasgupta et al., 2017), to similar but non-sparse models in an\nembodied navigation task. This requires a model to control steering by\ncomparing current visual inputs to memories stored along a training route. We\nconcluded the FlyHash model is more efficient than others, especially in terms\nof data encoding.\n", "date": "2023-03-14", "categories": []}
{"id": "2304.01015", "title": "Adaptive structure evolution and biologically plausible synaptic\n  plasticity for recurrent spiking neural networks", "abstract": "  The architecture design and multi-scale learning principles of the human\nbrain that evolved over hundreds of millions of years are crucial to realizing\nhuman-like intelligence. Spiking Neural Network (SNN) based Liquid State\nMachine (LSM) serves as a suitable architecture to study brain-inspired\nintelligence because of its brain-inspired structure and the potential for\nintegrating multiple biological principles. Existing researches on LSM focus on\ndifferent certain perspectives, including high-dimensional encoding or\noptimization of the liquid layer, network architecture search, and application\nto hardware devices. There is still a lack of in-depth inspiration from the\nlearning and structural evolution mechanism of the brain. Considering these\nlimitations, this paper presents a novel LSM learning model that integrates\nadaptive structural evolution and multi-scale biological learning rules. For\nstructural evolution, an adaptive evolvable LSM model is developed to optimize\nthe neural architecture design of liquid layer with separation property. For\nbrain-inspired learning of LSM, we propose a dopamine-modulated\nBienenstock-Cooper-Munros (DA-BCM) method that incorporates global long-term\ndopamine regulation and local trace-based BCM synaptic plasticity. Comparative\nexperimental results on different decision-making tasks show that introducing\nstructural evolution of the liquid layer, and the DA-BCM regulation of the\nliquid layer and the readout layer could improve the decision-making ability of\nLSM and flexibly adapt to rule reversal. This work is committed to exploring\nhow evolution can help to design more appropriate network architectures and how\nmulti-scale neuroplasticity principles coordinated to enable the optimization\nand learning of LSMs for relatively complex decision-making tasks.\n", "date": "2023-03-31", "categories": []}
{"id": "2210.06756", "title": "Decoding Visual Neural Representations by Multimodal Learning of\n  Brain-Visual-Linguistic Features", "abstract": "  Decoding human visual neural representations is a challenging task with great\nscientific significance in revealing vision-processing mechanisms and\ndeveloping brain-like intelligent machines. Most existing methods are difficult\nto generalize to novel categories that have no corresponding neural data for\ntraining. The two main reasons are 1) the under-exploitation of the multimodal\nsemantic knowledge underlying the neural data and 2) the small number of paired\n(stimuli-responses) training data. To overcome these limitations, this paper\npresents a generic neural decoding method called BraVL that uses multimodal\nlearning of brain-visual-linguistic features. We focus on modeling the\nrelationships between brain, visual and linguistic features via multimodal deep\ngenerative models. Specifically, we leverage the mixture-of-product-of-experts\nformulation to infer a latent code that enables a coherent joint generation of\nall three modalities. To learn a more consistent joint representation and\nimprove the data efficiency in the case of limited brain activity data, we\nexploit both intra- and inter-modality mutual information maximization\nregularization terms. In particular, our BraVL model can be trained under\nvarious semi-supervised scenarios to incorporate the visual and textual\nfeatures obtained from the extra categories. Finally, we construct three\ntrimodal matching datasets, and the extensive experiments lead to some\ninteresting conclusions and cognitive insights: 1) decoding novel visual\ncategories from human brain activity is practically possible with good\naccuracy; 2) decoding models using the combination of visual and linguistic\nfeatures perform much better than those using either of them alone; 3) visual\nperception may be accompanied by linguistic influences to represent the\nsemantics of visual stimuli. Code and data: https://github.com/ChangdeDu/BraVL.\n", "date": "2022-10-13", "categories": []}
{"id": "2211.06956", "title": "Seeing Beyond the Brain: Conditional Diffusion Model with Sparse Masked\n  Modeling for Vision Decoding", "abstract": "  Decoding visual stimuli from brain recordings aims to deepen our\nunderstanding of the human visual system and build a solid foundation for\nbridging human and computer vision through the Brain-Computer Interface.\nHowever, reconstructing high-quality images with correct semantics from brain\nrecordings is a challenging problem due to the complex underlying\nrepresentations of brain signals and the scarcity of data annotations. In this\nwork, we present MinD-Vis: Sparse Masked Brain Modeling with Double-Conditioned\nLatent Diffusion Model for Human Vision Decoding. Firstly, we learn an\neffective self-supervised representation of fMRI data using mask modeling in a\nlarge latent space inspired by the sparse coding of information in the primary\nvisual cortex. Then by augmenting a latent diffusion model with\ndouble-conditioning, we show that MinD-Vis can reconstruct highly plausible\nimages with semantically matching details from brain recordings using very few\npaired annotations. We benchmarked our model qualitatively and quantitatively;\nthe experimental results indicate that our method outperformed state-of-the-art\nin both semantic mapping (100-way semantic classification) and generation\nquality (FID) by 66% and 41% respectively. An exhaustive ablation study was\nalso conducted to analyze our framework.\n", "date": "2022-11-13", "categories": []}
{"id": "2303.13651", "title": "Building artificial neural circuits for domain-general cognition: a\n  primer on brain-inspired systems-level architecture", "abstract": "  There is a concerted effort to build domain-general artificial intelligence\nin the form of universal neural network models with sufficient computational\nflexibility to solve a wide variety of cognitive tasks but without requiring\nfine-tuning on individual problem spaces and domains. To do this, models need\nappropriate priors and inductive biases, such that trained models can\ngeneralise to out-of-distribution examples and new problem sets. Here we\nprovide an overview of the hallmarks endowing biological neural networks with\nthe functionality needed for flexible cognition, in order to establish which\nfeatures might also be important to achieve similar functionality in artificial\nsystems. We specifically discuss the role of system-level distribution of\nnetwork communication and recurrence, in addition to the role of short-term\ntopological changes for efficient local computation. As machine learning models\nbecome more complex, these principles may provide valuable directions in an\notherwise vast space of possible architectures. In addition, testing these\ninductive biases within artificial systems may help us to understand the\nbiological principles underlying domain-general cognition.\n", "date": "2023-03-21", "categories": []}
{"id": "2212.06726", "title": "Semantic Brain Decoding: from fMRI to conceptually similar image\n  reconstruction of visual stimuli", "abstract": "  Brain decoding is a field of computational neuroscience that uses measurable\nbrain activity to infer mental states or internal representations of perceptual\ninputs. Therefore, we propose a novel approach to brain decoding that also\nrelies on semantic and contextual similarity. We employ an fMRI dataset of\nnatural image vision and create a deep learning decoding pipeline inspired by\nthe existence of both bottom-up and top-down processes in human vision. We\ntrain a linear brain-to-feature model to map fMRI activity features to visual\nstimuli features, assuming that the brain projects visual information onto a\nspace that is homeomorphic to the latent space represented by the last\nconvolutional layer of a pretrained convolutional neural network, which\ntypically collects a variety of semantic features that summarize and highlight\nsimilarities and differences between concepts. These features are then\ncategorized in the latent space using a nearest-neighbor strategy, and the\nresults are used to condition a generative latent diffusion model to create\nnovel images. From fMRI data only, we produce reconstructions of visual stimuli\nthat match the original content very well on a semantic level, surpassing the\nstate of the art in previous literature. We evaluate our work and obtain good\nresults using a quantitative semantic metric (the Wu-Palmer similarity metric\nover the WordNet lexicon, which had an average value of 0.57) and perform a\nhuman evaluation experiment that resulted in correct evaluation, according to\nthe multiplicity of human criteria in evaluating image similarity, in over 80%\nof the test set.\n", "date": "2022-12-13", "categories": []}
{"id": "2206.01685", "title": "Toward a realistic model of speech processing in the brain with\n  self-supervised learning", "abstract": "  Several deep neural networks have recently been shown to generate activations\nsimilar to those of the brain in response to the same input. These algorithms,\nhowever, remain largely implausible: they require (1) extraordinarily large\namounts of data, (2) unobtainable supervised labels, (3) textual rather than\nraw sensory input, and / or (4) implausibly large memory (e.g. thousands of\ncontextual words). These elements highlight the need to identify algorithms\nthat, under these limitations, would suffice to account for both behavioral and\nbrain responses. Focusing on the issue of speech processing, we here\nhypothesize that self-supervised algorithms trained on the raw waveform\nconstitute a promising candidate. Specifically, we compare a recent\nself-supervised architecture, Wav2Vec 2.0, to the brain activity of 412\nEnglish, French, and Mandarin individuals recorded with functional Magnetic\nResonance Imaging (fMRI), while they listened to ~1h of audio books. Our\nresults are four-fold. First, we show that this algorithm learns brain-like\nrepresentations with as little as 600 hours of unlabelled speech -- a quantity\ncomparable to what infants can be exposed to during language acquisition.\nSecond, its functional hierarchy aligns with the cortical hierarchy of speech\nprocessing. Third, different training regimes reveal a functional\nspecialization akin to the cortex: Wav2Vec 2.0 learns sound-generic,\nspeech-specific and language-specific representations similar to those of the\nprefrontal and temporal cortices. Fourth, we confirm the similarity of this\nspecialization with the behavior of 386 additional participants. These\nelements, resulting from the largest neuroimaging benchmark to date, show how\nself-supervised learning can account for a rich organization of speech\nprocessing in the brain, and thus delineate a path to identify the laws of\nlanguage acquisition which shape the human brain.\n", "date": "2022-06-03", "categories": []}
{"id": "2303.08877", "title": "ROSE: A Neurocomputational Architecture for Syntax", "abstract": "  A comprehensive model of natural language processing in the brain must\naccommodate four components: representations, operations, structures and\nencoding. It further requires a principled account of how these components\nmechanistically, and causally, relate to each another. While previous models\nhave isolated regions of interest for structure-building and lexical access,\nmany gaps remain with respect to bridging distinct scales of neural complexity.\nBy expanding existing accounts of how neural oscillations can index various\nlinguistic processes, this article proposes a neurocomputational architecture\nfor syntax, termed the ROSE model (Representation, Operation, Structure,\nEncoding). Under ROSE, the basic data structures of syntax are atomic features,\ntypes of mental representations (R), and are coded at the single-unit and\nensemble level. Elementary computations (O) that transform these units into\nmanipulable objects accessible to subsequent structure-building levels are\ncoded via high frequency gamma activity. Low frequency synchronization and\ncross-frequency coupling code for recursive categorial inferences (S). Distinct\nforms of low frequency coupling and phase-amplitude coupling (delta-theta\ncoupling via pSTS-IFG; theta-gamma coupling via IFG to conceptual hubs) then\nencode these structures onto distinct workspaces (E). Causally connecting R to\nO is spike-phase/LFP coupling; connecting O to S is phase-amplitude coupling;\nconnecting S to E is a system of frontotemporal traveling oscillations;\nconnecting E to lower levels is low-frequency phase resetting of spike-LFP\ncoupling. ROSE is reliant on neurophysiologically plausible mechanisms, is\nsupported at all four levels by a range of recent empirical research, and\nprovides an anatomically precise and falsifiable grounding for the basic\nproperty of natural language syntax: hierarchical, recursive\nstructure-building.\n", "date": "2023-03-15", "categories": []}
{"id": "2303.04391", "title": "A Deep-Learning-Based Neural Decoding Framework for Emotional\n  Brain-Computer Interfaces", "abstract": "  Reading emotions precisely from segments of neural activity is crucial for\nthe development of emotional brain-computer interfaces. Among all neural\ndecoding algorithms, deep learning (DL) holds the potential to become the most\npromising one, yet progress has been limited in recent years. One possible\nreason is that the efficacy of DL strongly relies on training samples, yet the\nneural data used for training are often from non-human primates and mixed with\nplenty of noise, which in turn mislead the training of DL models. Given it is\ndifficult to accurately determine animals' emotions from humans' perspective,\nwe assume the dominant noise in neural data representing different emotions is\nthe labeling error. Here, we report the development and application of a neural\ndecoding framework called Emo-Net that consists of a confidence learning (CL)\ncomponent and a DL component. The framework is fully data-driven and is capable\nof decoding emotions from multiple datasets obtained from behaving monkeys. In\naddition to improving the decoding ability, Emo-Net significantly improves the\nperformance of the base DL models, making emotion recognition in animal models\npossible. In summary, this framework may inspire novel understandings of the\nneural basis of emotion and drive the realization of close-loop emotional\nbrain-computer interfaces.\n", "date": "2023-03-08", "categories": []}
{"id": "2212.04938", "title": "Emergent Computations in Trained Artificial Neural Networks and Real\n  Brains", "abstract": "  Synaptic plasticity allows cortical circuits to learn new tasks and to adapt\nto changing environments. How do cortical circuits use plasticity to acquire\nfunctions such as decision-making or working memory? Neurons are connected in\ncomplex ways, forming recurrent neural networks, and learning modifies the\nstrength of their connections. Moreover, neurons communicate emitting brief\ndiscrete electric signals. Here we describe how to train recurrent neural\nnetworks in tasks like those used to train animals in neuroscience\nlaboratories, and how computations emerge in the trained networks.\nSurprisingly, artificial networks and real brains can use similar computational\nstrategies.\n", "date": "2022-12-09", "categories": []}
{"id": "2212.02409", "title": "Decoding natural image stimuli from fMRI data with a surface-based\n  convolutional network", "abstract": "  Due to the low signal-to-noise ratio and limited resolution of functional MRI\ndata, and the high complexity of natural images, reconstructing a visual\nstimulus from human brain fMRI measurements is a challenging task. In this\nwork, we propose a novel approach for this task, which we call Cortex2Image, to\ndecode visual stimuli with high semantic fidelity and rich fine-grained detail.\nIn particular, we train a surface-based convolutional network model that maps\nfrom brain response to semantic image features first (Cortex2Semantic). We then\ncombine this model with a high-quality image generator (Instance-Conditioned\nGAN) to train another mapping from brain response to fine-grained image\nfeatures using a variational approach (Cortex2Detail). Image reconstructions\nobtained by our proposed method achieve state-of-the-art semantic fidelity,\nwhile yielding good fine-grained similarity with the ground-truth stimulus. Our\ncode is available at: https://github.com/zijin-gu/meshconv-decoding.git.\n", "date": "2022-12-05", "categories": []}
{"id": "2302.14389", "title": "Information-Restricted Neural Language Models Reveal Different Brain\n  Regions' Sensitivity to Semantics, Syntax and Context", "abstract": "  A fundamental question in neurolinguistics concerns the brain regions\ninvolved in syntactic and semantic processing during speech comprehension, both\nat the lexical (word processing) and supra-lexical levels (sentence and\ndiscourse processing). To what extent are these regions separated or\nintertwined? To address this question, we trained a lexical language model,\nGlove, and a supra-lexical language model, GPT-2, on a text corpus from which\nwe selectively removed either syntactic or semantic information. We then\nassessed to what extent these information-restricted models were able to\npredict the time-courses of fMRI signal of humans listening to naturalistic\ntext. We also manipulated the size of contextual information provided to GPT-2\nin order to determine the windows of integration of brain regions involved in\nsupra-lexical processing. Our analyses show that, while most brain regions\ninvolved in language are sensitive to both syntactic and semantic variables,\nthe relative magnitudes of these effects vary a lot across these regions.\nFurthermore, we found an asymmetry between the left and right hemispheres, with\nsemantic and syntactic processing being more dissociated in the left hemisphere\nthan in the right, and the left and right hemispheres showing respectively\ngreater sensitivity to short and long contexts. The use of\ninformation-restricted NLP models thus shed new light on the spatial\norganization of syntactic processing, semantic processing and compositionality.\n", "date": "2023-02-28", "categories": []}
{"id": "2302.13378", "title": "Puppeteer and Marionette: Learning Anticipatory Quadrupedal Locomotion\n  Based on Interactions of a Central Pattern Generator and Supraspinal Drive", "abstract": "  Quadruped animal locomotion emerges from the interactions between the spinal\ncentral pattern generator (CPG), sensory feedback, and supraspinal drive\nsignals from the brain. Computational models of CPGs have been widely used for\ninvestigating the spinal cord contribution to animal locomotion control in\ncomputational neuroscience and in bio-inspired robotics. However, the\ncontribution of supraspinal drive to anticipatory behavior, i.e. motor behavior\nthat involves planning ahead of time (e.g. of footstep placements), is not yet\nproperly understood. In particular, it is not clear whether the brain modulates\nCPG activity and/or directly modulates muscle activity (hence bypassing the\nCPG) for accurate foot placements. In this paper, we investigate the\ninteraction of supraspinal drive and a CPG in an anticipatory locomotion\nscenario that involves stepping over gaps. By employing deep reinforcement\nlearning (DRL), we train a neural network policy that replicates the\nsupraspinal drive behavior. This policy can either modulate the CPG dynamics,\nor directly change actuation signals to bypass the CPG dynamics. Our results\nindicate that the direct supraspinal contribution to the actuation signal is a\nkey component for a high gap crossing success rate. However, the CPG dynamics\nin the spinal cord are beneficial for gait smoothness and energy efficiency.\nMoreover, our investigation shows that sensing the front feet distances to the\ngap is the most important and sufficient sensory information for learning gap\ncrossing. Our results support the biological hypothesis that cats and horses\nmainly control the front legs for obstacle avoidance, and that hind limbs\nfollow an internal memory based on the front limbs' information. Our method\nenables the quadruped robot to cross gaps of up to 20 cm (50% of body-length)\nwithout any explicit dynamics modeling or Model Predictive Control (MPC).\n", "date": "2023-02-26", "categories": []}
{"id": "2302.11162", "title": "Sparse, Geometric Autoencoder Models of V1", "abstract": "  The classical sparse coding model represents visual stimuli as a linear\ncombination of a handful of learned basis functions that are Gabor-like when\ntrained on natural image data. However, the Gabor-like filters learned by\nclassical sparse coding far overpredict well-tuned simple cell receptive field\n(SCRF) profiles. A number of subsequent models have either discarded the sparse\ndictionary learning framework entirely or have yet to take advantage of the\nsurge in unrolled, neural dictionary learning architectures. A key missing\ntheme of these updates is a stronger notion of \\emph{structured sparsity}. We\npropose an autoencoder architecture whose latent representations are\nimplicitly, locally organized for spectral clustering, which begets artificial\nneurons better matched to observed primate data. The weighted-$\\ell_1$ (WL)\nconstraint in the autoencoder objective function maintains core ideas of the\nsparse coding framework, yet also offers a promising path to describe the\ndifferentiation of receptive fields in terms of a discriminative hierarchy in\nfuture work.\n", "date": "2023-02-22", "categories": []}
{"id": "2302.08589", "title": "Syntactic Structure Processing in the Brain while Listening", "abstract": "  Syntactic parsing is the task of assigning a syntactic structure to a\nsentence. There are two popular syntactic parsing methods: constituency and\ndependency parsing. Recent works have used syntactic embeddings based on\nconstituency trees, incremental top-down parsing, and other word syntactic\nfeatures for brain activity prediction given the text stimuli to study how the\nsyntax structure is represented in the brain's language network. However, the\neffectiveness of dependency parse trees or the relative predictive power of the\nvarious syntax parsers across brain areas, especially for the listening task,\nis yet unexplored. In this study, we investigate the predictive power of the\nbrain encoding models in three settings: (i) individual performance of the\nconstituency and dependency syntactic parsing based embedding methods, (ii)\nefficacy of these syntactic parsing based embedding methods when controlling\nfor basic syntactic signals, (iii) relative effectiveness of each of the\nsyntactic embedding methods when controlling for the other. Further, we explore\nthe relative importance of syntactic information (from these syntactic\nembedding methods) versus semantic information using BERT embeddings. We find\nthat constituency parsers help explain activations in the temporal lobe and\nmiddle-frontal gyrus, while dependency parsers better encode syntactic\nstructure in the angular gyrus and posterior cingulate cortex. Although\nsemantic signals from BERT are more effective compared to any of the syntactic\nfeatures or embedding methods, syntactic embedding methods explain additional\nvariance for a few brain regions.\n", "date": "2023-02-16", "categories": []}
{"id": "2210.03155", "title": "Understanding Neural Coding on Latent Manifolds by Sharing Features and\n  Dividing Ensembles", "abstract": "  Systems neuroscience relies on two complementary views of neural data,\ncharacterized by single neuron tuning curves and analysis of population\nactivity. These two perspectives combine elegantly in neural latent variable\nmodels that constrain the relationship between latent variables and neural\nactivity, modeled by simple tuning curve functions. This has recently been\ndemonstrated using Gaussian processes, with applications to realistic and\ntopologically relevant latent manifolds. Those and previous models, however,\nmissed crucial shared coding properties of neural populations. We propose\nfeature sharing across neural tuning curves which significantly improves\nperformance and helps optimization. We also propose a solution to the ensemble\ndetection problem, where different groups of neurons, i.e., ensembles, can be\nmodulated by different latent manifolds. Achieved through a soft clustering of\nneurons during training, this allows for the separation of mixed neural\npopulations in an unsupervised manner. These innovations lead to more\ninterpretable models of neural population activity that train well and perform\nbetter even on mixtures of complex latent manifolds. Finally, we apply our\nmethod on a recently published grid cell dataset, and recover distinct\nensembles, infer toroidal latents and predict neural tuning curves in a single\nintegrated modeling framework.\n", "date": "2022-10-06", "categories": []}
{"id": "2301.06340", "title": "Deep Learning Models to Study Sentence Comprehension in the Human Brain", "abstract": "  Recent artificial neural networks that process natural language achieve\nunprecedented performance in tasks requiring sentence-level understanding. As\nsuch, they could be interesting models of the integration of linguistic\ninformation in the human brain. We review works that compare these artificial\nlanguage models with human brain activity and we assess the extent to which\nthis approach has improved our understanding of the neural processes involved\nin natural language comprehension. Two main results emerge. First, the neural\nrepresentation of word meaning aligns with the context-dependent, dense word\nvectors used by the artificial neural networks. Second, the processing\nhierarchy that emerges within artificial neural networks broadly matches the\nbrain, but is surprisingly inconsistent across studies. We discuss current\nchallenges in establishing artificial neural networks as process models of\nnatural language comprehension. We suggest exploiting the highly structured\nrepresentational geometry of artificial neural networks when mapping\nrepresentations to brain data.\n", "date": "2023-01-16", "categories": []}
{"id": "2301.09565", "title": "Estimating the energy requirements for long term memory formation", "abstract": "  Brains consume metabolic energy to process information, but also to store\nmemories. The energy required for memory formation can be substantial, for\ninstance in fruit flies memory formation leads to a shorter lifespan upon\nsubsequent starvation (Mery and Kawecki, 2005). Here we estimate that the\nenergy required corresponds to about 10mJ/bit and compare this to biophysical\nestimates as well as energy requirements in computer hardware. We conclude that\nbiological memory storage is expensive, but the reason behind it is not known.\n", "date": "2023-01-16", "categories": []}
{"id": "2301.08391", "title": "Brain Model State Space Reconstruction Using an LSTM Neural Network", "abstract": "  Objective\n  Kalman filtering has previously been applied to track neural model states and\nparameters, particularly at the scale relevant to EEG. However, this approach\nlacks a reliable method to determine the initial filter conditions and assumes\nthat the distribution of states remains Gaussian. This study presents an\nalternative, data-driven method to track the states and parameters of neural\nmass models (NMMs) from EEG recordings using deep learning techniques,\nspecifically an LSTM neural network.\n  Approach\n  An LSTM filter was trained on simulated EEG data generated by a neural mass\nmodel using a wide range of parameters. With an appropriately customised loss\nfunction, the LSTM filter can learn the behaviour of NMMs. As a result, it can\noutput the state vector and parameters of NMMs given observation data as the\ninput.\n  Main Results\n  Test results using simulated data yielded correlations with R squared of\naround 0.99 and verified that the method is robust to noise and can be more\naccurate than a nonlinear Kalman filter when the initial conditions of the\nKalman filter are not accurate. As an example of real-world application, the\nLSTM filter was also applied to real EEG data that included epileptic seizures,\nand revealed changes in connectivity strength parameters at the beginnings of\nseizures.\n  Significance\n  Tracking the state vector and parameters of mathematical brain models is of\ngreat importance in the area of brain modelling, monitoring, imaging and\ncontrol. This approach has no need to specify the initial state vector and\nparameters, which is very difficult to do in practice because many of the\nvariables being estimated cannot be measured directly in physiological\nexperiments. This method may be applied using any neural mass model and,\ntherefore, provides a general, novel, efficient approach to estimate brain\nmodel variables that are often difficult to measure.\n", "date": "2023-01-20", "categories": []}
{"id": "2209.14406", "title": "Biological connectomes as a representation for the architecture of\n  artificial neural networks", "abstract": "  Grand efforts in neuroscience are working toward mapping the connectomes of\nmany new species, including the near completion of the Drosophila melanogaster.\nIt is important to ask whether these models could benefit artificial\nintelligence. In this work we ask two fundamental questions: (1) where and when\nbiological connectomes can provide use in machine learning, (2) which design\nprinciples are necessary for extracting a good representation of the\nconnectome. Toward this end, we translate the motor circuit of the C. Elegans\nnematode into artificial neural networks at varying levels of biophysical\nrealism and evaluate the outcome of training these networks on motor and\nnon-motor behavioral tasks. We demonstrate that biophysical realism need not be\nupheld to attain the advantages of using biological circuits. We also establish\nthat, even if the exact wiring diagram is not retained, the architectural\nstatistics provide a valuable prior. Finally, we show that while the C. Elegans\nlocomotion circuit provides a powerful inductive bias on locomotion problems,\nits structure may hinder performance on tasks unrelated to locomotion such as\nvisual classification problems.\n", "date": "2022-09-28", "categories": []}
{"id": "2212.10367", "title": "Modeling Human Eye Movements with Neural Networks in a Maze-Solving Task", "abstract": "  From smoothly pursuing moving objects to rapidly shifting gazes during visual\nsearch, humans employ a wide variety of eye movement strategies in different\ncontexts. While eye movements provide a rich window into mental processes,\nbuilding generative models of eye movements is notoriously difficult, and to\ndate the computational objectives guiding eye movements remain largely a\nmystery. In this work, we tackled these problems in the context of a canonical\nspatial planning task, maze-solving. We collected eye movement data from human\nsubjects and built deep generative models of eye movements using a novel\ndifferentiable architecture for gaze fixations and gaze shifts. We found that\nhuman eye movements are best predicted by a model that is optimized not to\nperform the task as efficiently as possible but instead to run an internal\nsimulation of an object traversing the maze. This not only provides a\ngenerative model of eye movements in this task but also suggests a\ncomputational theory for how humans solve the task, namely that humans use\nmental simulation.\n", "date": "2022-12-20", "categories": []}
{"id": "2212.08205", "title": "A unified information-theoretic model of EEG signatures of human\n  language processing", "abstract": "  We advance an information-theoretic model of human language processing in the\nbrain, in which incoming linguistic input is processed at two levels, in terms\nof a heuristic interpretation and in terms of error correction. We propose that\nthese two kinds of information processing have distinct electroencephalographic\nsignatures, corresponding to the well-documented N400 and P600 components of\nlanguage-related event-related potentials (ERPs). Formally, we show that the\ninformation content (surprisal) of a word in context can be decomposed into two\nquantities: (A) heuristic surprise, which signals processing difficulty of word\ngiven its inferred context, and corresponds with the N400 signal; and (B)\ndiscrepancy signal, which reflects divergence between the true context and the\ninferred context, and corresponds to the P600 signal. Both of these quantities\ncan be estimated using modern NLP techniques. We validate our theory by\nsuccessfully simulating ERP patterns elicited by a variety of linguistic\nmanipulations in previously-reported experimental data from Ryskin et al.\n(2021). Our theory is in principle compatible with traditional cognitive\ntheories assuming a `good-enough' heuristic interpretation stage, but with\nprecise information-theoretic formulation.\n", "date": "2022-12-16", "categories": []}
{"id": "2212.08626", "title": "Hippocampus-Inspired Cognitive Architecture (HICA) for Operant\n  Conditioning", "abstract": "  The neural implementation of operant conditioning with few trials is unclear.\nWe propose a Hippocampus-Inspired Cognitive Architecture (HICA) as a neural\nmechanism for operant conditioning. HICA explains a learning mechanism in which\nagents can learn a new behavior policy in a few trials, as mammals do in\noperant conditioning experiments. HICA is composed of two different types of\nmodules. One is a universal learning module type that represents a cortical\ncolumn in the neocortex gray matter. The working principle is modeled as\nModulated Heterarchical Prediction Memory (mHPM). In mHPM, each module learns\nto predict a succeeding input vector given the sequence of the input vectors\nfrom lower layers and the context vectors from higher layers. The prediction is\nfed into the lower layers as a context signal (top-down feedback signaling),\nand into the higher layers as an input signal (bottom-up feedforward\nsignaling). Rewards modulate the learning rate in those modules to memorize\nmeaningful sequences effectively. In mHPM, each module updates in a local and\ndistributed way compared to conventional end-to-end learning with\nbackpropagation of the single objective loss. This local structure enables the\nheterarchical network of modules. The second type is an innate, special-purpose\nmodule representing various organs of the brain's subcortical system. Modules\nmodeling organs such as the amygdala, hippocampus, and reward center are\npre-programmed to enable instinctive behaviors. The hippocampus plays the role\nof the simulator. It is an autoregressive prediction model of the top-most\nlevel signal with a loop structure of memory, while cortical columns are lower\nlayers that provide detailed information to the simulation. The simulation\nbecomes the basis for learning with few trials and the deliberate planning\nrequired for operant conditioning.\n", "date": "2022-12-16", "categories": []}
{"id": "2212.04589", "title": "Optimizing Integrated Information with a Prior Guided Random Search\n  Algorithm", "abstract": "  Integrated information theory (IIT) is a theoretical framework that provides\na quantitative measure to estimate when a physical system is conscious, its\ndegree of consciousness, and the complexity of the qualia space that the system\nis experiencing. Formally, IIT rests on the assumption that if a surrogate\nphysical system can fully embed the phenomenological properties of\nconsciousness, then the system properties must be constrained by the properties\nof the qualia being experienced. Following this assumption, IIT represents the\nphysical system as a network of interconnected elements that can be thought of\nas a probabilistic causal graph, $\\mathcal{G}$, where each node has an\ninput-output function and all the graph is encoded in a transition probability\nmatrix. Consequently, IIT's quantitative measure of consciousness, $\\Phi$, is\ncomputed with respect to the transition probability matrix and the present\nstate of the graph. In this paper, we provide a random search algorithm that is\nable to optimize $\\Phi$ in order to investigate, as the number of nodes\nincreases, the structure of the graphs that have higher $\\Phi$. We also provide\narguments that show the difficulties of applying more complex black-box search\nalgorithms, such as Bayesian optimization or metaheuristics, in this particular\nproblem. Additionally, we suggest specific research lines for these techniques\nto enhance the search algorithm that guarantees maximal $\\Phi$.\n", "date": "2022-12-08", "categories": []}
{"id": "2212.04219", "title": "The Neural Correlates of Linguistic Structure Building: Comments on\n  Kazanina & Tavano (2022)", "abstract": "  A recent perspective paper by Kazanina & Tavano (referred to as the KT\nperspective in the following) argues how neural oscillations cannot provide a\npotential neural correlate for syntactic structure building. The view that\nneural oscillations can provide a potential neural correlate for syntactic\nstructure building is largely attributed to a study by Ding, Melloni, Zhang,\nTian, and Poeppel in 2016 (referred to as the DMZTP study).\n  The KT perspective is thought provoking, but has severe misinterpretations\nabout the arguments in DMZTP and other studies, and contains contradictory\nconclusions in different parts of the perspective, making it impossible to\nunderstand the position of the authors. In the following, I summarize a few\nmisinterpretations and inconsistent arguments in the KT perspective, and put\nforward a few suggestions for future studies.\n", "date": "2022-12-08", "categories": []}
{"id": "2209.12894", "title": "Biologically-Plausible Determinant Maximization Neural Networks for\n  Blind Separation of Correlated Sources", "abstract": "  Extraction of latent sources of complex stimuli is critical for making sense\nof the world. While the brain solves this blind source separation (BSS) problem\ncontinuously, its algorithms remain unknown. Previous work on\nbiologically-plausible BSS algorithms assumed that observed signals are linear\nmixtures of statistically independent or uncorrelated sources, limiting the\ndomain of applicability of these algorithms. To overcome this limitation, we\npropose novel biologically-plausible neural networks for the blind separation\nof potentially dependent/correlated sources. Differing from previous work, we\nassume some general geometric, not statistical, conditions on the source\nvectors allowing separation of potentially dependent/correlated sources.\nConcretely, we assume that the source vectors are sufficiently scattered in\ntheir domains which can be described by certain polytopes. Then, we consider\nrecovery of these sources by the Det-Max criterion, which maximizes the\ndeterminant of the output correlation matrix to enforce a similar spread for\nthe source estimates. Starting from this normative principle, and using a\nweighted similarity matching approach that enables arbitrary linear\ntransformations adaptable by local learning rules, we derive two-layer\nbiologically-plausible neural network algorithms that can separate mixtures\ninto sources coming from a variety of source domains. We demonstrate that our\nalgorithms outperform other biologically-plausible BSS algorithms on correlated\nsource separation problems.\n", "date": "2022-09-27", "categories": []}
{"id": "2211.12935", "title": "Functional Connectome: Approximating Brain Networks with Artificial\n  Neural Networks", "abstract": "  We aimed to explore the capability of deep learning to approximate the\nfunction instantiated by biological neural circuits-the functional connectome.\nUsing deep neural networks, we performed supervised learning with firing rate\nobservations drawn from synthetically constructed neural circuits, as well as\nfrom an empirically supported Boundary Vector Cell-Place Cell network. The\nperformance of trained networks was quantified using a range of criteria and\ntasks. Our results show that deep neural networks were able to capture the\ncomputations performed by synthetic biological networks with high accuracy, and\nwere highly data efficient and robust to biological plasticity. We show that\ntrained deep neural networks are able to perform zero-shot generalisation in\nnovel environments, and allows for a wealth of tasks such as decoding the\nanimal's location in space with high accuracy. Our study reveals a novel and\npromising direction in systems neuroscience, and can be expanded upon with a\nmultitude of downstream applications, for example, goal-directed reinforcement\nlearning.\n", "date": "2022-11-23", "categories": []}
{"id": "2211.12100", "title": "Simulating Human Gaze with Neural Visual Attention", "abstract": "  Existing models of human visual attention are generally unable to incorporate\ndirect task guidance and therefore cannot model an intent or goal when\nexploring a scene. To integrate guidance of any downstream visual task into\nattention modeling, we propose the Neural Visual Attention (NeVA) algorithm. To\nthis end, we impose to neural networks the biological constraint of foveated\nvision and train an attention mechanism to generate visual explorations that\nmaximize the performance with respect to the downstream task. We observe that\nbiologically constrained neural networks generate human-like scanpaths without\nbeing trained for this objective. Extensive experiments on three common\nbenchmark datasets show that our method outperforms state-of-the-art\nunsupervised human attention models in generating human-like scanpaths.\n", "date": "2022-11-22", "categories": []}
{"id": "2211.10419", "title": "A Neural Active Inference Model of Perceptual-Motor Learning", "abstract": "  The active inference framework (AIF) is a promising new computational\nframework grounded in contemporary neuroscience that can produce human-like\nbehavior through reward-based learning. In this study, we test the ability for\nthe AIF to capture the role of anticipation in the visual guidance of action in\nhumans through the systematic investigation of a visual-motor task that has\nbeen well-explored -- that of intercepting a target moving over a ground plane.\nPrevious research demonstrated that humans performing this task resorted to\nanticipatory changes in speed intended to compensate for semi-predictable\nchanges in target speed later in the approach. To capture this behavior, our\nproposed \"neural\" AIF agent uses artificial neural networks to select actions\non the basis of a very short term prediction of the information about the task\nenvironment that these actions would reveal along with a long-term estimate of\nthe resulting cumulative expected free energy. Systematic variation revealed\nthat anticipatory behavior emerged only when required by limitations on the\nagent's movement capabilities, and only when the agent was able to estimate\naccumulated free energy over sufficiently long durations into the future. In\naddition, we present a novel formulation of the prior function that maps a\nmulti-dimensional world-state to a uni-dimensional distribution of free-energy.\nTogether, these results demonstrate the use of AIF as a plausible model of\nanticipatory visually guided behavior in humans.\n", "date": "2022-11-16", "categories": []}
{"id": "2211.09295", "title": "Testing for context-dependent changes in neural encoding in naturalistic\n  experiments", "abstract": "  We propose a decoding-based approach to detect context effects on neural\ncodes in longitudinal neural recording data. The approach is agnostic to how\ninformation is encoded in neural activity, and can control for a variety of\npossible confounding factors present in the data. We demonstrate our approach\nby determining whether it is possible to decode location encoding from\nprefrontal cortex in the mouse and, further, testing whether the encoding\nchanges due to task engagement.\n", "date": "2022-11-17", "categories": []}
{"id": "2211.09096", "title": "Testing geometric representation hypotheses from simulated place cell\n  recordings", "abstract": "  Hippocampal place cells can encode spatial locations of an animal in physical\nor task-relevant spaces. We simulated place cell populations that encoded\neither Euclidean- or graph-based positions of a rat navigating to goal nodes in\na maze with a graph topology, and used manifold learning methods such as UMAP\nand Autoencoders (AE) to analyze these neural population activities. The\nstructure of the latent spaces learned by the AE reflects their true geometric\nstructure, while PCA fails to do so and UMAP is less robust to noise. Our\nresults support future applications of AE architectures to decipher the\ngeometry of spatial encoding in the brain.\n", "date": "2022-11-16", "categories": []}
{"id": "2210.02684", "title": "Conformal Isometry of Lie Group Representation in Recurrent Network of\n  Grid Cells", "abstract": "  The activity of the grid cell population in the medial entorhinal cortex\n(MEC) of the mammalian brain forms a vector representation of the self-position\nof the animal. Recurrent neural networks have been proposed to explain the\nproperties of the grid cells by updating the neural activity vector based on\nthe velocity input of the animal. In doing so, the grid cell system effectively\nperforms path integration. In this paper, we investigate the algebraic,\ngeometric, and topological properties of grid cells using recurrent network\nmodels. Algebraically, we study the Lie group and Lie algebra of the recurrent\ntransformation as a representation of self-motion. Geometrically, we study the\nconformal isometry of the Lie group representation where the local displacement\nof the activity vector in the neural space is proportional to the local\ndisplacement of the agent in the 2D physical space. Topologically, the compact\nabelian Lie group representation automatically leads to the torus topology\ncommonly assumed and observed in neuroscience. We then focus on a simple\nnon-linear recurrent model that underlies the continuous attractor neural\nnetworks of grid cells. Our numerical experiments show that conformal isometry\nleads to hexagon periodic patterns in the grid cell responses and our model is\ncapable of accurate path integration. Code is available at\n\\url{https://github.com/DehongXu/grid-cell-rnn}.\n", "date": "2022-10-06", "categories": []}
{"id": "2210.14961", "title": "A Neural Network Based Automated IFT-20 Sensory Neuron Classifier for\n  Caenorhabditis elegans", "abstract": "  Determining neuronal identity in imaging data is an essential task in\nneuroscience, facilitating the comparison of neural activity across organisms.\nCross-organism comparison, in turn, enables a wide variety of research\nincluding whole-brain analysis of functional networks and linking the activity\nof specific neurons to behavior or environmental stimuli. The recent\ndevelopment of three-dimensional, pan-neuronal imaging with single-cell\nresolution within Caenorhabditis elegans has brought neuron identification,\ntracking, and activity monitoring all within reach. The nematode C. elegans is\noften used as a model organism to study neuronal activity due to factors such\nas its transparency and well-understood nervous system. The principal barrier\nto high-accuracy neuron identification is that in adult C. elegans, the\nposition of neuronal cell bodies is not stereotyped. Existing approaches to\naddress this issue use genetically encoded markers as an additional identifying\nfeature. For example, the NeuroPAL strain uses multicolored fluorescent\nreporters. However, this approach has limited use due to the negative effects\nof excessive genetic modification. In this study, I propose an alternative\nneuronal identification technique using only single-color fluorescent images. I\ndesigned a novel neural network based classifier that automatically labels\nsensory neurons using an iterative, landmark-based neuron identification\nprocess inspired by the manual annotation procedures that humans employ. This\ndesign labels sensory neurons in C. elegans with 91.61% accuracy.\n", "date": "2022-10-24", "categories": []}
{"id": "2210.16587", "title": "Relating Human Perception of Musicality to Prediction in a Predictive\n  Coding Model", "abstract": "  We explore the use of a neural network inspired by predictive coding for\nmodeling human music perception. This network was developed based on the\ncomputational neuroscience theory of recurrent interactions in the hierarchical\nvisual cortex. When trained with video data using self-supervised learning, the\nmodel manifests behaviors consistent with human visual illusions. Here, we\nadapt this network to model the hierarchical auditory system and investigate\nwhether it will make similar choices to humans regarding the musicality of a\nset of random pitch sequences. When the model is trained with a large corpus of\ninstrumental classical music and popular melodies rendered as mel spectrograms,\nit exhibits greater prediction errors for random pitch sequences that are rated\nless musical by human subjects. We found that the prediction error depends on\nthe amount of information regarding the subsequent note, the pitch interval,\nand the temporal context. Our findings suggest that predictability is\ncorrelated with human perception of musicality and that a predictive coding\nneural network trained on music can be used to characterize the features and\nmotifs contributing to human perception of music.\n", "date": "2022-10-29", "categories": []}
{"id": "2210.16062", "title": "Neural Network based Formation of Cognitive Maps of Semantic Spaces and\n  the Emergence of Abstract Concepts", "abstract": "  The hippocampal-entorhinal complex plays a major role in the organization of\nmemory and thought. The formation of and navigation in cognitive maps of\narbitrary mental spaces via place and grid cells can serve as a representation\nof memories and experiences and their relations to each other. The multi-scale\nsuccessor representation is proposed to be the mathematical principle\nunderlying place and grid cell computations. Here, we present a neural network,\nwhich learns a cognitive map of a semantic space based on 32 different animal\nspecies encoded as feature vectors. The neural network successfully learns the\nsimilarities between different animal species, and constructs a cognitive map\nof 'animal space' based on the principle of successor representations with an\naccuracy of around 30% which is near to the theoretical maximum regarding the\nfact that all animal species have more than one possible successor, i.e.\nnearest neighbor in feature space. Furthermore, a hierarchical structure, i.e.\ndifferent scales of cognitive maps, can be modeled based on multi-scale\nsuccessor representations. We find that, in fine-grained cognitive maps, the\nanimal vectors are evenly distributed in feature space. In contrast, in\ncoarse-grained maps, animal vectors are highly clustered according to their\nbiological class, i.e. amphibians, mammals and insects. This could be a\npossible mechanism explaining the emergence of new abstract semantic concepts.\nFinally, even completely new or incomplete input can be represented by\ninterpolation of the representations from the cognitive map with remarkable\nhigh accuracy of up to 95%. We conclude that the successor representation can\nserve as a weighted pointer to past memories and experiences, and may therefore\nbe a crucial building block for future machine learning to include prior\nknowledge, and to derive context knowledge from novel input.\n", "date": "2022-10-28", "categories": []}
{"id": "2208.12025", "title": "Integrating Statistical and Machine Learning Approaches to Identify\n  Receptive Field Structure in Neural Populations", "abstract": "  Neurons can code for multiple variables simultaneously and neuroscientists\nare often interested in classifying neurons based on their receptive field\nproperties. Statistical models provide powerful tools for determining the\nfactors influencing neural spiking activity and classifying individual neurons.\nHowever, as neural recording technologies have advanced to produce simultaneous\nspiking data from massive populations, classical statistical methods often lack\nthe computational efficiency required to handle such data. Machine learning\n(ML) approaches are known for enabling efficient large scale data analyses;\nhowever, they typically require massive training sets with balanced data, along\nwith accurate labels to fit well. Additionally, model assessment and\ninterpretation are often more challenging for ML than for classical statistical\nmethods. To address these challenges, we develop an integrated framework,\ncombining statistical modeling and machine learning approaches to identify the\ncoding properties of neurons from large populations. In order to demonstrate\nthis framework, we apply these methods to data from a population of neurons\nrecorded from rat hippocampus to characterize the distribution of spatial\nreceptive fields in this region.\n", "date": "2022-07-25", "categories": []}
{"id": "2210.15044", "title": "Generative modeling of the enteric nervous system employing point\n  pattern analysis and graph construction", "abstract": "  We describe a generative network model of the architecture of the enteric\nnervous system (ENS) in the colon employing data from images of human and mouse\ntissue samples obtained through confocal microscopy. Our models combine spatial\npoint pattern analysis with graph generation to characterize the spatial and\ntopological properties of the ganglia (clusters of neurons and glial cells),\nthe inter-ganglionic connections, and the neuronal organization within the\nganglia. We employ a hybrid hardcore-Strauss process for spatial patterns and a\nplanar random graph generation for constructing the spatially embedded network.\nWe show that our generative model may be helpful in both basic and\ntranslational studies, and it is sufficiently expressive to model the ENS\narchitecture of individuals who vary in age and health status. Increased\nunderstanding of the ENS connectome will enable the use of neuromodulation\nstrategies in treatment and clarify anatomic diagnostic criteria for people\nwith bowel motility disorders.\n", "date": "2022-10-26", "categories": []}
{"id": "2210.12068", "title": "Grid cells and their potential application in AI", "abstract": "  Since their Nobel Prize winning discovery in 2005, grid cells have been\nstudied extensively by neuroscientists. Their multi-scale periodic firing rates\ntiling the environment as the animal moves around has been shown as critical\nfor path integration. Multiple experiments have shown that grid cells also fire\nfor other representations such as olfactory, attention mechanisms, imagined\nmovement, and concept organization potentially acting as a form of neural\nrecycling and showing the possible brain mechanism for cognitive maps that\nTolman envisioned in 1948. Grid cell integration into artificial neural\nnetworks may enable more robust, generalized, and smarter computers. In this\npaper we give an overview of grid cell research since their discovery, their\nrole in neuroscience and cognitive science, and possible future directions of\nartificial intelligence research.\n", "date": "2022-10-12", "categories": []}
{"id": "2210.03667", "title": "CommsVAE: Learning the brain's macroscale communication dynamics using\n  coupled sequential VAEs", "abstract": "  Communication within or between complex systems is commonplace in the natural\nsciences and fields such as graph neural networks. The brain is a perfect\nexample of such a complex system, where communication between brain regions is\nconstantly being orchestrated. To analyze communication, the brain is often\nsplit up into anatomical regions that each perform certain computations. These\nregions must interact and communicate with each other to perform tasks and\nsupport higher-level cognition. On a macroscale, these regions communicate\nthrough signal propagation along the cortex and along white matter tracts over\nlonger distances. When and what types of signals are communicated over time is\nan unsolved problem and is often studied using either functional or structural\ndata. In this paper, we propose a non-linear generative approach to\ncommunication from functional data. We address three issues with common\nconnectivity approaches by explicitly modeling the directionality of\ncommunication, finding communication at each timestep, and encouraging\nsparsity. To evaluate our model, we simulate temporal data that has sparse\ncommunication between nodes embedded in it and show that our model can uncover\nthe expected communication dynamics. Subsequently, we apply our model to\ntemporal neural data from multiple tasks and show that our approach models\ncommunication that is more specific to each task. The specificity of our method\nmeans it can have an impact on the understanding of psychiatric disorders,\nwhich are believed to be related to highly specific communication between brain\nregions compared to controls. In sum, we propose a general model for dynamic\ncommunication learning on graphs, and show its applicability to a subfield of\nthe natural sciences, with potential widespread scientific impact.\n", "date": "2022-10-07", "categories": []}
{"id": "2210.01767", "title": "Supervised Parameter Estimation of Neuron Populations from Multiple\n  Firing Events", "abstract": "  The firing dynamics of biological neurons in mathematical models is often\ndetermined by the model's parameters, representing the neurons' underlying\nproperties. The parameter estimation problem seeks to recover those parameters\nof a single neuron or a neuron population from their responses to external\nstimuli and interactions between themselves. Most common methods for tackling\nthis problem in the literature use some mechanistic models in conjunction with\neither a simulation-based or solution-based optimization scheme. In this paper,\nwe study an automatic approach of learning the parameters of neuron populations\nfrom a training set consisting of pairs of spiking series and parameter labels\nvia supervised learning. Unlike previous work, this automatic learning does not\nrequire additional simulations at inference time nor expert knowledge in\nderiving an analytical solution or in constructing some approximate models. We\nsimulate many neuronal populations with different parameter settings using a\nstochastic neuron model. Using that data, we train a variety of supervised\nmachine learning models, including convolutional and deep neural networks,\nrandom forest, and support vector regression. We then compare their performance\nagainst classical approaches including a genetic search, Bayesian sequential\nestimation, and a random walk approximate model. The supervised models almost\nalways outperform the classical methods in parameter estimation and spike\nreconstruction errors, and computation expense. Convolutional neural network,\nin particular, is the best among all models across all metrics. The supervised\nmodels can also generalize to out-of-distribution data to a certain extent.\n", "date": "2022-10-02", "categories": []}
{"id": "2210.01769", "title": "Mind Reader: Reconstructing complex images from brain activities", "abstract": "  Understanding how the brain encodes external stimuli and how these stimuli\ncan be decoded from the measured brain activities are long-standing and\nchallenging questions in neuroscience. In this paper, we focus on\nreconstructing the complex image stimuli from fMRI (functional magnetic\nresonance imaging) signals. Unlike previous works that reconstruct images with\nsingle objects or simple shapes, our work aims to reconstruct image stimuli\nthat are rich in semantics, closer to everyday scenes, and can reveal more\nperspectives. However, data scarcity of fMRI datasets is the main obstacle to\napplying state-of-the-art deep learning models to this problem. We find that\nincorporating an additional text modality is beneficial for the reconstruction\nproblem compared to directly translating brain signals to images. Therefore,\nthe modalities involved in our method are: (i) voxel-level fMRI signals, (ii)\nobserved images that trigger the brain signals, and (iii) textual description\nof the images. To further address data scarcity, we leverage an aligned\nvision-language latent space pre-trained on massive datasets. Instead of\ntraining models from scratch to find a latent space shared by the three\nmodalities, we encode fMRI signals into this pre-aligned latent space. Then,\nconditioned on embeddings in this space, we reconstruct images with a\ngenerative model. The reconstructed images from our pipeline balance both\nnaturalness and fidelity: they are photo-realistic and capture the ground truth\nimage contents well.\n", "date": "2022-09-30", "categories": []}
{"id": "2209.13561", "title": "Adapting Brain-Like Neural Networks for Modeling Cortical Visual\n  Prostheses", "abstract": "  Cortical prostheses are devices implanted in the visual cortex that attempt\nto restore lost vision by electrically stimulating neurons. Currently, the\nvision provided by these devices is limited, and accurately predicting the\nvisual percepts resulting from stimulation is an open challenge. We propose to\naddress this challenge by utilizing 'brain-like' convolutional neural networks\n(CNNs), which have emerged as promising models of the visual system. To\ninvestigate the feasibility of adapting brain-like CNNs for modeling visual\nprostheses, we developed a proof-of-concept model to predict the perceptions\nresulting from electrical stimulation. We show that a neurologically-inspired\ndecoding of CNN activations produces qualitatively accurate phosphenes,\ncomparable to phosphenes reported by real patients. Overall, this is an\nessential first step towards building brain-like models of electrical\nstimulation, which may not just improve the quality of vision provided by\ncortical prostheses but could also further our understanding of the neural code\nof vision.\n", "date": "2022-09-27", "categories": []}
{"id": "2209.10992", "title": "Modeling cognitive load as a self-supervised brain rate with\n  electroencephalography and deep learning", "abstract": "  The principal reason for measuring mental workload is to quantify the\ncognitive cost of performing tasks to predict human performance. Unfortunately,\na method for assessing mental workload that has general applicability does not\nexist yet. This research presents a novel self-supervised method for mental\nworkload modelling from EEG data employing Deep Learning and a continuous brain\nrate, an index of cognitive activation, without requiring human declarative\nknowledge. This method is a convolutional recurrent neural network trainable\nwith spatially preserving spectral topographic head-maps from EEG data to fit\nthe brain rate variable. Findings demonstrate the capacity of the convolutional\nlayers to learn meaningful high-level representations from EEG data since\nwithin-subject models had a test Mean Absolute Percentage Error average of 11%.\nThe addition of a Long-Short Term Memory layer for handling sequences of\nhigh-level representations was not significant, although it did improve their\naccuracy. Findings point to the existence of quasi-stable blocks of learnt\nhigh-level representations of cognitive activation because they can be induced\nthrough convolution and seem not to be dependent on each other over time,\nintuitively matching the non-stationary nature of brain responses.\nAcross-subject models, induced with data from an increasing number of\nparticipants, thus containing more variability, obtained a similar accuracy to\nthe within-subject models. This highlights the potential generalisability of\nthe induced high-level representations across people, suggesting the existence\nof subject-independent cognitive activation patterns. This research contributes\nto the body of knowledge by providing scholars with a novel computational\nmethod for mental workload modelling that aims to be generally applicable, does\nnot rely on ad-hoc human-crafted models supporting replicability and\nfalsifiability.\n", "date": "2022-09-21", "categories": []}
{"id": "2209.08572", "title": "A Computational Model of Learning Flexible Navigation in a Maze by\n  Layout-Conforming Replay of Place Cells", "abstract": "  Recent experimental observations have shown that the reactivation of\nhippocampal place cells (PC) during sleep or immobility depicts trajectories\nthat can go around barriers and can flexibly adapt to a changing maze layout.\nSuch layout-conforming replay sheds a light on how the activity of place cells\nsupports the learning of flexible navigation of an animal in a dynamically\nchanging maze. However, existing computational models of replay fall short of\ngenerating layout-conforming replay, restricting their usage to simple\nenvironments, like linear tracks or open fields. In this paper, we propose a\ncomputational model that generates layout-conforming replay and explains how\nsuch replay drives the learning of flexible navigation in a maze. First, we\npropose a Hebbian-like rule to learn the inter-PC synaptic strength during\nexploring a maze. Then we use a continuous attractor network (CAN) with\nfeedback inhibition to model the interaction among place cells and hippocampal\ninterneurons. The activity bump of place cells drifts along a path in the maze,\nwhich models layout-conforming replay. During replay in rest, the synaptic\nstrengths from place cells to striatal medium spiny neurons (MSN) are learned\nby a novel dopamine-modulated three-factor rule to store place-reward\nassociations. During goal-directed navigation, the CAN periodically generates\nreplay trajectories from the animal's location for path planning, and the\ntrajectory leading to a maximal MSN activity is followed by the animal. We have\nimplemented our model into a high-fidelity virtual rat in the MuJoCo physics\nsimulator. Extensive experiments have demonstrated that its superior\nflexibility during navigation in a maze is due to a continuous re-learning of\ninter-PC and PC-MSN synaptic strength.\n", "date": "2022-09-18", "categories": []}
{"id": "2209.02387", "title": "MARTI-4: new model of human brain, considering neocortex and basal\n  ganglia -- learns to play Atari game by reinforcement learning on a single\n  CPU", "abstract": "  We present Deep Control - new ML architecture of cortico-striatal brain\ncircuits, which use whole cortical column as a structural element, instead of a\nsinge neuron. Based on this architecture, we present MARTI - new model of human\nbrain, considering neocortex and basal ganglia. This model is de-signed to\nimplement expedient behavior and is capable to learn and achieve goals in\nunknown environments. We introduce a novel surprise feeling mechanism, that\nsignificantly improves reinforcement learning process through inner rewards. We\nuse OpenAI Gym environment to demonstrate MARTI learning on a single CPU just\nin several hours.\n", "date": "2022-08-18", "categories": []}
{"id": "2208.12032", "title": "How (and Why) to Think that the Brain is Literally a Computer", "abstract": "  The relationship between brains and computers is often taken to be merely\nmetaphorical. However, genuine computational systems can be implemented in\nvirtually any media; thus, one can take seriously the view that brains\nliterally compute. But without empirical criteria for what makes a physical\nsystem genuinely a computational one, computation remains a matter of\nperspective, especially for natural systems (e.g., brains) that were not\nexplicitly designed and engineered to be computers. Considerations from real\nexamples of physical computers-both analog and digital, contemporary and\nhistorical-make clear what those empirical criteria must be. Finally, applying\nthose criteria to the brain shows how we can view the brain as a computer\n(probably an analog one at that), which, in turn, illuminates how that claim is\nboth informative and falsifiable.\n", "date": "2022-08-24", "categories": []}
{"id": "2208.03666", "title": "See What You See: Self-supervised Cross-modal Retrieval of Visual\n  Stimuli from Brain Activity", "abstract": "  Recent studies demonstrate the use of a two-stage supervised framework to\ngenerate images that depict human perception to visual stimuli from EEG,\nreferring to EEG-visual reconstruction. They are, however, unable to reproduce\nthe exact visual stimulus, since it is the human-specified annotation of\nimages, not their data, that determines what the synthesized images are.\nMoreover, synthesized images often suffer from noisy EEG encodings and unstable\ntraining of generative models, making them hard to recognize. Instead, we\npresent a single-stage EEG-visual retrieval paradigm where data of two\nmodalities are correlated, as opposed to their annotations, allowing us to\nrecover the exact visual stimulus for an EEG clip. We maximize the mutual\ninformation between the EEG encoding and associated visual stimulus through\noptimization of a contrastive self-supervised objective, leading to two\nadditional benefits. One, it enables EEG encodings to handle visual classes\nbeyond seen ones during training, since learning is not directed at class\nannotations. In addition, the model is no longer required to generate every\ndetail of the visual stimulus, but rather focuses on cross-modal alignment and\nretrieves images at the instance level, ensuring distinguishable model output.\nEmpirical studies are conducted on the largest single-subject EEG dataset that\nmeasures brain activities evoked by image stimuli. We demonstrate the proposed\napproach completes an instance-level EEG-visual retrieval task which existing\nmethods cannot. We also examine the implications of a range of EEG and visual\nencoder structures. Furthermore, for a mostly studied semantic-level EEG-visual\nclassification task, despite not using class annotations, the proposed method\noutperforms state-of-the-art supervised EEG-visual reconstruction approaches,\nparticularly on the capability of open class recognition.\n", "date": "2022-08-07", "categories": []}
{"id": "2208.02845", "title": "Decision SincNet: Neurocognitive models of decision making that predict\n  cognitive processes from neural signals", "abstract": "  Human decision making behavior is observed with choice-response time data\nduring psychological experiments. Drift-diffusion models of this data consist\nof a Wiener first-passage time (WFPT) distribution and are described by\ncognitive parameters: drift rate, boundary separation, and starting point.\nThese estimated parameters are of interest to neuroscientists as they can be\nmapped to features of cognitive processes of decision making (such as speed,\ncaution, and bias) and related to brain activity. The observed patterns of RT\nalso reflect the variability of cognitive processes from trial to trial\nmediated by neural dynamics. We adapted a SincNet-based shallow neural network\narchitecture to fit the Drift-Diffusion model using EEG signals on every\nexperimental trial. The model consists of a SincNet layer, a depthwise spatial\nconvolution layer, and two separate FC layers that predict drift rate and\nboundary for each trial in-parallel. The SincNet layer parametrized the kernels\nin order to directly learn the low and high cutoff frequencies of bandpass\nfilters that are applied to the EEG data to predict drift and boundary\nparameters. During training, model parameters were updated by minimizing the\nnegative log likelihood function of WFPT distribution given trial RT. We\ndeveloped separate decision SincNet models for each participant performing a\ntwo-alternative forced-choice task. Our results showed that single-trial\nestimates of drift and boundary performed better at predicting RTs than the\nmedian estimates in both training and test data sets, suggesting that our model\ncan successfully use EEG features to estimate meaningful single-trial Diffusion\nmodel parameters. Furthermore, the shallow SincNet architecture identified time\nwindows of information processing related to evidence accumulation and caution\nand the EEG frequency bands that reflect these processes within each\nparticipant.\n", "date": "2022-08-04", "categories": []}
{"id": "2208.06382", "title": "Layers, Folds, and Semi-Neuronal Information Processing", "abstract": "  What role does phenotypic complexity play in the systems-level function of an\nembodied agent? The organismal phenotype is a topologically complex structure\nthat interacts with a genotype, developmental physics, and an informational\nenvironment. Using this observation as inspiration, we utilize a type of\nembodied agent that exhibits layered representational capacity: meta-brain\nmodels. Meta-brains are used to demonstrate how phenotypes process information\nand exhibit self-regulation from development to maturity. We focus on two\ncandidate structures that potentially explain this capacity: folding and\nlayering. As layering and folding can be observed in a host of biological\ncontexts, they form the basis for our representational investigations. First,\nan innate starting point (genomic encoding) is described. The generative output\nof this encoding is a differentiation tree, which results in a layered\nphenotypic representation. Then we specify a formal meta-brain model of the\ngut, which exhibits folding and layering in development in addition to\ndifferent degrees of representation of processed information. This organ\ntopology is retained in maturity, with the potential for additional folding and\nrepresentational drift in response to inflammation. Next, we consider\ntopological remapping using the developmental Braitenberg Vehicle (dBV) as a\ntoy model. During topological remapping, it is shown that folding of a layered\nneural network can introduce a number of distortions to the original model,\nsome with functional implications. The paper concludes with a discussion on how\nthe meta-brains method can assist us in the investigation of enactivism,\nholism, and cognitive processing in the context of biological simulation.\n", "date": "2022-07-07", "categories": []}
{"id": "2208.04707", "title": "Context sequence theory: a common explanation for multiple types of\n  learning", "abstract": "  Although principles of neuroscience like reinforcement learning, visual\nperception and attention have been applied in machine learning models, there is\na huge gap between machine learning and mammalian learning. Based on the\nadvances in neuroscience, we propose the context sequence theory to give a\ncommon explanation for multiple types of learning in mammals and hope that can\nprovide a new insight into the construct of machine learning models.\n", "date": "2022-07-17", "categories": []}
{"id": "2208.00352", "title": "Neural Correlates of Face Familiarity Perception", "abstract": "  In the domain of face recognition, there exists a puzzling timing discrepancy\nbetween results from macaque neurophysiology on the one hand and human\nelectrophysiology on the other. Single unit recordings in macaques have\ndemonstrated face identity specific responses in extra-striate visual cortex\nwithin 100 milliseconds of stimulus onset. In EEG and MEG experiments with\nhumans, however, a consistent distinction between neural activity corresponding\nto unfamiliar and familiar faces has been reported to emerge around 250 ms.\nThis points to the possibility that there may be a hitherto undiscovered early\ncorrelate of face familiarity perception in human electrophysiological traces.\nWe report here a successful search for such a correlate in dense MEG recordings\nusing pattern classification techniques. Our analyses reveal markers of face\nfamiliarity as early as 85 ms after stimulus onset. Low-level attributes of the\nimages, such as luminance and color distributions, are unable to account for\nthis early emerging response difference. These results help reconcile human and\nmacaque data, and provide clues regarding neural mechanisms underlying familiar\nface perception.\n", "date": "2022-07-31", "categories": []}
{"id": "2207.14349", "title": "Bridging the Gap between Deep Learning and Hypothesis-Driven Analysis\n  via Permutation Testing", "abstract": "  A fundamental approach in neuroscience research is to test hypotheses based\non neuropsychological and behavioral measures, i.e., whether certain factors\n(e.g., related to life events) are associated with an outcome (e.g.,\ndepression). In recent years, deep learning has become a potential alternative\napproach for conducting such analyses by predicting an outcome from a\ncollection of factors and identifying the most \"informative\" ones driving the\nprediction. However, this approach has had limited impact as its findings are\nnot linked to statistical significance of factors supporting hypotheses. In\nthis article, we proposed a flexible and scalable approach based on the concept\nof permutation testing that integrates hypothesis testing into the data-driven\ndeep learning analysis. We apply our approach to the yearly self-reported\nassessments of 621 adolescent participants of the National Consortium of\nAlcohol and Neurodevelopment in Adolescence (NCANDA) to predict negative\nvalence, a symptom of major depressive disorder according to the NIMH Research\nDomain Criteria (RDoC). Our method successfully identifies categories of risk\nfactors that further explain the symptom.\n", "date": "2022-07-28", "categories": []}
{"id": "2207.14470", "title": "Significant changes in EEG neural oscillations during different phases\n  of three-dimensional multiple object tracking task (3D-MOT) imply different\n  roles for attention and working memory", "abstract": "  Our ability to track multiple objects in a dynamic environment enables us to\nperform everyday tasks such as driving, playing team sports, and walking in a\ncrowded mall. Despite more than three decades of literature on multiple object\ntracking (MOT) tasks, the underlying and intertwined neural mechanisms remain\npoorly understood. Here we looked at the electroencephalography (EEG) neural\ncorrelates and their changes across the three phases of a 3D-MOT task, namely\nidentification, tracking and recall. We recorded the EEG activity of 24\nparticipants while they were performing a 3D-MOT task with either 1, 2 or 3\ntargets where some trials were lateralized and some were not. We observed what\nseems to be a handoff between focused attention and working memory processes\nwhen going from tracking to recall. Our findings revealed a strong inhibition\nin delta and theta frequencies from the frontal region during tracking,\nfollowed by a strong (re)activation of these same frequencies during recall.\nOur results also showed contralateral delay activity (CDA) for the lateralized\ntrials, in both the identification and recall phases but not during tracking.\n", "date": "2022-07-29", "categories": []}
{"id": "2207.13845", "title": "EEG2Mel: Reconstructing Sound from Brain Responses to Music", "abstract": "  Information retrieval from brain responses to auditory and visual stimuli has\nshown success through classification of song names and image classes presented\nto participants while recording EEG signals. Information retrieval in the form\nof reconstructing auditory stimuli has also shown some success, but here we\nimprove on previous methods by reconstructing music stimuli well enough to be\nperceived and identified independently. Furthermore, deep learning models were\ntrained on time-aligned music stimuli spectrum for each corresponding\none-second window of EEG recording, which greatly reduces feature extraction\nsteps needed when compared to prior studies. The NMED-Tempo and NMED-Hindi\ndatasets of participants passively listening to full length songs were used to\ntrain and validate Convolutional Neural Network (CNN) regressors. The efficacy\nof raw voltage versus power spectrum inputs and linear versus mel spectrogram\noutputs were tested, and all inputs and outputs were converted into 2D images.\nThe quality of reconstructed spectrograms was assessed by training classifiers\nwhich showed 81% accuracy for mel-spectrograms and 72% for linear spectrograms\n(10% chance accuracy). Lastly, reconstructions of auditory music stimuli were\ndiscriminated by listeners at an 85% success rate (50% chance) in a\ntwo-alternative match-to-sample task.\n", "date": "2022-07-28", "categories": []}
{"id": "2207.08591", "title": "The Brain-Inspired Decoder for Natural Visual Image Reconstruction", "abstract": "  Decoding images from brain activity has been a challenge. Owing to the\ndevelopment of deep learning, there are available tools to solve this problem.\nThe decoded image, which aims to map neural spike trains to low-level visual\nfeatures and high-level semantic information space. Recently, there are a few\nstudies of decoding from spike trains, however, these studies pay less\nattention to the foundations of neuroscience and there are few studies that\nmerged receptive field into visual image reconstruction. In this paper, we\npropose a deep learning neural network architecture with biological properties\nto reconstruct visual image from spike trains. As far as we know, we\nimplemented a method that integrated receptive field property matrix into loss\nfunction at the first time. Our model is an end-to-end decoder from neural\nspike trains to images. We not only merged Gabor filter into auto-encoder which\nused to generate images but also proposed a loss function with receptive field\nproperties. We evaluated our decoder on two datasets which contain macaque\nprimary visual cortex neural spikes and salamander retina ganglion cells (RGCs)\nspikes. Our results show that our method can effectively combine receptive\nfield features to reconstruct images, providing a new approach to visual\nreconstruction based on neural information.\n", "date": "2022-07-18", "categories": []}
{"id": "2207.03901", "title": "Reproducing sensory induced hallucinations via neural fields", "abstract": "  Understanding sensory-induced cortical patterns in the primary visual cortex\nV1 is an important challenge both for physiological motivations and for\nimproving our understanding of human perception and visual organisation. In\nthis work, we focus on pattern formation in the visual cortex when the cortical\nactivity is driven by a geometric visual hallucination-like stimulus. In\nparticular, we present a theoretical framework for sensory-induced\nhallucinations which allows one to reproduce novel psychophysical results such\nas the MacKay effect (Nature, 1957) and the Billock and Tsou experiences (PNAS,\n2007).\n", "date": "2022-07-08", "categories": []}
{"id": "2207.03380", "title": "Neural Language Models are not Born Equal to Fit Brain Data, but\n  Training Helps", "abstract": "  Neural Language Models (NLMs) have made tremendous advances during the last\nyears, achieving impressive performance on various linguistic tasks.\nCapitalizing on this, studies in neuroscience have started to use NLMs to study\nneural activity in the human brain during language processing. However, many\nquestions remain unanswered regarding which factors determine the ability of a\nneural language model to capture brain activity (aka its 'brain score'). Here,\nwe make first steps in this direction and examine the impact of test loss,\ntraining corpus and model architecture (comparing GloVe, LSTM, GPT-2 and BERT),\non the prediction of functional Magnetic Resonance Imaging timecourses of\nparticipants listening to an audiobook. We find that (1) untrained versions of\neach model already explain significant amount of signal in the brain by\ncapturing similarity in brain responses across identical words, with the\nuntrained LSTM outperforming the transformerbased models, being less impacted\nby the effect of context; (2) that training NLP models improves brain scores in\nthe same brain regions irrespective of the model's architecture; (3) that\nPerplexity (test loss) is not a good predictor of brain score; (4) that\ntraining data have a strong influence on the outcome and, notably, that\noff-the-shelf models may lack statistical power to detect brain activations.\nOverall, we outline the impact of modeltraining choices, and suggest good\npractices for future studies aiming at explaining the human language system\nusing neural language models.\n", "date": "2022-07-07", "categories": []}
{"id": "2207.02457", "title": "Brain-inspired probabilistic generative model for double articulation\n  analysis of spoken language", "abstract": "  The human brain, among its several functions, analyzes the double\narticulation structure in spoken language, i.e., double articulation analysis\n(DAA). A hierarchical structure in which words are connected to form a sentence\nand words are composed of phonemes or syllables is called a double articulation\nstructure. Where and how DAA is performed in the human brain has not been\nestablished, although some insights have been obtained. In addition, existing\ncomputational models based on a probabilistic generative model (PGM) do not\nincorporate neuroscientific findings, and their consistency with the brain has\nnot been previously discussed. This study compared, mapped, and integrated\nthese existing computational models with neuroscientific findings to bridge\nthis gap, and the findings are relevant for future applications and further\nresearch. This study proposes a PGM for a DAA hypothesis that can be realized\nin the brain based on the outcomes of several neuroscientific surveys. The\nstudy involved (i) investigation and organization of anatomical structures\nrelated to spoken language processing, and (ii) design of a PGM that matches\nthe anatomy and functions of the region of interest. Therefore, this study\nprovides novel insights that will be foundational to further exploring DAA in\nthe brain.\n", "date": "2022-07-06", "categories": []}
{"id": "2207.00636", "title": "Action-modulated midbrain dopamine activity arises from distributed\n  control policies", "abstract": "  Animal behavior is driven by multiple brain regions working in parallel with\ndistinct control policies. We present a biologically plausible model of\noff-policy reinforcement learning in the basal ganglia, which enables learning\nin such an architecture. The model accounts for action-related modulation of\ndopamine activity that is not captured by previous models that implement\non-policy algorithms. In particular, the model predicts that dopamine activity\nsignals a combination of reward prediction error (as in classic models) and\n\"action surprise,\" a measure of how unexpected an action is relative to the\nbasal ganglia's current policy. In the presence of the action surprise term,\nthe model implements an approximate form of Q-learning. On benchmark navigation\nand reaching tasks, we show empirically that this model is capable of learning\nfrom data driven completely or in part by other policies (e.g. from other brain\nregions). By contrast, models without the action surprise term suffer in the\npresence of additional policies, and are incapable of learning at all from\nbehavior that is completely externally driven. The model provides a\ncomputational account for numerous experimental findings about dopamine\nactivity that cannot be explained by classic models of reinforcement learning\nin the basal ganglia. These include differing levels of action surprise signals\nin dorsal and ventral striatum, decreasing amounts movement-modulated dopamine\nactivity with practice, and representations of action initiation and kinematics\nin dopamine activity. It also provides further predictions that can be tested\nwith recordings of striatal dopamine activity.\n", "date": "2022-07-01", "categories": []}
{"id": "2207.00815", "title": "Simulating reaction time for Eureka effect in visual object recognition\n  using artificial neural network", "abstract": "  The human brain can recognize objects hidden in even severely degraded images\nafter observing them for a while, which is known as a type of Eureka effect,\npossibly associated with human creativity. A previous psychological study\nsuggests that the basis of this \"Eureka recognition\" is neural processes of\ncoincidence of multiple stochastic activities. Here we constructed an\nartificial-neural-network-based model that simulated the characteristics of the\nhuman Eureka recognition.\n", "date": "2022-06-30", "categories": []}
{"id": "2206.12503", "title": "Multi-Modal and Multi-Factor Branching Time Active Inference", "abstract": "  Active inference is a state-of-the-art framework for modelling the brain that\nexplains a wide range of mechanisms such as habit formation, dopaminergic\ndischarge and curiosity. Recently, two versions of branching time active\ninference (BTAI) based on Monte-Carlo tree search have been developed to handle\nthe exponential (space and time) complexity class that occurs when computing\nthe prior over all possible policies up to the time horizon. However, those two\nversions of BTAI still suffer from an exponential complexity class w.r.t the\nnumber of observed and latent variables being modelled. In the present paper,\nwe resolve this limitation by first allowing the modelling of several\nobservations, each of them having its own likelihood mapping. Similarly, we\nallow each latent state to have its own transition mapping. The inference\nalgorithm then exploits the factorisation of the likelihood and transition\nmappings to accelerate the computation of the posterior. Those two\noptimisations were tested on the dSprites environment in which the metadata of\nthe dSprites dataset was used as input to the model instead of the dSprites\nimages. On this task, $BTAI_{VMP}$ (Champion et al., 2022b,a) was able to solve\n96.9\\% of the task in 5.1 seconds, and $BTAI_{BF}$ (Champion et al., 2021a) was\nable to solve 98.6\\% of the task in 17.5 seconds. Our new approach\n($BTAI_{3MF}$) outperformed both of its predecessors by solving the task\ncompletly (100\\%) in only 2.559 seconds. Finally, $BTAI_{3MF}$ has been\nimplemented in a flexible and easy to use (python) package, and we developed a\ngraphical user interface to enable the inspection of the model's beliefs,\nplanning process and behaviour.\n", "date": "2022-06-24", "categories": []}
{"id": "2206.13217", "title": "Center-Embedding and Constituency in the Brain and a New\n  Characterization of Context-Free Languages", "abstract": "  A computational system implemented exclusively through the spiking of neurons\nwas recently shown capable of syntax, that is, of carrying out the dependency\nparsing of simple English sentences. We address two of the most important\nquestions left open by that work: constituency (the identification of key parts\nof the sentence such as the verb phrase) and the processing of dependent\nsentences, especially center-embedded ones. We show that these two aspects of\nlanguage can also be implemented by neurons and synapses in a way that is\ncompatible with what is known, or widely believed, about the structure and\nfunction of the language organ. Surprisingly, the way we implement center\nembedding points to a new characterization of context-free languages.\n", "date": "2022-06-27", "categories": []}
{"id": "2206.10821", "title": "Coupling Visual Semantics of Artificial Neural Networks and Human Brain\n  Function via Synchronized Activations", "abstract": "  Artificial neural networks (ANNs), originally inspired by biological neural\nnetworks (BNNs), have achieved remarkable successes in many tasks such as\nvisual representation learning. However, whether there exists semantic\ncorrelations/connections between the visual representations in ANNs and those\nin BNNs remains largely unexplored due to both the lack of an effective tool to\nlink and couple two different domains, and the lack of a general and effective\nframework of representing the visual semantics in BNNs such as human functional\nbrain networks (FBNs). To answer this question, we propose a novel\ncomputational framework, Synchronized Activations (Sync-ACT), to couple the\nvisual representation spaces and semantics between ANNs and BNNs in human brain\nbased on naturalistic functional magnetic resonance imaging (nfMRI) data. With\nthis approach, we are able to semantically annotate the neurons in ANNs with\nbiologically meaningful description derived from human brain imaging for the\nfirst time. We evaluated the Sync-ACT framework on two publicly available\nmovie-watching nfMRI datasets. The experiments demonstrate a) the significant\ncorrelation and similarity of the semantics between the visual representations\nin FBNs and those in a variety of convolutional neural networks (CNNs) models;\nb) the close relationship between CNN's visual representation similarity to\nBNNs and its performance in image classification tasks. Overall, our study\nintroduces a general and effective paradigm to couple the ANNs and BNNs and\nprovides novel insights for future studies such as brain-inspired artificial\nintelligence.\n", "date": "2022-06-22", "categories": []}
{"id": "2206.11228", "title": "Adversarially trained neural representations may already be as robust as\n  corresponding biological neural representations", "abstract": "  Visual systems of primates are the gold standard of robust perception. There\nis thus a general belief that mimicking the neural representations that\nunderlie those systems will yield artificial visual systems that are\nadversarially robust. In this work, we develop a method for performing\nadversarial visual attacks directly on primate brain activity. We then leverage\nthis method to demonstrate that the above-mentioned belief might not be well\nfounded. Specifically, we report that the biological neurons that make up\nvisual systems of primates exhibit susceptibility to adversarial perturbations\nthat is comparable in magnitude to existing (robustly trained) artificial\nneural networks.\n", "date": "2022-06-19", "categories": []}
{"id": "2206.08666", "title": "The Sensorium competition on predicting large-scale mouse primary visual\n  cortex activity", "abstract": "  The neural underpinning of the biological visual system is challenging to\nstudy experimentally, in particular as the neuronal activity becomes\nincreasingly nonlinear with respect to visual input. Artificial neural networks\n(ANNs) can serve a variety of goals for improving our understanding of this\ncomplex system, not only serving as predictive digital twins of sensory cortex\nfor novel hypothesis generation in silico, but also incorporating bio-inspired\narchitectural motifs to progressively bridge the gap between biological and\nmachine vision. The mouse has recently emerged as a popular model system to\nstudy visual information processing, but no standardized large-scale benchmark\nto identify state-of-the-art models of the mouse visual system has been\nestablished. To fill this gap, we propose the Sensorium benchmark competition.\nWe collected a large-scale dataset from mouse primary visual cortex containing\nthe responses of more than 28,000 neurons across seven mice stimulated with\nthousands of natural images, together with simultaneous behavioral measurements\nthat include running speed, pupil dilation, and eye movements. The benchmark\nchallenge will rank models based on predictive performance for neuronal\nresponses on a held-out test set, and includes two tracks for model input\nlimited to either stimulus only (Sensorium) or stimulus plus behavior\n(Sensorium+). We provide a starting kit to lower the barrier for entry,\nincluding tutorials, pre-trained baseline models, and APIs with one line\ncommands for data loading and submission. We would like to see this as a\nstarting point for regular challenges and data releases, and as a standard tool\nfor measuring progress in large-scale neural system identification models of\nthe mouse visual system and beyond.\n", "date": "2022-06-17", "categories": []}
{"id": "2206.01930", "title": "Investigating Brain Connectivity with Graph Neural Networks and\n  GNNExplainer", "abstract": "  Functional connectivity plays an essential role in modern neuroscience. The\nmodality sheds light on the brain's functional and structural aspects,\nincluding mechanisms behind multiple pathologies. One such pathology is\nschizophrenia which is often followed by auditory verbal hallucinations. The\nlatter is commonly studied by observing functional connectivity during speech\nprocessing. In this work, we have made a step toward an in-depth examination of\nfunctional connectivity during a dichotic listening task via deep learning for\nthree groups of people: schizophrenia patients with and without auditory verbal\nhallucinations and healthy controls. We propose a graph neural network-based\nframework within which we represent EEG data as signals in the graph domain.\nThe framework allows one to 1) predict a brain mental disorder based on EEG\nrecording, 2) differentiate the listening state from the resting state for each\ngroup and 3) recognize characteristic task-depending connectivity. Experimental\nresults show that the proposed model can differentiate between the above groups\nwith state-of-the-art performance. Besides, it provides a researcher with\nmeaningful information regarding each group's functional connectivity, which we\nvalidated on the current domain knowledge.\n", "date": "2022-06-04", "categories": []}
{"id": "2606.21940", "title": "DevoTG: Temporal Graph Neural Networks for Modeling C. elegans Developmental Connectomics", "abstract": "Understanding how a nervous system wires itself from birth to adulthood is a fundamental challenge in developmental neuroscience. We present DevoTG, a temporal graph framework that applies Temporal Graph Neural Networks (TGNs) to two complementary representations of C. elegans neural development: a Continuous-Time Dynamic Graph (CTDG) of cell division events derived from cell lineage data, and a Discrete-Time Dynamic Graph (DTDG) of the developing synaptic connectome spanning eight reconstructed electron-microscopy datasets. On the lineage prediction task, our TGN achieves a mean test AUC of 0.839 +/- 0.007 (5 seeds; validation AUC 0.937 +/- 0.001), outperforming a static GNN with the identical architecture by 26 AUC points (0.577 +/- 0.080), demonstrating that temporal memory is the decisive factor. Applied to the connectome DTDG, DevoTG identifies three connection stability classes (stable, developmental, and variable) across 225 neurons and 858 to 2,496 connections over development (L1 birth to adult), providing a temporal-graph-theoretic complement to the individual-variability classification of Witvliet et al. Analysis of hub command interneurons AVA, AVB, and AVE reveals their persistent centrality and how their integration roles are progressively reinforced across larval stages. Accompanying interactive visualizations (3D animated networks, centrality heatmaps, and a spatiotemporal lineage graph) make developmental dynamics accessible for biological hypothesis generation. DevoTG is open-source and designed for extension to other developing nervous systems. Code is publicly available at https://github.com/DevoLearn/DevoGraph/tree/main/DevoTG.", "date": "2026-06-20", "categories": []}
{"id": "2606.26279", "title": "Beyond Single-Source Cognitive Taskonomy:Multi-Source Task Relations through fMRI Transfer Learning", "abstract": "Cognitive tasks are organized by shared and specialized neural processes. Masked fMRI reconstruction provides a common self-supervised objective for quantifying transfer relations among task states, but existing reconstruction-based taskonomies mainly study one-to-one transfer from a single source task to a target. Here, we extend an fMRI cognitive taskonomy from single-source to multi-source transfer across 23 Human Connectome Project task states and use Boolean Integer Programming (BIP) to analyze budget-constrained task allocation. We train 1,127 task-specific and transfer models. Single-source transfer is directional and paradigm structured: motor states transfer well within the motor paradigm but provide limited support to most non-motor targets, consistent with a shared sensorimotor execution system and effector-specific representations. Multi-source transfer depends on the composition of the source set, suggesting that many-to-one task relations are not fully captured by pairwise taskonomy alone. Across supervision budgets, BIP repeatedly allocates direct supervision to several 0-back and 2-back working-memory states, although these states are not consistently the strongest individual sources. This pattern may reflect the integration of perceptual, attentional, and executive processes in working-memory tasks. Together, these findings reveal a cross-paradigm-limited motor cluster and working-memory states with high priority under the specified global allocation objective. Our study extends reconstruction-based fMRI taskonomy from one-to-one transfer to many-to-one task relations and budget-constrained task dependencies.", "date": "2026-06-24", "categories": []}
{"id": "2606.27783", "title": "CANNs: A Toolkit for Research on Continuous Attractor Neural Networks", "abstract": "Continuous attractor neural networks (CANNs) are the canonical computational framework for how the brain encodes continuous variables such as spatial position, head direction, and movement direction, and explain the activity of hippocampal place cells, entorhinal grid cells, and head-direction cells. CANN research, however, is fragmented: most results rest on lab-specific implementations, general-purpose simulators lack CANN-specific abstractions, and the path from spike trains to attractor geometry in real recordings lacks a standardized toolkit. Here, we present a comprehensive open-source toolkit that unifies the full CANN research workflow. It combines three tightly integrated components: 1) canns, a Python library on BrainPy/JAX that provides standardized 1D/2D CANNs, spike-frequency-adaptation variants, grid cell networks, hierarchical path-integration models, and brain-inspired attractor architectures, together with curated datasets, task generators, an analyzer module and trainer modules for biologically plausible plasticity; 2) canns-lib, a Rust acceleration backend delivering hundreds-of-times speedups for spatial-navigation workloads and modest gains for Ripser-based persistent homology; 3) ASA (Attractor Structure Analyzer), a PySide6 pipeline applying persistent homology and cohomology to experimental neural recordings to detect ring-like and toroidal attractor signatures in real data. The toolkit ships with full-detail reproducible pipelines that recover recent CANN results including SFA-driven anticipative tracking, theta sweeps in head-direction/place/grid systems, and hierarchical path integration.", "date": "2026-06-26", "categories": []}
{"id": "2606.30319", "title": "BrainJanus: A Unified Model for Understanding and Generation across Brain, Vision, and Language", "abstract": "Modeling the bidirectional correspondence between external sensory stimuli and internal neural activity has emerged as a critical frontier in neuroscience. However, existing approaches predominantly treat brain encoding and decoding as isolated tasks, relying heavily on unimodal alignment and external priors while overlooking the brain's intrinsic nature as a multimodal integration system. To address these limitations, we propose BrainJanus, the first unified brain model that integrates brain, vision, and language within a single framework. Specifically, we introduce a Unified Brain Tokenizer to quantize continuous neural dynamics into discrete tokens aligned with visual and linguistic representations in a shared Omni space. Building on this, we utilize an All-in-One autoregressive architecture that leverages next-token prediction to enable seamless any-to-any generation, which encompasses image-to-brain and text-to-brain encoding, and brain-to-image and brain-to-text decoding. Extensive experiments demonstrate that BrainJanus achieves superior performance across diverse benchmarks. Furthermore, our framework exhibits zero-shot generalization and preserves interpretable biological topography, highlighting its potential as a general-purpose brain modeling paradigm. The code is available at \\href{https://github.com/HaitaoWuTJU/BrainJanus}{GitHub}.", "date": "2026-06-29", "categories": []}
{"id": "2606.31819", "title": "Creating Intelligence: A Computational Foundation for AGI", "abstract": "This work introduces a new computational theory of mind grounded in set theory and hyperdimensional computing. Whereas traditional neural networks rely on continuous weights and matrix multiplication, this framework works with sparse binary data. It represents information as discrete sets, directly modeling biological neural population codes. I demonstrate that associative memory emerges naturally from network topologies featuring a combinatorially expanded hidden layer. Learning is driven by topological plasticity rather than scalar weight adjustments. This architecture unifies auto-associative and hetero-associative learning under a single core algorithm: information retrieval via subset pattern matching and exact nearest-neighbor search. Operating with constant-time complexity, these mechanisms bridge perceptual data (sparse distributed representations) and symbols (sparse holographic representations) without continuous bottlenecks. Mapping this framework to neuroanatomy, I propose that both the cerebellum and the neocortex implement variants of this algorithm, making subset pattern matching the fundamental engine of cognition. Because it relies on discrete logic rather than matrix arithmetic, this algorithm translates directly into in-memory hardware. This opens a new route toward synthetic intelligence with human-level energy efficiency.", "date": "2026-06-30", "categories": []}