{"id": "2605.21528", "title": "A Reproducible Log-Driven AutoML Framework for Interpretable Pipeline Optimization in Healthcare Risk Prediction", "abstract": "Accurate disease risk prediction is challenged by heterogeneous features, limited data, and class imbalance. This study presents yvsoucom-iterkit, a deterministic AutoML framework that models pipeline optimization as a configuration-level system with full reproducibility and traceable execution logs, enabling systematic analysis of component attribution, interactions, similarity, and cross-seed robustness.\n  Experiments on the Pima Indians Diabetes and Stroke datasets across more than 18,000 pipeline configurations reveal a structured yet partially redundant search space, where performance is dominated by a small subset of interacting components.\n  Ensemble models achieve stable performance, reaching a Weighted-F1 of 0.89 on Pima and 0.94 on Stroke. Macro-F1 reaches approximately 0.88 on Pima but drops to 0.6560 on Stroke due to severe imbalance. Cross-seed experiments show that ensembles reduce variance compared to single models. Friedman testing ($p < 0.05$) confirms significant ranking differences across configurations.\n  Based on analysis of component attribution, interaction, and similarity, optimal configuration design reveals dataset-dependent behavior. For the Pima dataset, computational efficiency benefits from simplified search spaces where redundant components can be removed, with split ratio playing a key role. In contrast, the Stroke dataset requires enhanced imbalance-aware strategies, where RandomOverSampler improves Macro-F1 from 0.6560 to 0.6766.\n  These findings demonstrate that effective AutoML optimization is achieved through optimal configuration design, where carefully constraining the search space to high-impact components can improve performance, stability, and interpretability while reducing unnecessary search complexity.", "date": "2026-05-19", "categories": []}
{"id": "2606.16000", "title": "GRACE-DS: a Guarded Reward-guided Agent Correction Environment in Data Science", "abstract": "We introduce GRACE-DS, a Guarded Reward-guided Agent Correction Environment in Data Science for pre-deployment evaluation of LLM-powered AutoML agents. GRACE-DS is a set of evaluation metrics in an isolated environment that can be applied to tabular ML tasks specific to a particular organization. It exposes agents to realistic workflow stages, from planning and data inspection through feature engineering, model development, validation, and code repair to final submission, while hidden executable validators measure not only final predictive performance but also leakage avoidance, reproducibility, protocol validity, correction behavior, and reward alignment. The strongest structured regime, flexible iterative interaction (our approach), achieves higher end-to-end normalized hidden-test quality than single-shot generation, unstructured interaction, and restart-based baselines, while also improving protocol-valid completion. Validated across more than 7,000 episodes, these results establish GRACE-DS as a robust platform for assessing the capacity of LLM-based AutoML agents to execute machine learning workflows under production-like conditions and in accordance with organization-specific requirements.", "date": "2026-06-14", "categories": []}
{"id": "2404.01965", "title": "Towards Leveraging AutoML for Sustainable Deep Learning: A Multi-Objective HPO Approach on Deep Shift Neural Networks", "abstract": "Deep Learning (DL) has advanced various fields by extracting complex patterns from large datasets. However, the computational demands of DL models pose environmental and resource challenges. Deep shift neural networks (DSNNs) offer a solution by leveraging shift operations to reduce computational complexity at inference. Following the insights from standard DNNs, we are interested in leveraging the full potential of DSNNs by means of AutoML techniques. We study the impact of hyperparameter optimization (HPO) to maximize DSNN performance while minimizing resource consumption. Since this combines multi-objective (MO) optimization with accuracy and energy consumption as potentially complementary objectives, we propose to combine state-of-the-art multi-fidelity (MF) HPO with multi-objective optimization. Experimental results demonstrate the effectiveness of our approach, resulting in models with over 80\\% in accuracy and low computational cost. Overall, our method accelerates efficient model development while enabling sustainable AI applications.", "date": "2024-04-02", "categories": []}
{"id": "2603.22376", "title": "Closing the Auto-Research Loop: An AI Co-Scientist for Production Search Ranking", "abstract": "We present an AI Co-Scientist framework that closes the research loop for the production search-ranking system of a large online travel platform -- pairing LLM agents with direct cloud-compute access so that idea generation, code implementation, GPU experimentation, and result analysis iterate end-to-end with a human scientist in the loop. The framework uses a hybrid agent architecture: single-LLM agents handle routine work, while multi-LLM consensus (GPT-5.2, Gemini Pro 3, Claude Opus 4.5) is invoked for higher-stakes decisions. On the production ranking task, a human-designed transformer baseline (V2) yielded $+0.118\\%$ over a pre-transformer baseline (V1); the AI Co-Scientist's automated loop on top of V2 contributed an additional $+0.083\\%$, for a combined $+0.201\\%$ offline gain delivered in roughly one extra week of wall-clock time (single-run numbers; statistical limits discussed in the paper). The most useful AI proposals -- unified long-sequence layouts, slot-type embeddings, and multi-phase learning-rate schedules -- are standard practice in NLP and Vision but were absent from our production stack, suggesting that LLM agents can serve as cross-disciplinary connectors for ranking teams. We also report deployment context, negative results, and lessons learned.", "date": "2026-03-23", "categories": []}
{"id": "2606.15497", "title": "Towards End-to-End Automation of AI Research", "abstract": "The automation of science is a long-standing ambition in the field of AI. While the community has made significant progress in automating individual components of the scientific process, a system that autonomously navigates the entire research lifecycle -- from conception to publication -- has remained out of reach. Here, we present the strongest demonstration to date toward automating the entire process end-to-end. We present The AI Scientist, which creates research ideas, writes code, runs experiments, plots and analyzes data, writes the entire scientific manuscript and performs its own peer review. Its ideas, execution, and presentation are of sufficient quality to produce a manuscript generated by an AI system that passes the first round of peer review at a major machine learning conference workshop. The workshop has an acceptance rate of 70 percent. Our system leverages modern foundation models within a complex agentic system. We evaluate The AI Scientist in two settings: a focused mode using human-provided code templates as an initial scaffold to conduct research on a specific topic, and a template-free, open-ended mode that leverages agentic search for wider scientific exploration. Both settings produce diverse ideas and automatically test, report on, and evaluate them. This achievement demonstrates AI's growing capacity for scientific contribution and signifies a potential paradigm shift in how research is conducted. As with any impactful new technology, there could be significant risks, including taxing overwhelmed review systems and adding noise to scientific literature. However, if developed responsibly, such autonomous systems could greatly accelerate scientific discovery.", "date": "2026-03-31", "categories": []}
{"id": "2601.08334", "title": "Automated Machine Learning in Radiomics: A Comparative Evaluation of Performance, Efficiency and Accessibility", "abstract": "Automated machine learning (AutoML) frameworks can lower technical barriers for predictive and prognostic model development in radiomics by enabling researchers without programming expertise to build models. However, their effectiveness in addressing radiomics-specific challenges remains unclear. This study evaluates the performance, efficiency, and accessibility of general-purpose and radiomics-specific AutoML frameworks on diverse radiomics classification tasks, thereby highlighting development needs for radiomics. Ten public/private radiomics datasets with varied imaging modalities (CT/MRI), sizes, anatomies and endpoints were used. Six general-purpose and five radiomics-specific frameworks were tested with predefined parameters using standardized cross-validation. Evaluation metrics included AUC, runtime, together with qualitative aspects related to software status, accessibility, and interpretability. Simplatab, a radiomics-specific tool with a no-code interface, achieved the highest average test AUC (81.81%) with a moderate runtime (~1 hour). LightAutoML, a general-purpose framework, showed the fastest execution with competitive performance (78.74% mean AUC in six minutes). Most radiomics-specific frameworks were excluded from the performance analysis due to obsolescence, extensive programming requirements, or computational inefficiency. Conversely, general-purpose frameworks demonstrated higher accessibility and ease of implementation. Simplatab provides an effective balance of performance, efficiency, and accessibility for radiomics classification problems. However, significant gaps remain, including the lack of accessible survival analysis support and the limited integration of feature reproducibility and harmonization within current AutoML frameworks. Future research should focus on adapting AutoML solutions to better address these radiomics-specific challenges.", "date": "2026-01-13", "categories": []}
{"id": "2606.05860", "title": "GenAutoML: An Agentic Framework for Dynamic Architecture Generation and Optimization in Time-Series Analysis", "abstract": "Designing neural architectures for time-series forecasting and anomaly detection remains a resource-intensive task that often requires substantial domain expertise. Traditional Automated Machine Learning (AutoML) systems typically rely on static, predefined search spaces, limiting their ability to adapt to diverse data characteristics. We present GenAutoML, an agentic framework that leverages Large Language Models (LLMs) as neural architects to bridge natural-language requirements and executable PyTorch implementations. The framework incorporates a Sandboxed Reflection Loop for autonomous code refinement and a Signature-Aware Runtime that enforces architectural consistency and execution safety. To improve robustness under non-stationary conditions, we further introduce a Dynamic Reversible Instance Normalization (Dyn-RevIN) wrapper. Experiments on the ETTh1, ETTm1, and Weather benchmarks demonstrate that GenAutoML can dynamically generate task-specific neural architectures tailored to dataset characteristics. Among the generated models, WaveInterferenceNet achieves inference latency below 0.01 ms per sample while maintaining competitive predictive performance. By emphasizing computational efficiency, architectural adaptability, and stable optimization behavior, GenAutoML enables the creation of ultra-lightweight neural networks suitable for resource-constrained and latency-sensitive Edge AI deployments.", "date": "2026-06-04", "categories": []}
{"id": "2606.12611", "title": "Evaluation of AutoML Frameworks for IDS under Imbalanced Data Conditions of the NSL-KDD Dataset", "abstract": "This work investigates the impact of severe class imbalance on the performance of automated machine learning (AutoML) frameworks for multiclass network intrusion detection using the NSL-KDD dataset. Unlike previous studies that simplify the problem through binary classification or minority-class removal, we preserve the original five-class distribution, including highly underrepresented attacks such as R2L and U2R, enabling a realistic evaluation of imbalance-sensitive learning behavior. Nine open-source AutoML frameworks were analyzed under a unified and reproducible experimental protocol, considering differences in architectural design, ensemble strategies, validation procedures, hyperparameter optimization, and imbalance-handling mechanisms. The results demonstrate that frameworks incorporating ensemble learning and imbalance-aware optimization achieve better minority-class discrimination. PyCaret obtained the best overall performance, reaching 66\\% macro-F1, followed by AutoGluon with 55\\%, whereas frameworks lacking native balancing support exhibited significant degradation in minority-class detection capability. The analysis further shows that accuracy-oriented optimization alone is insufficient for highly imbalanced IDS scenarios, since high-weighted metrics may coexist with poor generalization on rare attack categories. As a contribution, this work establishes a standardized benchmark for AutoML-based intrusion detection under severe multiclass imbalance, highlighting current architectural limitations and the need for native integration of imbalance-aware optimization, resampling, and stratified evaluation strategies into automated learning pipelines. The source code is publicly available.", "date": "2026-06-10", "categories": []}
{"id": "2512.08343", "title": "Soil Compaction Parameters Prediction Based on Automated Machine Learning Approach", "abstract": "Soil compaction is critical in construction engineering to ensure the stability of structures like road embankments and earth dams. Traditional methods for determining optimum moisture content (OMC) and maximum dry density (MDD) involve labor-intensive laboratory experiments, and empirical regression models have limited applicability and accuracy across diverse soil types. In recent years, artificial intelligence (AI) and machine learning (ML) techniques have emerged as alternatives for predicting these compaction parameters. However, ML models often struggle with prediction accuracy and generalizability, particularly with heterogeneous datasets representing various soil types. This study proposes an automated machine learning (AutoML) approach to predict OMC and MDD. AutoML automates algorithm selection and hyperparameter optimization, potentially improving accuracy and scalability. Through extensive experimentation, the study found that the Extreme Gradient Boosting (XGBoost) algorithm provided the best performance, achieving R-squared values of 80.4% for MDD and 89.1% for OMC on a separate dataset. These results demonstrate the effectiveness of AutoML in predicting compaction parameters across different soil types. The study also highlights the importance of heterogeneous datasets in improving the generalization and performance of ML models. Ultimately, this research contributes to more efficient and reliable construction practices by enhancing the prediction of soil compaction parameters.", "date": "2025-12-09", "categories": []}
{"id": "2606.08212", "title": "Public Machine Learning Solver Framework for Novices in the Machine Learning Domain", "abstract": "Solving machine learning problems is complex and typically reserved for experts. Over the past two decades, systems have emerged to support non-experts. Based on our review, we identify three categories: (1) fully automated AutoML systems, (2) expert cheat sheets for algorithm selection, and (3) decision-support systems using selection criteria (accuracy, transparency, data requirements). We propose a new platform combining categories 2 and 3 to deliver semi-automated, intelligent solution recommendations for non-experts. Unlike existing approaches that recommend a single algorithm, our platform suggests a complete pipeline tailored to the user's problem. It integrates expert-defined selection criteria with transfer learning and automatically extracts data characteristics (e.g., class imbalance, missing values) from user-provided datasets. The platform uses first-order logic to reason over its knowledge base and recommends suitable algorithms ranked by relevance. It features a user-friendly interface and connects to a crowdsourcing platform for ML experts, ensuring continuous updates. The platform is built incrementally, allowing seamless integration of new algorithms, criteria, and domain knowledge. To our knowledge, this is the first free, publicly accessible online framework that systematically captures and operationalizes expert knowledge to guide non-experts in solving ML problems in a structured, transparent manner.", "date": "2026-06-06", "categories": []}
{"id": "2606.06473", "title": "MLEvolve: A Self-Evolving Framework for Automated Machine Learning Algorithm Discovery", "abstract": "Large language model (LLM) agents are increasingly applied to long-horizon tasks such as scientific discovery and machine learning engineering (MLE), where sustained self-evolution becomes a key capability. However, existing MLE agents suffer from inter-branch information isolation, memoryless search, and lack of hierarchical control, which together hinder long-horizon optimization. We present MLEvolve, an LLM-based self-evolving multi-agent framework for end-to-end machine learning algorithm discovery. By extending tree search to Progressive MCGS, MLEvolve enables cross-branch information flow through graph-based reference edges and gradually shifts the search from broad exploration to focused exploitation with an entropy-inspired progressive schedule. To allow the agent to evolve with accumulated experience, we introduce Retrospective Memory, which combines a cold-start domain knowledge base with a dynamic global memory for task-specific experience retrieval and reuse. For stable long-horizon iteration, we further decouple strategic planning from code generation with adaptive coding modes. Evaluation on MLE-Bench shows that MLEvolve achieves state-of-the-art performance across multiple dimensions including average medal rate and valid submission rate under a 12-hour budget (half the standard runtime). Moreover, MLEvolve also outperforms specialized algorithm discovery methods including AlphaEvolve on mathematical algorithm optimization tasks, demonstrating strong cross-domain generalization. Our code is available at https://github.com/InternScience/MLEvolve.", "date": "2026-06-04", "categories": []}
{"id": "2509.25289", "title": "ClustRecNet: A Novel End-to-End Deep Learning Framework for Clustering Algorithm Recommendation", "abstract": "Identifying an effective clustering algorithm for a given dataset remains a fundamental unsupervised learning issue. We introduce ClustRecNet, a novel end-to-end deep learning framework that recommends suitable clustering algorithm(s) by directly learning high-order representations of raw tabular data. To facilitate robust meta-learning, we first construct a comprehensive repository of 34,000 synthetic datasets encompassing a large variety of clustering scenarios, run 10 popular clustering algorithms, and use Adjusted Rand Index (ARI) to establish ground-truth labels. ClustRecNet's architecture incorporates a convolution block, two residual blocks, and an attention block to capture local and global structural patterns, effectively bypassing the knowledge bottleneck associated with manual feature engineering. Extensive evaluation on both synthetic and real-world benchmarks demonstrates that ClustRecNet consistently outperforms traditional internal cluster validity indices such as Silhouette, Calinski-Harabasz, Davies-Bouldin, and Dunn as well as state-of-the-art Automated Machine Learning (AutoML) approaches such as ML2DAC, AutoCluster, and AutoML4Clust. For example, our framework achieves an average 0.497 ARI gain over the Calinski-Harabasz cluster validity index on synthetic data and an average 44.16% ARI improvement over the leading AutoML approach (ML2DAC) on real-world benchmarks. Code and data are available at: https://github.com/mrbakhtyari/ClustRecNet", "date": "2025-09-29", "categories": []}
{"id": "2601.20800", "title": "Conditional PED-ANOVA: Hyperparameter Importance in Hierarchical & Dynamic Search Spaces", "abstract": "We propose conditional PED-ANOVA (condPED-ANOVA), a principled framework for estimating hyperparameter importance (HPI) in conditional search spaces, where the presence or domain of a hyperparameter can depend on other hyperparameters. Although the original PED-ANOVA provides a fast and efficient way to estimate HPI within the top-performing regions of the search space, it assumes a fixed, unconditional search space and therefore cannot properly handle conditional hyperparameters. To address this, we introduce a conditional HPI for top-performing regions and derive a closed-form estimator that accurately reflects conditional activation and domain changes. Experiments show that naive adaptations of existing HPI estimators yield misleading or uninterpretable importances in conditional settings, whereas condPED-ANOVA consistently provides meaningful importances that reflect the underlying conditional structure. Our code is publicly available at https://github.com/kAIto47802/condPED-ANOVA.", "date": "2026-01-28", "categories": []}
{"id": "2603.23420", "title": "Bilevel Autoresearch: Meta-Autoresearching Itself", "abstract": "If autoresearch is itself a form of research, then autoresearch can be applied to research itself. We present Bilevel Autoresearch, a bilevel framework in which an outer autoresearch loop improves an inner autoresearch loop by reading its code and traces, identifying bottlenecks, and generating injectable Python search mechanisms at runtime. The inner loop optimizes task performance; the outer loop optimizes how the inner loop searches. Both loops use the same LLM, so improvements come from the bilevel architecture rather than a stronger meta-level model, although the outer loop consumes additional inference and wall-clock budget.\n  On Karpathy's GPT pretraining benchmark, the meta-autoresearch outer loop achieves a 5x improvement over the standard inner loop alone (-0.045 vs. -0.009 val_bpb), while parameter-level adjustment without mechanism change yields no reliable gain. The outer loop instantiates mechanisms from adjacent search domains, including combinatorial optimization, multi-armed bandits, and design of experiments, without human specification of the final mechanism design. Trace analysis suggests that these mechanisms break deterministic search patterns and force exploration of directions the LLM's priors avoid.\n  The experiments demonstrate, on this benchmark, a first bilevel step: an outer loop improves the search behavior of an inner loop. Code is the mechanism carrier in this implementation, but skills, prompts, workflows, evaluators, domain principles, world-model assumptions, and memory schemas can also encode mechanisms that shape future agent behavior. This suggests a path toward recursive bootstrapping, where mechanisms discovered for the inner loop can be fed back to improve the meta-level loop itself.", "date": "2026-03-24", "categories": []}
{"id": "2606.04261", "title": "Can Generalist Agents Automate Data Curation?", "abstract": "Curating training data is among the most consequential yet labor-intensive parts of modern AI development: practitioners iteratively propose, implement, evaluate, and revise data policies against noisy benchmark feedback. We ask whether generalist coding agents can automate this data-curation loop. We introduce *Curation-Bench*, an agent-centric benchmark that fixes the model, training recipe, and evaluation suite while giving agents command-line access to inspect data, implement policies, submit them to a fixed training/evaluation pipeline, and revise. In a vision-language instruction-tuning instantiation, out-of-the-box agents reach strong published data-selection baselines within ten iterations. However, trajectory analysis reveals a persistent *execution-research gap*: agents mainly tune local policy variants rather than explore new policy families, even when given strategy guides and paper references. Scaffolds requiring each iteration to cite, instantiate, and adapt a prior method shift agents toward method-guided exploration. The scaffolded agent autonomously composes -- without human design input -- a data-selection policy that outperforms strong published baselines at one-tenth their data budget. Overall, current agents can run the curation loop, but reliable data research requires scaffolded method adaptation, not open-ended prompting alone. Code and benchmark are open-sourced.", "date": "2026-06-02", "categories": []}
{"id": "2606.04971", "title": "Be Fair! Can Machine Learning Engineering Agents Adhere to Fairness Constraints?", "abstract": "Machine learning engineering (MLE) agents promise to automate end-to-end ML pipeline development from raw data and natural language instructions, potentially making ML accessible to non-technical domain experts. However, in sensitive and regulated domains, this abstraction creates a responsibility gap: end-users may lack visibility into design choices that affect correctness, robustness, fairness, and regulatory compliance. We argue that existing benchmarks are insufficient to assess whether MLE agents can be safely applied in such settings. We propose desiderata for a responsibility-centered evaluation framework and conduct an exploratory study on melanoma classification, focusing on fairness across skin tones as a responsibility constraint. When evaluating two recent MLE agents, we find that agent-generated pipelines show high variance and consistently underperform manually designed baselines in both predictive quality and fairness, despite fairness-oriented prompts. These preliminary results suggest that further research is needed towards redesigning MLE agents to allow humans to guide the search process and reliably assess the compliance and quality of the generated ML pipelines.", "date": "2026-06-03", "categories": []}
{"id": "2606.05139", "title": "BBOmix: A Tabular Benchmark for Hyperparameter Optimization of Unsupervised Biological Representation Learning", "abstract": "The rapid advancement of high-throughput sequencing has led to large, high-dimensional omics datasets. Deep unsupervised learning architectures, particularly Autoencoders (AEs), are increasingly used for dimensionality reduction and representation learning in this domain. However, AEs are highly sensitive to architectural choices and hyperparameters, and unsupervised optimization typically relies on reconstruction loss, which may be a poor proxy for downstream utility. Exhaustive hyperparameter optimization (HPO) is computationally expensive, leading researchers to frequently rely on suboptimal default configurations. To democratize access to large-scale unsupervised HPO research, we introduce $\\textbf{BBOmix}$, the first open-source tabular benchmark for unsupervised representation learning on real-world biological data. Our benchmark includes 105,000 evaluations across four AE architectures and seven multi-omics modalities from the TCGA and SCHC datasets. We quantify the correlation between reconstruction loss and downstream task performance and provide an extensive evaluation of state-of-the-art single-fidelity, multi-fidelity, and transfer learning HPO methods, establishing a rigorous baseline for future research in unsupervised biological representation learning.", "date": "2026-06-03", "categories": []}
{"id": "2504.10552", "title": "LEMUR Neural Network Dataset: Towards Seamless AutoML", "abstract": "Neural networks are the backbone of modern artificial intelligence, but designing, evaluating, and comparing them remains labor-intensive. While numerous datasets exist for training, there are few standardized collections of the models themselves. We introduce LEMUR, an open-source dataset and framework that provides a large collection of PyTorch-based neural networks across tasks such as classification, segmentation, detection, and natural language processing. Each model follows a unified template, with configurations and results stored in a structured database to ensure consistency and reproducibility. LEMUR integrates automated hyperparameter optimization via Optuna, includes statistical analysis and visualization tools, and offers an API for seamless access to performance data. The framework is extensible, allowing researchers to add new models, datasets, or metrics without breaking compatibility. By standardizing implementations and unifying evaluation, LEMUR aims to accelerate AutoML research, enable fair benchmarking, and reduce barriers to large-scale neural network experimentation. To support adoption and collaboration, LEMUR and its plugins are released under the MIT license at: https://github.com/ABrain-One/nn-dataset https://github.com/ABrain-One/nn-plots https://github.com/ABrain-One/nn-vr", "date": "2025-04-14", "categories": []}
{"id": "2511.20333", "title": "NNGPT: Rethinking AutoML with Large Language Models", "abstract": "Building self-improving AI systems remains a fundamental challenge in the AI domain. We present NNGPT, an open-source framework that turns a large language model (LLM) into a self-improving AutoML engine for neural network development, primarily for computer vision. Unlike previous frameworks, NNGPT extends the dataset of neural networks by generating new models, enabling continuous fine-tuning of LLMs based on closed-loop system of generation, assessment, and self-improvement. It integrates within one unified workflow five synergistic LLM-based pipelines: zero-shot architecture synthesis, hyperparameter optimization (HPO), code-aware accuracy/early-stop prediction, retrieval-augmented synthesis of scope-closed PyTorch blocks (NN-RAG), and reinforcement learning. Built on the LEMUR dataset as an audited corpus with reproducible metrics, NNGPT emits from a single prompt and validates network architecture, preprocessing code, and hyperparameters, executes them end-to-end, and learns from result. The PyTorch adapter makes NNGPT framework-agnostic, enabling strong performance: NN-RAG achieves 73% executability on 1,289 targets, 3-shot prompting boosts accuracy on common datasets, and hash-based deduplication saves hundreds of runs. One-shot prediction matches search-based AutoML, reducing the need for numerous trials. HPO on LEMUR achieves RMSE 0.60, outperforming Optuna (0.64), while the code-aware predictor reaches RMSE 0.14 with Pearson r=0.78. The system has already generated over 5K validated models, proving NNGPT as an autonomous AutoML engine. Upon acceptance, the code, prompts, and checkpoints will be released for public access to enable reproducibility and facilitate community usage.", "date": "2025-11-25", "categories": []}
{"id": "2602.13937", "title": "iML: Executable, Problem-Grounded, and Broadly Exploratory Code-Driven AutoML", "abstract": "Automated Machine Learning (AutoML) has improved access to machine learning, yet existing techniques often remain limited in flexibility, transparency, and execution reliability. Code-driven AutoML offers a promising direction by synthesizing executable code for preprocessing, model training, and evaluation. However, current LLM-based approaches frequently generate code that is plausible in text yet brittle in execution, insufficiently grounded in the actual dataset, or restricted to narrow solution paths. In this paper, we introduce iML, a multi-agent code-driven AutoML framework designed around three requirements: executability, problem grounding, and broad exploration of valid solutions. iML first analyzes the task and profiles the data, then synthesizes a structured blueprint that guides modular code generation across multiple implementation tracks, including traditional ML,pretrained adaptation, and custom neural architectures. To improve reliability, iML enforces interface checking, dynamic execution, and iterative debugging during integration. We evaluate iML on MLE-BENCH and the newly introduced iML-BENCH, covering diverse Kaggle-style tasks. On MLE-BENCH, iML attains a 90% valid submission rate and a 45% medal rate, and an APS of 0.82, improving the average standardized performance score (APS) over the LLM-based baselines by 52%-273%. On iML-BENCH, it achieves the highest APS and demonstrates robust performance even when task descriptions are substantially stripped. These results establish iML as a reliable and competitive framework for code-driven AutoML.", "date": "2026-02-15", "categories": []}
{"id": "2605.11359", "title": "CVEvolve: Autonomous Algorithm Discovery for Unstructured Scientific Data Processing", "abstract": "Scientific data processing often requires task-specific algorithms or AI models, creating a barrier for domain scientists who need to analyze their data but may not have extensive computing or image-processing expertise. This barrier is especially pronounced when data are noisy, have a high dynamic range, are sparsely labeled, or are only loosely specified. We introduce CVEvolve, an autonomous agentic harness with a zero-code interface for scientific data-processing algorithm discovery. CVEvolve combines a multi-round search strategy with tools for code execution, evaluation implementation, history management, holdout testing, and optional inspection of scientific data and visual outputs. The search alternates between discovery and improvement actions, and uses lineage-aware stochastic candidate sampling to balance exploration and exploitation. We demonstrate CVEvolve on X-ray fluorescence microscopy image registration, Bragg peak detection, high-energy diffraction microscopy image segmentation, and hybrid analytical-learning-based affine registration. Across these tasks, CVEvolve discovers algorithms that improve over baseline methods, while holdout test tracking helps identify candidates that generalize better than later over-optimized alternatives. These results show that zero-code, autonomous LLM-powered algorithm development can help domain scientists turn unstructured scientific image data into practical algorithms and downstream scientific discoveries.", "date": "2026-05-12", "categories": []}
{"id": "2606.00708", "title": "MOSAIC: Modular Orchestration for Structured Agentic Intelligence and Composition", "abstract": "Automated data science is a structured model-selection problem. A solution must choose data transformations, feature representations, architecture, training procedure, evaluation protocol, and refinement strategy for a task. AutoML systems automate parts of this process, but typically search within predefined pipeline, model, and hyperparameter spaces. LLM-based agents offer greater flexibility through retrieval, code generation, and execution feedback, yet their modelling decisions are often unstructured, difficult to verify, and hard to reuse. We introduce \\textsc{MOSAIC} (Modular Orchestration for Structured Agentic Intelligence and Composition), a structured agentic framework for memory-grounded model selection and workflow construction. Given a task and dataset, \\textsc{MOSAIC} builds a semantic task profile, retrieves prior cases and source-code modules, and constructs a blueprint: an intermediate representation specifying selected modelling components, composition, interface constraints, and execution requirements. This blueprint turns model selection into a staged, context-grounded search and grounds LLM-based code generation in retrieved evidence rather than unconstrained synthesis. Candidate models are validated by execution and refined using diagnostic feedback, training traces, task metrics, and a failure-aware reinforcement learning policy. We instantiate \\textsc{MOSAIC} on financial time-series forecasting and generation, where models must satisfy predictive accuracy, distributional fidelity, execution reliability, and downstream financial criteria such as risk and tail behaviour. Experiments against AutoML and agentic baselines show that \\textsc{MOSAIC} improves task performance, execution success, and decision traceability, demonstrating the value of treating automated data science as structured, reusable, and execution-grounded model selection.", "date": "2026-05-30", "categories": []}
{"id": "2605.17373", "title": "FML-bench: A Controlled Study of AI Research Agent Strategies from the Perspective of Search Dynamics", "abstract": "AI research agents accelerate ML research by automating hypothesis generation, experimentation, and empirical refinement. Existing agent strategies range from greedy hill-climbing to tree search and evolutionary optimization, yet which strategy choices drive performance remains unclear. Answering this question requires a benchmark that separates agent strategy (e.g., search topology) from execution infrastructure (e.g., code editor), so that performance differences are attributable to strategy rather than infrastructure, and that provides process-level metrics beyond final scores to analyze exploration behaviors. Existing benchmarks offer limited support. We propose FML-Bench, a benchmark of 18 fundamental ML research tasks across 10 domains that separates agent strategy from execution infrastructure and defines 12 process-level behavioral metrics. Evaluating six representative agents, we find that: (1) strategy complexity alone does not guarantee strong performance: a simple greedy hill-climber nearly matches the best-performing tree-search agent, both well above the remaining agents; (2) our analysis suggests this pattern relates to improvement opportunity structure: greedy search tends to be more effective when opportunities are dense, while tree-search and evolutionary strategies tend to be more effective when opportunities are sparse; an adaptive agent built on this insight switches to broader exploration upon detecting improvement stagnation and outperforms the other six agents, lending initial support to this observation; and (3) process-level analysis reveals that early convergence and directionally focused exploration are significantly associated with final performance, while solution diversity and compute cost are not. Our benchmark is available at: https://github.com/qrzou/FML-bench.", "date": "2026-05-17", "categories": []}
{"id": "2602.11171", "title": "A Language-Guided Bayesian Optimization for Efficient LoRA Hyperparameter Search", "abstract": "Fine-tuning Large Language Models (LLMs) with Low-Rank Adaptation (LoRA) offers a resource-efficient way to personalize or specialize. However, LoRA is highly sensitive to hyperparameter choices, and exhaustive hyperparameter search is computationally expensive. To address this, we propose a Bayesian Optimization (BO) framework that leverages the domain knowledge of pre-trained LLMs to efficiently search for LoRA hyperparameters. Our approach repurposes a pre-trained LLM as a discrete-to-continuous mapping module to link hyperparameters and their domain knowledge to a continuous vector space, where BO is conducted. We design and control the mapping via language prompting, providing a domain-aware textual prompt that describes the relationships among hyperparameters and their respective roles. This allows us to explicitly inject domain knowledge about LoRA into the LLM in natural language. We also introduce an additional learnable token to capture residual information that is difficult to describe linguistically in the prompt. This aids BO to sample more high-performing hyperparameters. In addition, by leveraging the strong correlation observed between the performance obtained from full and subset training datasets in LoRA training regimes, we introduce proxy training and evaluation using a data subset. This significantly improves the efficiency of our method. We demonstrate that our hyperparameter, discovered with only about 30 iterations, achieves more than 20% performance improvement over standard hyperparameters found from about 45,000 combinations. Project page: https://baekseongeun.github.io/lora-bo/", "date": "2026-01-19", "categories": []}
{"id": "2605.27873", "title": "AIBuildAI-2: A Knowledge-Enhanced Agent for Automatically Building AI Models", "abstract": "AI models underpin data-centric applications from image and text processing to scientific discovery in biology, physics, and chemistry. Yet developing them remains heavily manual, requiring practitioners to design architectures, build training pipelines, and iteratively refine solutions, making it challenging for natural scientists without specialized AI engineering expertise to build the high-performing models their research demands. To reduce this burden and broaden access to AI for scientific discovery, agents that automatically build AI models have been proposed. However, the performance of these agents is largely limited by the parametric knowledge of their underlying large language models, which is static, often outdated, and sparse on practical AI model engineering know-how. To address this limitation, we introduce AIBuildAI-2, a knowledge-enhanced agent with an external, evolving knowledge system for automatically building AI models. The knowledge system of AIBuildAI-2 is hierarchical, organizing curated AI development knowledge into high-level knowledge instructions over topical categories and low-level knowledge documents under each category, from which the agent dynamically loads only the context relevant to its current state and the AI task being solved, grounding each design and implementation decision in concrete, externally verifiable expertise. The system is initialized by collecting and cleaning AI-development-related documents from the web and organizing them into the corresponding categories, and continually evolves from the agent's own experience by distilling each completed run on an AI task into structured takeaways that are written back into the knowledge system. AIBuildAI-2 achieves state-of-the-art results, ranking first on MLE-Bench with a 70.7% medal rate and placing in the top 6.6% among 4,370 human-expert teams in a heart disease prediction competition.", "date": "2026-05-27", "categories": []}
{"id": "2502.03946", "title": "CleanSurvival: Automated data preprocessing for time-to-event models using reinforcement learning", "abstract": "Data preprocessing is often paid little attention in machine learning, despite its potentially significant impact on model performance. While automated machine learning pipelines are starting to recognize and integrate data preprocessing into their solutions for classification and regression tasks, this integration is lacking for more specialized tasks like time-to-event models for censored data. As a result, survival analysis not only faces the general challenges of data preprocessing but also suffers from the lack of tailored, automated solutions in this area. To address this gap, this paper presents CleanSurvival, a reinforcement-learning-based solution for optimizing preprocessing pipelines, extended specifically for survival analysis. The framework can handle continuous and categorical variables. It builds upon Learn2Clean's Q-learning to select which combination of data imputation, outlier detection and feature extraction techniques achieves optimal performance for a Cox, random forest, neural network or user-supplied time-to-event model. The Python package is available on GitHub: https://github.com/datasciapps/CleanSurvival. Experimental benchmarks on real-world datasets show that the Q-learning-based data preprocessing can improve predictive performance relative to simple baselines, while runtime behavior is condition-dependent and most clearly interpretable in the best-covered benchmark cells. Furthermore, a simulation study demonstrates effectiveness across different types and levels of missingness and noise. With an increase in the use of machine learning, it becomes important to generalise AutoML pipelines to a variety of models now present, including survival analysis. Tools like CleanSurvival, which integrate preprocessing for survival analysis, can make survival studies easier and quicker to perform, as well as make the results more robust.", "date": "2025-02-06", "categories": []}
{"id": "2604.05550", "title": "AutoSOTA: An End-to-End Automated Research System for State-of-the-Art AI Model Discovery", "abstract": "Artificial intelligence research increasingly depends on prolonged cycles of reproduction, debugging, and iterative refinement to achieve State-Of-The-Art (SOTA) performance, creating a growing need for systems that can accelerate the full pipeline of empirical model optimization. In this work, we introduce AutoSOTA, an end-to-end automated research system that advances the latest SOTA models published in top-tier AI papers to reproducible and empirically improved new SOTA models. We formulate this problem through three tightly coupled stages: resource preparation and goal setting; experiment evaluation; and reflection and ideation. To tackle this problem, AutoSOTA adopts a multi-agent architecture with eight specialized agents that collaboratively ground papers to code and dependencies, initialize and repair execution environments, track long-horizon experiments, generate and schedule optimization ideas, and supervise validity to avoid spurious gains. We evaluate AutoSOTA on recent research papers collected from eight top-tier AI conferences under filters for code availability and execution cost. Across these papers, AutoSOTA achieves strong end-to-end performance in both automated replication and subsequent optimization. Specifically, it successfully discovers 105 new SOTA models that surpass the original reported methods, averaging approximately five hours per paper. Case studies spanning LLM, NLP, computer vision, time series, and optimization further show that the system can move beyond routine hyperparameter tuning to identify architectural innovation, algorithmic redesigns, and workflow-level improvements. These results suggest that end-to-end research automation can serve not only as a performance optimizer, but also as a new form of research infrastructure that reduces repetitive experimental burden and helps redirect human attention toward higher-level scientific creativity.", "date": "2026-04-07", "categories": []}
{"id": "2602.13473", "title": "NeuroWeaver: An Autonomous Evolutionary Agent for Exploring the Programmatic Space of EEG Analysis Pipelines", "abstract": "Although foundation models have demonstrated remarkable success in general domains, the application of these models to electroencephalography (EEG) analysis is constrained by substantial data requirements and high parameterization. These factors incur prohibitive computational costs, thereby impeding deployment in resource-constrained clinical environments. Conversely, general-purpose automated machine learning frameworks are often ill-suited for this domain, as exploration within an unbounded programmatic space fails to incorporate essential neurophysiological priors and frequently yields solutions that lack scientific plausibility. To address these limitations, we propose NeuroWeaver, a unified autonomous evolutionary agent designed to generalize across diverse EEG datasets and tasks by reformulating pipeline engineering as a discrete constrained optimization problem. Specifically, we employ a Domain-Informed Subspace Initialization to confine the search to neuroscientifically plausible manifolds, coupled with a Multi-Objective Evolutionary Optimization that dynamically balances performance, novelty, and efficiency via self-reflective refinement. Empirical evaluations across five heterogeneous benchmarks demonstrate that NeuroWeaver synthesizes lightweight solutions that consistently outperform state-of-the-art task-specific methods and achieve performance comparable to large-scale foundation models, despite utilizing significantly fewer parameters.", "date": "2026-02-13", "categories": []}
{"id": "2602.09851", "title": "CoFEH: LLM-driven Feature Engineering Empowered by Collaborative Bayesian Hyperparameter Optimization", "abstract": "Feature Engineering (FE) is pivotal in automated machine learning (AutoML) but remains a bottleneck for traditional methods, which operate within rigid search spaces and lack domain awareness. While Large Language Models (LLMs) offer a promising alternative to generate unbounded operators with semantic reasoning, existing methods focus on isolated subtasks such as feature generation, falling short of free-form FE pipelines. Moreover, they are rarely coupled with hyperparameter optimization (HPO) of the downstream ML model, leading to greedy \"FE-then-HPO\" workflows that cannot capture strong FE-HPO interactions. In this paper, we present CoFEH, a collaborative framework that interleaves LLM-based FE and Bayesian HPO for robust end-to-end AutoML. CoFEH uses an LLM-driven FE optimizer powered by Tree of Thought (TOT) to explore flexible FE pipelines, a Bayesian optimization (BO) module to solve HPO, and a dynamic optimizer selector that adaptively interleaves FE and HPO steps. Crucially, we introduce a mutual conditioning mechanism that shares context between LLM and BO, enabling mutually informed decisions. Experiments show that CoFEH outperforms both traditional and LLM-based baselines in both standalone FE and joint FE+HPO settings.", "date": "2026-02-10", "categories": []}
{"id": "2602.02660", "title": "MARS: Modular Agent with Reflective Search for Automated AI Research", "abstract": "A critical bottleneck in automating AI research is the execution of complex machine learning engineering (MLE) tasks. MLE differs from general software engineering due to computationally expensive evaluation (e.g., model training) and opaque performance attribution. Current LLM-based agents struggle here, often generating monolithic scripts that ignore execution costs and causal factors. We introduce MARS (Modular Agent with Reflective Search), a framework optimized for autonomous AI research. MARS relies on three pillars: (1) Budget-Aware Planning via cost-constrained Monte Carlo Tree Search (MCTS) to explicitly balance performance with execution expense; (2) Modular Construction, employing a \"Design-Decompose-Implement\" pipeline to manage complex research repositories; and (3) Comparative Reflective Memory, which addresses credit assignment by analyzing solution differences to distill high-signal insights. MARS achieves state-of-the-art performance among open-source frameworks on MLE-Bench under comparable settings, maintaining competitiveness with the global leaderboard's top methods. Furthermore, the system exhibits qualitative \"Aha!\" moments, where 63% of all utilized lessons originate from cross-branch transfer, demonstrating that the agent effectively generalizes insights across search paths.", "date": "2026-02-02", "categories": []}
{"id": "2603.01712", "title": "FT-Dojo: Towards Autonomous LLM Fine-Tuning with Language Agents", "abstract": "Fine-tuning large language models for vertical domains remains labor-intensive, requiring practitioners to curate data, configure training, and iteratively diagnose model behavior. Despite growing interest in autonomous machine learning and language agents, end-to-end LLM fine-tuning has not been systematically studied as an interactive agent task. We introduce FT-Dojo, an interactive benchmark environment for autonomous LLM fine-tuning, comprising 13 tasks across 5 domains. Rather than a new collection of static datasets, FT-Dojo standardizes a task interface, shared raw-data repository, sandboxed execution environment, structured feedback protocol, and held-out evaluation procedure. We further develop FT-Agent, a fine-tuning-oriented autonomous framework that uses structured iteration planning, fail-fast validation, and multi-level feedback analysis to refine data and training strategies. Experiments show that FT-Agent provides a strong initial baseline, achieving the best performance on 10 out of 13 tasks, with additional controlled comparisons against frontier agents, open-source planning backbones, and multi-run statistics supporting the main findings. Case studies show that agents can recover from failures through cumulative learning, while still exposing limitations in causal diagnosis and long-horizon planning. The implementation is available at https://github.com/microsoft/rd-agent.", "date": "2026-03-02", "categories": []}
{"id": "2605.18764", "title": "From Intent to AI Pipelines: A Controlled Agentic Framework for Non-AI Expert Scientists", "abstract": "Artificial Intelligence (AI) pipelines have become integral to modern research, supporting fields such as Medical Sciences, Agriculture, and Social Sciences, and enabling large-scale data analysis, predictive modeling, and the automation of complex tasks. However, designing and implementing AI solutions remains challenging for many researchers due to the expertise required in the design and development of end-to-end AI systems. To address this gap, we present Domain-Driven Adaptable AI Pipelines (DDAP), a controlled, human-in-the-loop, agentic framework that leverages large language models to guide users in a systematic construction of AI pipelines and their corresponding implementation code. DDAP structures the development process into four stages: problem definition, compute environment specification, pipeline generation, and code generation. Through this staged interaction, the framework adapts to domain context, user expertise, and resource constraints, while maintaining user control over key decisions. We evaluate DDAP across multiple datasets spanning business, biology, and health science domains by comparing its AI models against expert-developed models. The experimental results show that DDAP achieves competitive results in several tasks compared to expert baselines, although performance varies across problem types, particularly for text-based clustering tasks. By combining guided interaction, adaptability, and reproducibility, DDAP demonstrates that a controlled agentic framework can generate competitive AI pipelines for non-expert users.", "date": "2026-04-10", "categories": []}
{"id": "2605.11518", "title": "AutoLLMResearch: Training Research Agents for Automating LLM Experiment Configuration - Learning from Cheap, Optimizing Expensive", "abstract": "Effectively configuring scalable large language model (LLM) experiments, spanning architecture design, hyperparameter tuning, and beyond, is crucial for advancing LLM research, as poor configuration choices can waste substantial computational resources and prevent models from realizing their full potential. Prior automated methods are designed for low-cost settings where repeated trial and error is feasible, but scalable LLM experiments are too expensive for such extensive iteration. To our knowledge, no work has addressed the automation of high-cost LLM experiment configurations, leaving this problem labor-intensive and dependent on expert intuition. Motivated by this gap, we propose AutoLLMResearch, an agentic framework that mimics how human researchers learn generalizable principles from low-fidelity experiments and extrapolate to efficiently identify promising configurations in expensive LLM settings. The core challenge is how to enable an agent to learn, through interaction with a multi-fidelity experimental environment that captures the structure of the LLM configuration landscape. To achieve this, we propose a systematic framework with two key components: 1) LLMConfig-Gym, a multi-fidelity environment encompassing four critical LLM experiment tasks, supported by over one million GPU hours of verifiable experiment outcomes; 2) A structured training pipeline that formulates configuration research as a long-horizon Markov Decision Process and accordingly incentivizes cross-fidelity extrapolation reasoning. Extensive evaluation against diverse strong baselines on held-out experiments demonstrates the effectiveness, generalization, and interpretability of our framework, supporting its potential as a practical and general solution for scalable real-world LLM experiment automation.", "date": "2026-05-12", "categories": []}
{"id": "2605.15461", "title": "DrugSAGE:Self-evolving Agent Experience for Efficient State-of-the-Art Drug Discovery", "abstract": "Building state-of-the-art (SOTA) predictive models for drug discovery requires expensive search over tools, architectures, and training strategies. Current LLM-based agents can find SOTA solutions through extensive trial and error, but they do not retain the experience accumulated along the way and therefore pay the full search cost on every new task. We propose \\method (Self-evolving Agent Experience), a framework that accumulates and reuses experience across tasks to build SOTA drug discovery models efficiently. \\method maintains a cross-task memory of verified skills, statistical evidence about effective strategies, and a record of recurring errors and their fixes. In some cases, \\method transfers a working solution directly without test-time search. In 33 molecular property prediction tasks, \\method ranks first among nine SOTA agents in a single-task setting. With memory accumulated from 16 smaller tasks, \\method achieves an averaged normalized score of 0.935 on 17 held-out tasks in a cross-task evaluation setting and outperforms all baseline agents by 10-30\\% in a zero-test-time search regime. In summary, our work shows the advantage of cross-task memory for efficient SOTA model development in drug discovery.", "date": "2026-05-14", "categories": []}
{"id": "2605.15871", "title": "Agentic Discovery of Neural Architectures: AIRA-Compose and AIRA-Design", "abstract": "Toward recursive self-improvement, we investigate LLM agents autonomously designing foundation models beyond standard Transformers. We introduce a dual-framework approach: AIRA-Compose for high-level architecture search, and AIRA-Design for low-level mechanistic implementation. AIRA-Compose uses 11 agents to explore fundamental computational primitives under a 24-hour budget. Agents evaluate million-parameter candidates, extrapolating top designs to 350M, 1B, and 3B scales. This yields 14 architectures across two families: AIRAformers (Transformer-based) and AIRAhybrids (Transformer-Mamba). Pre-trained at 1B scale, these consistently outperform Llama 3.2 and Composer-found baselines. On downstream tasks, AIRAformer-D and AIRAhybrid-D improve accuracy by 2.4% and 3.8% over Llama 3.2. Furthermore, AIRA-Compose finds models with highly efficient scaling frontiers: AIRAformer-C scales 54% and 71% faster than Llama 3.2 and Composer's best Transformer, while AIRAhybrid-C outscales Nemotron-2 by 23% and Composer's best hybrid by 37%. AIRA-Design tasks 20 agents with writing novel attention mechanisms for long-range dependencies and high-performing training scripts. On the Long Range Arena benchmark, agent-designed architectures reach within 2.3% and 2.6% of human state-of-the-art on document matching and text classification. On the Autoresearch benchmark, Greedy Opus 4.5 achieves 0.968 validation bits-per-byte under a fixed time budget, surpassing the published minimum. Together, these frameworks show AI agents can autonomously discover architectures and algorithmic optimizations matching or surpassing hand-designed baselines. This establishes a powerful paradigm for discovering next-generation foundation models, marking a clear step toward recursive self-improvement.", "date": "2026-05-15", "categories": []}
{"id": "2503.14434", "title": "LLM-FE: Automated Feature Engineering for Tabular Data with LLMs as Evolutionary Optimizers", "abstract": "Automated feature engineering plays a critical role in improving predictive model performance for tabular learning tasks. Traditional automated feature engineering methods are limited by their reliance on pre-defined transformations within fixed, manually designed search spaces, often neglecting domain knowledge. Recent advances using Large Language Models (LLMs) have enabled the integration of domain knowledge into the feature engineering process. However, existing LLM-based approaches use direct prompting or rely solely on validation scores for feature selection, failing to leverage insights from prior feature discovery experiments or establish meaningful reasoning between feature generation and data-driven performance. To address these challenges, we propose LLM-FE, a novel framework that combines evolutionary search with the domain knowledge and reasoning capabilities of LLMs to automatically discover effective features for tabular learning tasks. LLM-FE formulates feature engineering as a program search problem, where LLMs propose new feature transformation programs iteratively, and data-driven feedback guides the search process. Our results demonstrate that LLM-FE consistently outperforms state-of-the-art baselines, significantly enhancing the performance of tabular prediction models across diverse classification and regression benchmarks. The code is available at: https://github.com/nikhilsab/LLMFE", "date": "2025-03-18", "categories": []}
{"id": "2605.08305", "title": "LLMSYS-HPOBench: Hyperparameter Optimization Benchmark Suite for Real-World LLM Systems", "abstract": "Large Language Model (LLM) systems have been the frontier of AI in many application domains, leading to new challenges and opportunities for hyperparameter optimization (HPO) for the AutoML community. However, this type of system exhibits an unprecedented compound space of hyperparameter configuration from both the AI and non-AI components; rich and nonlinear implications from the fidelity factors; and diverse costs of measuring hyperparameter configurations, none of which have been fully captured in existing benchmarks. This paper presents the first (live) benchmark suite and datasets for HPO of real-world LLM systems, dubbed LLMSYS-HPOBench, covering data related to the inference objective values of hyperparameter configurations profiled from running the LLM systems. Currently, LLMSYS-HPOBench contains 364,450 hyperparameter configurations with a dimensionality of 12-23, 3-5 dimensions of fidelity factor leading to 932 settings, 3-9 inference objective metrics, and 2-10 cost metrics, together with generated logs from measuring the LLM systems. What we seek to advocate is not only a revalidation of the existing HPO algorithms over the frontier LLM systems, but also to provide an evolving platform for the AutoML community to explore new directions of research in this regard. The benchmark suite has been made available at: https://github.com/ideas-labo/llmsys-hpobench", "date": "2026-05-08", "categories": []}
{"id": "2601.12355", "title": "Tree-Structured Synergy of Large Language Models and Bayesian Optimization for Efficient CASH", "abstract": "To lower the expertise barrier in machine learning, the AutoML community has focused on the CASH problem, which jointly automates algorithm selection and hyperparameter tuning. While traditional methods like Bayesian Optimization (BO) struggle with cold-start issues, Large Language Models (LLMs) can mitigate these through semantic priors. However, existing LLM-based optimizers generalize poorly to high-dimensional, structured CASH spaces. In this paper, we propose LB-MCTS, a trajectory-structured optimization framework that uses a Monte Carlo Tree Search tree as a shared state for algorithm selection, hyperparameter refinement, and BO-LLM proposer synergy. Within this shared state, BO provides algorithm-specific surrogate modeling for quantitative search, while the LLM exploits path-aware selective memory to generate semantic proposals and reflections. As the surrogate model improves, a reliability-aware proposer policy adaptively shifts from LLM-driven to BO-driven proposals within a unified search trajectory. Experiments on 104 AMLB datasets demonstrate that LB-MCTS consistently outperforms BO-based, LLM-based, and hybrid baselines.", "date": "2026-01-18", "categories": []}
{"id": "2601.20375", "title": "LLM-AutoDP: Automatic Data Processing via LLM Agents for Model Fine-tuning", "abstract": "Large Language Models (LLMs) can be fine-tuned on domain-specific data to enhance their performance in specialized fields. However, such data often contains numerous low-quality samples, necessitating effective data processing (DP). In practice, DP strategies are typically developed through iterative manual analysis and trial-and-error adjustment. These processes inevitably incur high labor costs and may lead to privacy issues in high-privacy domains like healthcare due to direct human access to sensitive data. Thus, achieving automated data processing without exposing the raw data has become a critical challenge. To address this challenge, we propose LLM-AutoDP, a novel framework that leverages LLMs as agents to automatically generate and optimize data processing strategies. Our method generates multiple candidate strategies and iteratively refines them using feedback signals and comparative evaluations. This iterative in-context learning mechanism enables the agent to converge toward high-quality processing pipelines without requiring direct human intervention or access to the underlying data. To further accelerate strategy search, we introduce three key techniques: Distribution Preserving Sampling, which reduces data volume while maintaining distributional integrity; Processing Target Selection, which uses a binary classifier to identify low-quality samples for focused processing; Cache-and-Reuse Mechanism}, which minimizes redundant computations by reusing prior processing results. Results show that models trained on data processed by our framework achieve over 80% win rates against models trained on unprocessed data. Compared to AutoML baselines based on LLM agents, LLM-AutoDP achieves approximately a 65% win rate. Moreover, our acceleration techniques reduce the total searching time by up to 10 times, demonstrating both effectiveness and efficiency.", "date": "2026-01-28", "categories": []}
{"id": "2605.05724", "title": "Auto Research with Specialist Agents Develops Effective and Non-Trivial Training Recipes", "abstract": "We study auto research as a closed empirical loop driven by external measurement. Each submitted trial carries a hypothesis, an executable code edit, an evaluator-owned outcome, and feedback that shapes the next proposal. The output is not a generated paper or a single model checkpoint, but an auditable trajectory of proposals, code diffs, experiments, scores, and failure labels. We instantiate this loop with specialist agents that partition recipe surfaces and share measured lineage across trials. The central empirical finding is that lineage feedback lets agents turn evaluator outcomes, including crashes, budget overruns, size failures, and accuracy-gate misses, into later program-level recipe edits rather than one-shot suggestions. Across 1,197 headline-run trials plus 600 Parameter Golf control trials after one-time setup and launch, humans did not choose proposals, edit recipes, override scores, or repair failed trials during the search. In the three headline runs, the same submitted-trial loop reduces Parameter Golf validation bpb by $0.81\\%$, raises NanoChat-D12 CORE by $38.7\\%$, and reduces CIFAR-10 Airbench96 wallclock by $4.59\\%$, with each task measured by its own external evaluator and legality checks. The trace includes a strict architecture-domain audit of 157 headline-run submissions and program rewrites such as a NanoChat attention-kernel path change. Within this scope the loop autonomously writes code, submits experiments, absorbs feedback, applies and combines known techniques inside each environment, and improves public starting recipes.", "date": "2026-05-07", "categories": []}
{"id": "2601.03166", "title": "Dynamic Hyperparameter Importance for Efficient Multi-Objective Optimization", "abstract": "Choosing a suitable ML model is a complex task that can depend on several objectives, e.g., accuracy, fairness, or energy consumption. In practice, this requires trading off multiple, often competing, objectives through multi-objective optimization (MOO). However, existing MOO methods typically treat all hyperparameters as equally important, disregarding that hyperparameter importance (HPI) can vary significantly across objectives. We propose a novel dynamic optimization approach that prioritizes the most influential hyperparameters based on varying objective trade-offs during the search, thereby accelerating empirical convergence. We advance prior work on HPI for MOO from post-analysis to direct, dynamic integration within the optimization, using the recent HPI method HyperSHAP. For this, we leverage the objective weightings naturally produced by the MOO algorithm ParEGO and reduce the configuration space by fixing the unimportant hyperparameters, allowing the search to focus on the important ones. Eventually, we evaluate our method on diverse tasks from PyMOO and YAHPO-Gym. For HPO, integrating HPI yields up to 24% improvement in final Pareto front quality, while on synthetic data, integrating HPI achieves 2x better final results.", "date": "2026-01-06", "categories": []}
{"id": "2605.03686", "title": "From Code to Prediction: Fine-Tuning LLMs for Neural Network Performance Classification in NNGPT", "abstract": "Automated Machine Learning (AutoML) frameworks increasingly leverage Large Language Models (LLMs) for tasks such as hyperparameter optimization and neural architecture code generation. However, current LLM-based approaches focus on generative outputs and evaluate them by training the produced artifacts. Whether LLMs can learn to reason about neural network performance across datasets remains underexplored. We present a classification task integrated into the NNGPT framework, in which a fine-tuned LLM predicts which of two image classification datasets a given neural network architecture achieves higher accuracy on. The task is built on the LEMUR dataset, which provides standardized PyTorch implementations with reproducible performance metrics. Three prompt configurations of increasing difficulty are evaluated: a normalized-accuracy baseline (trivially reaching 100%), a metadata-enriched prompt replacing accuracies with dataset properties, and a code-only prompt presenting only architecture source code and dataset names. Using DeepSeek-Coder-7B-Instruct fine-tuned with LoRA, the code-only prompt reaches 80% peak accuracy over 15 epochs, while the metadata prompt peaks at 70%. Perdataset analysis reveals complementary strengths: metadata excels for datasets with distinctive properties (CelebAGender at 90.9%) but degrades for overlapping characteristics, whereas the code-only prompt shows more balanced performance. A comparison with DeepSeek-Coder1.3B confirms that model capacity affects this form of architectural reasoning. The results establish that LLMs can be fine-tuned to predict cross-dataset suitability from neural network code, suggesting that architecture source code contains richer discriminative signal than dataset metadata alone.", "date": "2026-05-05", "categories": []}
{"id": "2605.02967", "title": "AutoRAGTuner: A Declarative Framework for Automatic Optimization of RAG Pipelines", "abstract": "Retrieval-Augmented Generation (RAG) enhances LLMs, but performance is highly sensitive to complex architecture designs and hyper-parameter configurations, which currently rely on inefficient manual tuning. We present AutoRAGTuner, a declarative, configuration-driven framework that automates the RAG life cycle: construction, execution,evaluation, and optimization. AutoRAGTuner employs a modular architecture to decouple pipeline stages through a component registration mechanism. To unify heterogeneous data, we introduce the Domain-Element Model (DEM), representing objects as atomic elements with bidirectional pointers to support nodes, edges, and hyperedges. Furthermore, AutoRAGTuner integrates an adaptive Bayesian optimization engine for end-to-end hyper-parameter tuning. Experimental results demonstrate AutoRAGTuner's architectural generality: across diverse RAG pipelines, ranging from vanilla to graph-based, the framework consistently outperforms default baselines. Notably, AutoRAGTuner significantly mitigates engineering overhead, where its declarative configuration language enables a up to 95\\% reduction in code churn for architectural adjustments. Overall, AutoRAGTuner provides a systematically optimizable foundation for building evolvable and reusable RAG systems.", "date": "2026-05-03", "categories": []}
{"id": "2604.27025", "title": "SCOPE-FE: Structured Control of Operator and Pairwise Exploration for Feature Engineering", "abstract": "Automatic feature engineering is an effective approach for improving predictive performance in tabular learning. However, expand-and-reduce methods, such as OpenFE, become increasingly computationally expensive as the input dimensionality grows. This limitation arises primarily from the combinatorial explosion of candidate features generated through operator-feature combinations. To address this issue, we propose SCOPE-FE, a structured search space control framework that improves efficiency by reducing the candidate space prior to feature generation. SCOPE-FE jointly regulates two major sources of combinatorial growth: the operator space and feature-pair space. First, OperatorProbing estimates the dataset-specific utility of candidate operators and eliminates low-contribution operators in advance. Second, FeatureClustering employs spectral embedding and fuzzy c-means clustering to group structurally related features, thereby restricting candidate generation to relevant within-cluster combinations. In addition, we introduce ReliabilityScoring, which incorporates variance across subsamples to stabilize pruning decisions. Experiments on ten benchmark datasets demonstrate that SCOPE-FE substantially reduces feature engineering time while maintaining competitive predictive performance relative to existing baselines. The efficiency gains are particularly pronounced for high-dimensional datasets. These results indicate that structured control of the search space is an effective strategy for scalable automatic feature engineering. The code will be made publicly available upon acceptance.", "date": "2026-04-29", "categories": []}
{"id": "2604.27096", "title": "Think it, Run it: Autonomous ML pipeline generation via self-healing multi-agent AI", "abstract": "The purpose of our paper is to develop a unified multi-agent architecture that automates end-to-end machine learning (ML) pipeline generation from datasets and natural-language (NL) goals, improving efficiency, robustness and explainability. A five-agent system is proposed to handle profiling, intent parsing, microservice recommendation, Directed Acyclic Graph (DAG) construction and execution. It integrates code-grounded Retrieval-Augmented Generation (RAG) for microservice understanding, an explainable hybrid recommender combining multiple criteria, a self-healing mechanism using Large Language Model (LLM)-based error interpretation and adaptive learning from execution history. The approach is evaluated on 150 ML tasks across diverse scenarios. The system achieves an 84.7% end-to-end pipeline success rate, outperforming baseline methods. It demonstrates improved robustness through self-healing and reduces workflow development time compared to manual construction. The study introduces a novel integration of code-grounded RAG, explainable recommendation, self-healing execution and adaptive learning within a single architecture, showing that tightly coupled intelligent components can outperform isolated solutions.", "date": "2026-04-29", "categories": []}
{"id": "2604.26211", "title": "OMEGA: Optimizing Machine Learning by Evaluating Generated Algorithms", "abstract": "In order to automate AI research we introduce a full, end-to-end framework, OMEGA: Optimizing Machine learning by Evaluating Generated Algorithms, that starts at idea generation and ends with executable code. Our system combines structured meta-prompt engineering with executable code generation to create new ML classifiers. The OMEGA framework has been utilized to generate several novel algorithms that outperform scikit-learn baselines across a robust selection of 20 benchmark datasets (infinity-bench). You can access models discussed in this paper and more in the python package: pip install omega-models.", "date": "2026-04-29", "categories": []}
{"id": "2505.20662", "title": "AutoReproduce: Automatic AI Experiment Reproduction with Paper Lineage", "abstract": "Efficient reproduction of research papers is pivotal to accelerating scientific progress. However, the increasing complexity of proposed methods often renders reproduction a labor-intensive endeavor, necessitating profound domain expertise. To address this, we introduce the paper lineage, which systematically mines implicit knowledge from the cited literature. This algorithm serves as the backbone of our proposed \\ours, a multi-agent framework designed to autonomously reproduce experimental code in a complete, end-to-end manner. To ensure code executability, \\ours incorporates a sampling-based unit testing strategy for rapid validation. To assess reproduction capabilities, we introduce \\ourbench, a benchmark featuring verified implementations, alongside comprehensive metrics for evaluating both reproduction and execution fidelity. Extensive evaluations on PaperBench and \\ourbench demonstrate that \\ours consistently surpasses existing baselines across all metrics. Notably, it yields substantial improvements in reproduction fidelity and final execution performance. The code is available at https://github.com/AI9Stars/AutoReproduce.", "date": "2025-05-27", "categories": []}
{"id": "2508.10177", "title": "KompeteAI: Accelerated Autonomous Multi-Agent System for End-to-End Pipeline Generation for Machine Learning Problems", "abstract": "Recent Large Language Model (LLM)-based AutoML systems demonstrate impressive capabilities but face significant limitations such as constrained exploration strategies and a severe execution bottleneck. Exploration is hindered by one-shot methods lacking diversity and Monte Carlo Tree Search (MCTS) approaches that fail to recombine strong partial solutions. The execution bottleneck arises from lengthy code validation cycles that stifle iterative refinement. To overcome these challenges, we introduce KompeteAI, a novel AutoML framework with dynamic solution space exploration. Unlike previous MCTS methods that treat ideas in isolation, KompeteAI introduces a merging stage that composes top candidates. We further expand the hypothesis space by integrating Retrieval-Augmented Generation (RAG), sourcing ideas from Kaggle notebooks and arXiv papers to incorporate real-world strategies. KompeteAI also addresses the execution bottleneck via a predictive scoring model and an accelerated debugging method, assessing solution potential using early stage metrics to avoid costly full-code execution. This approach accelerates pipeline evaluation 6.9 times. KompeteAI outperforms leading methods (e.g., RD-agent, AIDE, and Ml-Master) by an average of 3\\% on the primary AutoML benchmark, MLE-Bench. Additionally, we propose Kompete-bench to address limitations in MLE-Bench, where KompeteAI also achieves state-of-the-art results", "date": "2025-08-13", "categories": []}
{"id": "2604.20938", "title": "HARBOR: Automated Harness Optimization", "abstract": "Long-horizon language-model agents are dominated, in lines of code and in operational complexity, not by their underlying model but by the harness that wraps it: context compaction, tool caching, semantic memory, trajectory reuse, speculative tool prediction, and the glue that binds the model to a sandboxed execution environment. We argue that harness design is a first-class machine-learning problem and that automated configuration search dominates manual stacking once the flag space exceeds a handful of bits. We defend this claim in two steps. First, we formalize automated harness optimization as constrained noisy Bayesian optimization over a mixed-variable, cost-heterogeneous configuration space with cold-start-corrected rewards and a posterior chance-constrained safety check, and give a reference solver, HARBOR (Harness Axis-aligned Regularized Bayesian Optimization Routine), built from a block-additive SAAS surrogate, multi-fidelity cost-aware acquisition, and TuRBO trust regions. Second, we instantiate the problem in a flag-gated harness over a production coding agent and report a controlled four-round manual-tuning case study against a fixed task suite and an end-to-end HARBOR run. The formulation itself is task-class agnostic: the configuration space, reward correction, acquisition, and safety check apply to any agent harness with a bounded flag space and a reproducible task suite.", "date": "2026-04-22", "categories": []}
{"id": "2604.14116", "title": "TREX: Automating LLM Fine-tuning via Agent-Driven Tree-based Exploration", "abstract": "While Large Language Models (LLMs) have empowered AI research agents to perform isolated scientific tasks, automating complex, real-world workflows, such as LLM training, remains a significant challenge. In this paper, we introduce TREX, a multi-agent system that automates the entire LLM training life-cycle. By orchestrating collaboration between two core modules-the Researcher and the Executor-the system seamlessly performs requirement analysis, open-domain literature and data research, formulation of training strategies, preparation of data recipes, and model training and evaluation. The multi-round experimental process is modeled as a search tree, enabling the system to efficiently plan exploration paths, reuse historical results, and distill high-level insights from iterative trials. To evaluate the capability of automated LLM training, we construct FT-Bench, a benchmark comprising 10 tasks derived from real-world scenarios, ranging from optimizing fundamental model capabilities to enhancing performance on domain-specific tasks. Experimental results demonstrate that the TREX agent consistently optimizes model performance on target tasks.", "date": "2026-04-15", "categories": []}
{"id": "2604.20261", "title": "Memory-Augmented LLM-based Multi-Agent System for Automated Feature Generation on Tabular Data", "abstract": "Automated feature generation extracts informative features from raw tabular data without manual intervention and is crucial for accurate, generalizable machine learning. Traditional methods rely on predefined operator libraries and cannot leverage task semantics, limiting their ability to produce diverse, high-value features for complex tasks. Recent Large Language Model (LLM)-based approaches introduce richer semantic signals, but still suffer from a restricted feature space due to fixed generation patterns and from the absence of feedback from the learning objective. To address these challenges, we propose a Memory-Augmented LLM-based Multi-Agent System (\\textbf{MALMAS}) for automated feature generation. MALMAS decomposes the generation process into agents with distinct responsibilities, and a Router Agent activates an appropriate subset of agents per iteration, further broadening exploration of the feature space. We further integrate a memory module comprising procedural memory, feedback memory, and conceptual memory, enabling iterative refinement that adaptively guides subsequent feature generation and improves feature quality and diversity. Extensive experiments on multiple public datasets against state-of-the-art baselines demonstrate the effectiveness of our approach. The code is available at https://github.com/fxdong24/MALMAS", "date": "2026-04-22", "categories": []}
{"id": "2603.24647", "title": "Can LLMs Beat Classical Hyperparameter Optimization Algorithms? A Study on autoresearch", "abstract": "The autoresearch repository enables an LLM agent to optimize hyperparameters by editing training code directly. We use it as a testbed to compare classical HPO algorithms against LLM-based methods on tuning the hyperparameters of a small language model under a fixed compute budget. When defining a fixed search space over autoresearch, classical methods such as CMA-ES and TPE consistently outperform LLM-based agents, where avoiding out-of-memory failures matters more than search diversity. Allowing the LLM to directly edit source code narrows the gap to the classical methods but does not close it, even with frontier models available at the time of writing such as Claude Opus 4.6 and Gemini 3.1 Pro Preview. We observe that LLMs struggle to track optimization state across trials. In contrast, classical methods lack the domain knowledge of LLMs. To combine the strengths of both, we introduce Centaur, a hybrid that shares CMA-ES's interpretable internal state, including mean vector, step-size, and covariance matrix, with an LLM. Centaur achieves the best result in our experiments, and a 0.8B LLM already suffices to outperform all classical and pure LLM methods. Unconstrained code editing requires larger models to be competitive with classical methods. We further analyze search diversity, model scaling from 0.8B to frontier models, and ablate the fraction of LLM-proposed trials in Centaur. All in all, our results suggest that LLMs are most effective as a complement to classical optimizers, not as a replacement.\n  Code is available at https://github.com/ferreirafabio/autoresearch-automl & interactive demo at https://ferreirafabio.github.io/autoresearch-automl.", "date": "2026-03-25", "categories": []}
{"id": "2604.16555", "title": "LLM as a Tool, Not an Agent: Code-Mined Tree Transformations for Neural Architecture Search", "abstract": "Neural Architecture Search (NAS) aims to automatically discover high-performing deep neural network (DNN) architectures. However, conventional algorithm-driven NAS relies on carefully hand-crafted search spaces to ensure executability, which restricts open-ended exploration. Recent coding-based agentic approaches using large language models (LLMs) reduce manual design, but current LLMs struggle to reliably generate complex, valid architectures, and their proposals are often biased toward a narrow set of patterns observed in their training data. To bridge reliable algorithmic search with powerful LLM assistance, we propose LLMasTool, a hierarchical tree-based NAS framework for stable and open-ended model evolution. Our method automatically extracts reusable modules from arbitrary source code and represents full architectures as hierarchical trees, enabling evolution through reliable tree transformations rather than code generation. At each evolution step, coarse-level planning is governed by a diversity-guided algorithm that leverages Bayesian modeling to improve exploration efficiency, while the LLM resolves the remaining degrees of freedom to ensure a meaningful evolutionary trajectory and an executable generated architecture. With this formulation, instead of fully agentic LLM approaches, our method explores diverse directions beyond the inherent biases in the LLM. Our method improves over existing NAS methods by 0.69, 1.83, and 2.68 points on CIFAR-10, CIFAR-100, and ImageNet16-120, demonstrating its effectiveness.", "date": "2026-04-17", "categories": []}
{"id": "2604.14455", "title": "AIBuildAI: An AI Agent for Automatically Building AI Models", "abstract": "AI models underpin modern intelligent systems, driving advances across science, medicine, finance, and technology. Yet developing high-performing AI models remains a labor-intensive process that requires expert practitioners to iteratively design architectures, engineer representations, implement training pipelines and refine approaches through empirical evaluation. Existing AutoML methods partially alleviate this burden but remain limited to narrow aspects such as hyperparameter optimization and model selection within predefined search spaces, leaving the full development lifecycle largely dependent on human expertise. To address this gap, we introduce AIBuildAI, an AI agent that automatically builds AI models from a task description and training data. AIBuildAI adopts a hierarchical agent architecture in which a manager agent coordinates three specialized sub-agents: a designer for modeling strategy, a coder for implementation and debugging, and a tuner for training and performance optimization. Each sub-agent is itself a large language model (LLM) based agent capable of multi-step reasoning and tool use, enabling end-to-end automation of the AI model development process that goes beyond the scope of existing AutoML approaches. We evaluate AIBuildAI on MLE-Bench, a benchmark of realistic Kaggle-style AI development tasks spanning visual, textual, time-series and tabular modalities. AIBuildAI ranks first on MLE-Bench with a medal rate of 63.1%, outperforming all existing baseline methods and matching the capability of highly experienced AI engineers. These results demonstrate that hierarchical agent systems can automate the full AI model development process from task specification to deployable model, suggesting a pathway toward broadly accessible AI development with minimal human intervention.", "date": "2026-04-15", "categories": []}
{"id": "2310.02540", "title": "Auto-FP: An Experimental Study of Automated Feature Preprocessing for Tabular Data", "abstract": "Classical machine learning models, such as linear models and tree-based models, are widely used in industry. These models are sensitive to data distribution, thus feature preprocessing, which transforms features from one distribution to another, is a crucial step to ensure good model quality. Manually constructing a feature preprocessing pipeline is challenging because data scientists need to make difficult decisions about which preprocessors to select and in which order to compose them. In this paper, we study how to automate feature preprocessing (Auto-FP) for tabular data. Due to the large search space, a brute-force solution is prohibitively expensive. To address this challenge, we interestingly observe that Auto-FP can be modelled as either a hyperparameter optimization (HPO) or a neural architecture search (NAS) problem. This observation enables us to extend a variety of HPO and NAS algorithms to solve the Auto-FP problem. We conduct a comprehensive evaluation and analysis of 15 algorithms on 45 public ML datasets. Overall, evolution-based algorithms show the leading average ranking. Surprisingly, the random search turns out to be a strong baseline. Many surrogate-model-based and bandit-based search algorithms, which achieve good performance for HPO and NAS, do not outperform random search for Auto-FP. We analyze the reasons for our findings and conduct a bottleneck analysis to identify the opportunities to improve these algorithms. Furthermore, we explore how to extend Auto-FP to support parameter search and compare two ways to achieve this goal. In the end, we evaluate Auto-FP in an AutoML context and discuss the limitations of popular AutoML tools. To the best of our knowledge, this is the first study on automated feature preprocessing. We hope our work can inspire researchers to develop new algorithms tailored for Auto-FP.", "date": "2023-10-04", "categories": []}
{"id": "2408.00601", "title": "AutoPV: Automatically Design Your Photovoltaic Power Forecasting Model", "abstract": "Photovoltaic power forecasting (PVPF) is a critical area in time series forecasting (TSF), enabling the efficient utilization of solar energy. With advancements in machine learning and deep learning, various models have been applied to PVPF tasks. However, constructing an optimal predictive architecture for specific PVPF tasks remains challenging, as it requires cross-domain knowledge and significant labor costs. To address this challenge, we introduce AutoPV, a novel framework for the automated search and construction of PVPF models based on neural architecture search (NAS) technology. We develop a brand new NAS search space that incorporates various data processing techniques from state-of-the-art (SOTA) TSF models and typical PVPF deep learning models. The effectiveness of AutoPV is evaluated on diverse PVPF tasks using a dataset from the Daqing Photovoltaic Station in China. Experimental results demonstrate that AutoPV can complete the predictive architecture construction process in a relatively short time, and the newly constructed architecture is superior to SOTA predefined models. This work bridges the gap in applying NAS to TSF problems, assisting non-experts and industries in automatically designing effective PVPF models.", "date": "2024-08-01", "categories": []}
{"id": "2604.11945", "title": "AutoSurrogate: An LLM-Driven Multi-Agent Framework for Autonomous Construction of Deep Learning Surrogate Models in Subsurface Flow", "abstract": "High-fidelity numerical simulation of subsurface flow is computationally intensive, especially for many-query tasks such as uncertainty quantification and data assimilation. Deep learning (DL) surrogates can significantly accelerate forward simulations, yet constructing them requires substantial machine learning (ML) expertise - from architecture design to hyperparameter tuning - that most domain scientists do not possess. Furthermore, the process is predominantly manual and relies heavily on heuristic choices. This expertise gap remains a key barrier to the broader adoption of DL surrogate techniques. For this reason, we present AutoSurrogate, a large-language-model-driven multi-agent framework that enables practitioners without ML expertise to build high-quality surrogates for subsurface flow problems through natural-language instructions. Given simulation data and optional preferences, four specialized agents collaboratively execute data profiling, architecture selection from a model zoo, Bayesian hyperparameter optimization, model training, and quality assessment against user-specified thresholds. The system also handles common failure modes autonomously, including restarting training with adjusted configurations when numerical instabilities occur and switching to alternative architectures when predictive accuracy falls short of targets. In our setting, a single natural-language sentence can be sufficient to produce a deployment-ready surrogate model, with minimum human intervention required at any intermediate stage. We demonstrate the utility of AutoSurrogate on a 3D geological carbon storage modeling task, mapping permeability fields to pressure and CO$_2$ saturation fields over 31 timesteps. Without any manual tuning, AutoSurrogate is able to outperform expert-designed baselines and domain-agnostic AutoML methods, demonstrating strong potential for practical deployment.", "date": "2026-04-13", "categories": []}
{"id": "2603.01692", "title": "Reasoning as Gradient: Scaling MLE Agents Beyond Tree Search", "abstract": "LLM-based agents for machine learning engineering (MLE) predominantly rely on tree search, a form of gradient-free optimization that uses scalar validation scores to rank candidates. As LLM reasoning capabilities improve, exhaustive enumeration becomes increasingly inefficient compared to directed updates, analogous to how accurate gradients enable efficient descent over random search. We introduce Gome, an MLE agent that operationalizes gradient-based optimization. Gome maps structured diagnostic reasoning to gradient computation, success memory to momentum, and multi-trace execution to distributed optimization. Under a closed-world protocol that isolates architectural effects from external knowledge, Gome achieves a state-of-the-art 35.1\\% any-medal rate on MLE-Bench with a restricted 12-hour budget on a single V100 GPU. Scaling experiments across 10 models reveal a critical crossover: with weaker models, tree search retains advantages by compensating for unreliable reasoning through exhaustive exploration; as reasoning capability strengthens, gradient-based optimization progressively outperforms, with the gap widening at frontier-tier models. Given the rapid advancement of reasoning-oriented LLMs, this positions gradient-based optimization as an increasingly favorable paradigm. We release our codebase and GPT-5 traces at https://github.com/microsoft/RD-Agent.", "date": "2026-03-02", "categories": []}
{"id": "2603.26499", "title": "AIRA_2: Overcoming Bottlenecks in AI Research Agents", "abstract": "Existing research has identified three structural performance bottlenecks in AI research agents: (1) synchronous single-GPU execution constrains sample throughput, limiting the benefit of search; (2) a generalization gap where validation-based selection causes overfitting and performance to degrade over extended search horizons; and (3) the limited capability of fixed, single-turn LLM operators imposes a ceiling on search performance. We introduce AIRA$_2$, which addresses these bottlenecks through three architectural choices: an asynchronous multi-GPU worker pool that increases experiment throughput linearly; a Hidden Consistent Evaluation protocol that delivers a reliable evaluation signal; and ReAct agents that dynamically scope their actions and debug interactively. On MLE-bench-30, AIRA$^{\\dagger}_{2}$ achieves a mean Percentile Rank of 81.5% at 24 hours and 83.1% at 72 hours, outperforming the strongest baseline, which achieves 72.7%. On AIRS-Bench, AIRA$_2$ exceeds human state-of-the-art on 6 out of 20 diverse research tasks. Ablations confirm that each architectural component is necessary, that performance follows a predictable scaling law that transfers across LLM backbones, and that the \"overfitting\" reported in prior work was driven by evaluation noise rather than true data memorization.", "date": "2026-03-27", "categories": []}
{"id": "2604.05225", "title": "fastml: Guarded Resampling Workflows for Safer Automated Machine Learning in R", "abstract": "Preprocessing leakage arises when scaling, imputation, or other data-dependent transformations are estimated before resampling, inflating apparent performance while remaining hard to detect. We present fastml, an R package that provides a single-call interface for leakage-aware machine learning through guarded resampling, where preprocessing is re-estimated inside each resample and applied to the corresponding assessment data. The package supports grouped and time-ordered resampling, blocks high-risk configurations, audits recipes for external dependencies, and includes sandboxed execution and integrated model explanation. We evaluate fastml with a Monte Carlo simulation contrasting global and fold-local normalization, a usability comparison with tidymodels under matched specifications, and survival benchmarks across datasets of different sizes. The simulation demonstrates that global preprocessing substantially inflates apparent performance relative to guarded resampling. fastml matched held-out performance obtained with tidymodels while reducing workflow orchestration, and it supported consistent benchmarking of multiple survival model classes through a unified interface.", "date": "2026-04-06", "categories": []}
{"id": "2604.09791", "title": "Pioneer Agent: Continual Improvement of Small Language Models in Production", "abstract": "Small language models are attractive for production deployment due to their low cost, fast inference, and ease of specialization. However, adapting them to a specific task remains a challenging engineering loop, driven not by training itself but by surrounding decisions: data curation, failure diagnosis, regression avoidance, and iteration control. We present Pioneer Agent, a closed-loop system that automates this lifecycle. In cold-start mode, given only a natural-language task description, the agent acquires data, constructs evaluation sets, and iteratively trains models by jointly optimizing data, hyperparameters, and learning strategy. In production mode, given a deployed model with labeled failures, it diagnoses error patterns, constructs targeted training data, and retrains under explicit regression constraints. To evaluate this setting, we introduce AdaptFT-Bench, a benchmark of synthetic inference logs with progressively increasing noise, designed to test the full adaptation loop: diagnosis, curriculum synthesis, retraining, and verification. Across eight cold-start benchmarks spanning reasoning, math, code generation, summarization, and classification, Pioneer Agent improves over base models by 1.6-83.8 points. On AdaptFT-Bench, it improves or preserves performance in all seven scenarios, while naive retraining degrades by up to 43 points. On two production-style deployments built from public benchmark tasks, it raises intent classification from 84.9% to 99.3% and Entity F1 from 0.345 to 0.810. Beyond performance gains, the agent often discovers effective training strategies, including chain-of-thought supervision, task-specific optimization, and quality-focused data curation, purely from downstream feedback.", "date": "2026-04-10", "categories": []}
{"id": "2604.10224", "title": "Exploring the impact of fairness-aware criteria in AutoML", "abstract": "Machine Learning (ML) systems are increasingly used to support decision-making processes that affect individuals. However, these systems often rely on biased data, which can lead to unfair outcomes against specific groups. With the growing adoption of Automated Machine Learning (AutoML), the risk of intensifying discriminatory behaviours increases, as most frameworks primarily focus on model selection to maximise predictive performance. Previous research on fairness in AutoML had largely followed this trend, integrating fairness awareness only in the model selection or hyperparameter tuning, while neglecting other critical stages of the ML pipeline. This paper aims to study the impact of integrating fairness directly into the optimisation component of an AutoML framework that constructs complete ML pipelines, from data selection and transformations to model selection and tuning. As selecting appropriate fairness metrics remains a key challenge, our work incorporates complementary fairness metrics to capture different dimensions of fairness during the optimisation. Their integration within AutoML resulted in measurable differences compared to a baseline focused solely on predictive performance. Despite a 9.4% decrease in predictive power, the average fairness improved by 14.5%, accompanied by a 35.7% reduction in data usage. Furthermore, fairness integration produced complete yet simpler final solutions, suggesting that model complexity is not always required to achieve balanced and fair ML solutions.", "date": "2026-04-11", "categories": []}
{"id": "2604.07610", "title": "Auto-Configured Networks for Multi-Scale Multi-Output Time-Series Forecasting", "abstract": "Industrial forecasting often involves multi-source asynchronous signals and multi-output targets, while deployment requires explicit trade-offs between prediction error and model complexity. Current practices typically fix alignment strategies or network designs, making it difficult to systematically co-design preprocessing, architecture, and hyperparameters in budget-limited training-based evaluations. To address this issue, we propose an auto-configuration framework that outputs a deployable Pareto set of forecasting models balancing error and complexity. At the model level, a Multi-Scale Bi-Branch Convolutional Neural Network (MS--BCNN) is developed, where short- and long-kernel branches capture local fluctuations and long-term trends, respectively, for multi-output regression. At the search level, we unify alignment operators, architectural choices, and training hyperparameters into a hierarchical-conditional mixed configuration space, and apply Player-based Hybrid Multi-Objective Evolutionary Algorithm (PHMOEA) to approximate the error--complexity Pareto frontier within a limited computational budget. Experiments on hierarchical synthetic benchmarks and a real-world sintering dataset demonstrate that our framework outperforms competitive baselines under the same budget and offers flexible deployment choices.", "date": "2026-04-08", "categories": []}
{"id": "2508.13657", "title": "In-Context Decision Making for Optimizing Complex AutoML Pipelines", "abstract": "Combined Algorithm Selection and Hyperparameter Optimization (CASH) has been fundamental to traditional AutoML systems. However, with the advancements of pre-trained models, modern ML workflows go beyond hyperparameter optimization and often require fine-tuning, ensembling, and other adaptation techniques. While the core challenge of identifying the best-performing model for a downstream task remains, the increasing heterogeneity of ML pipelines demands novel AutoML approaches. This work extends the CASH framework to select and adapt modern ML pipelines. We propose PS-PFN to efficiently explore and exploit adapting ML pipelines by extending Posterior Sampling (PS) to the max k-armed bandit problem setup. PS-PFN leverages prior-data fitted networks (PFNs) to efficiently estimate the posterior distribution of the maximal value via in-context learning. We show how to extend this method to consider varying costs of pulling arms and to use different PFNs to model reward distributions individually per arm. Experimental results on one novel and two existing standard benchmark tasks demonstrate the superior performance of PS-PFN compared to other bandit and AutoML strategies. We make our code and data available at https://github.com/amirbalef/CASHPlus.", "date": "2025-08-19", "categories": []}
{"id": "2501.12215", "title": "Automatic selection of the best neural architecture for time series forecasting", "abstract": "Time series forecasting plays a pivotal role in a wide range of applications, including weather prediction, healthcare, structural health monitoring, predictive maintenance, energy systems, and financial markets. While models such as LSTM, GRU, Transformers, and State-Space Models (SSMs) have become standard tools in this domain, selecting the optimal architecture remains a challenge. Performance comparisons often depend on evaluation metrics and the datasets under analysis, making the choice of a universally optimal model controversial. In this work, we introduce a flexible automated framework for time series forecasting that systematically designs and evaluates diverse network architectures by integrating LSTM, GRU, multi-head Attention, and SSM blocks. Using a multi-objective optimization approach, our framework determines the number, sequence, and combination of blocks to align with specific requirements and evaluation objectives. From the resulting Pareto-optimal architectures, the best model for a given context is selected via a user-defined preference function. We validate our framework across four distinct real-world applications. Results show that a single-layer GRU or LSTM is usually optimal when minimizing training time alone. However, when maximizing accuracy or balancing multiple objectives, the best architectures are often composite designs incorporating multiple block types in specific configurations. By employing a weighted preference function, users can resolve trade-offs between objectives, revealing novel, context-specific optimal architectures. Our findings underscore that no single neural architecture is universally optimal for time series forecasting. Instead, the best-performing model emerges as a data-driven composite architecture tailored to user-defined criteria and evaluation objectives.", "date": "2025-01-21", "categories": []}
{"id": "2603.29640", "title": "ASI-Evolve: AI Accelerates AI", "abstract": "Can AI accelerate the development of AI itself? While recent agentic systems have shown strong performance on well-scoped tasks with rapid feedback, it remains unclear whether they can tackle the costly, long-horizon, and weakly supervised research loops that drive real AI progress. We present ASI-Evolve, an agentic framework for AI-for-AI research that closes this loop through a learn-design-experiment-analyze cycle. ASI-Evolve augments standard evolutionary agents with two key components: a cognition base that injects accumulated human priors into each round of exploration, and a dedicated analyzer that distills complex experimental outcomes into reusable insights for future iterations. To our knowledge, ASI-Evolve is the first unified framework to demonstrate AI-driven discovery across three central components of AI development: data, architectures, and learning algorithms. In neural architecture design, it discovered 105 SOTA linear attention architectures, with the best discovered model surpassing DeltaNet by +0.97 points, nearly 3x the gain of recent human-designed improvements. In pretraining data curation, the evolved pipeline improves average benchmark performance by +3.96 points, with gains exceeding 18 points on MMLU. In reinforcement learning algorithm design, discovered algorithms outperform GRPO by up to +12.5 points on AMC32, +11.67 points on AIME24, and +5.04 points on OlympiadBench. We further provide initial evidence that this AI-for-AI paradigm can transfer beyond the AI stack through experiments in mathematics and biomedicine. Together, these results suggest that ASI-Evolve represents a promising step toward enabling AI to accelerate AI across the foundational stages of development, offering early evidence for the feasibility of closed-loop AI research.", "date": "2026-03-31", "categories": []}
{"id": "2603.27415", "title": "Greedy Is a Strong Default: Agents as Iterative Optimizers", "abstract": "Classical optimization algorithms--hill climbing, simulated annealing, population-based methods--generate candidate solutions via random perturbations. We replace the random proposal generator with an LLM agent that reasons about evaluation diagnostics to propose informed candidates, and ask: does the classical optimization machinery still help when the proposer is no longer random? We evaluate on four tasks spanning discrete, mixed, and continuous search spaces (all replicated across 3 independent runs): rule-based classification on Breast Cancer (test accuracy 86.0% to 96.5%), mixed hyperparameter optimization for MobileNetV3-Small on STL-10 (84.5% to 85.8%, zero catastrophic failures vs. 60% for random search), LoRA fine-tuning of Qwen2.5-0.5B on SST-2 (89.5% to 92.7%, matching Optuna TPE with 2x efficiency), and XGBoost on Adult Census (AUC 0.9297 to 0.9317, tying CMA-ES with 3x fewer evaluations). Empirically, on these tasks: a cross-task ablation shows that simulated annealing, parallel investigators, and even a second LLM model (OpenAI Codex) provide no benefit over greedy hill climbing while requiring 2-3x more evaluations. In our setting, the LLM's learned prior appears strong enough that acceptance-rule sophistication has limited impact--round 1 alone delivers the majority of improvement, and variants converge to similar configurations across strategies. The practical implication is surprising simplicity: greedy hill climbing with early stopping is a strong default. Beyond accuracy, the framework produces human-interpretable artifacts--the discovered cancer classification rules independently recapitulate established cytopathology principles.", "date": "2026-03-28", "categories": []}
{"id": "2603.25813", "title": "MAGNET: Autonomous Expert Model Generation via Decentralized Autoresearch and BitNet Training", "abstract": "We present MAGNET (Model Autonomously Growing Network), a decentralized system for autonomous generation, training, and serving of domain-expert language models across commodity hardware. MAGNET integrates four components: (1) autoresearch, an autonomous ML research pipeline that automates dataset generation, hyperparameter exploration, evaluation, and error-driven iteration; (2) BitNet b1.58 ternary training, enabling CPU-native inference via bitnet.cpp without GPU hardware; (3) DiLoCo-based distributed merging for communication-efficient aggregation of domain specialists; and (4) on-chain contribution tracking on the HOOTi EVM chain. We validate autoresearch through three case studies: video safety classification (balanced accuracy 0.9287 to 0.9851), cryptocurrency directional prediction (41% to 54.9% hit rate), and BitNet hyperparameter optimization (10-phase sweep, -16.7% validation loss).", "date": "2026-03-26", "categories": []}
{"id": "2508.05144", "title": "PSEO: Optimizing Post-hoc Stacking Ensemble Through Hyperparameter Tuning", "abstract": "The Combined Algorithm Selection and Hyperparameter Optimization (CASH) problem is fundamental in Automated Machine Learning (AutoML). Inspired by the success of ensemble learning, recent AutoML systems construct post-hoc ensembles for final predictions rather than relying on the best single model. However, while most CASH methods conduct extensive searches for the optimal single model, they typically employ fixed strategies during the ensemble phase that fail to adapt to specific task characteristics. To tackle this issue, we propose PSEO, a framework for post-hoc stacking ensemble optimization. First, we conduct base model selection through binary quadratic programming, with a trade-off between diversity and performance. Furthermore, we introduce two mechanisms to fully realize the potential of multi-layer stacking. Finally, PSEO builds a hyperparameter space and searches for the optimal post-hoc ensemble strategy within it. Empirical results on 80 public datasets show that \\sys achieves the best average test rank (2.96) among 16 methods, including post-hoc designs in recent AutoML systems and state-of-the-art ensemble learning methods.", "date": "2025-08-07", "categories": []}
{"id": "2603.23951", "title": "From AI Assistant to AI Scientist: Autonomous Discovery of LLM-RL Algorithms with LLM Agents", "abstract": "Discovering improved policy optimization algorithms for language models remains a costly manual process requiring repeated mechanism-level modification and validation. Unlike simple combinatorial code search, this problem requires searching over algorithmic mechanisms tightly coupled with training dynamics while reusing empirical evidence across iterations. We propose POISE, a closed-loop framework for automated discovery of policy optimization algorithms for language models. POISE maintains a structured, genealogically linked archive linking proposals, executable implementations, standardized evaluations, and natural-language reflections to support evidence-driven iteration. In mathematical reasoning experiments starting from GRPO, POISE evaluates 64 candidate algorithms and discovers improved mechanisms, including analytic-variance scaling and validity masking. The best variant improves weighted Overall from 47.8 to 52.5 (+4.6) and increases AIME25 pass@32 from 26.7% to 43.3%, demonstrating the feasibility of automated policy optimization discovery while supporting interpretable design principles.", "date": "2026-03-25", "categories": []}
{"id": "2603.21698", "title": "A Blueprint for Self-Evolving Coding Agents in Vehicle Aerodynamic Drag Prediction", "abstract": "High-fidelity vehicle drag evaluation is constrained less by solver runtime than by workflow friction: geometry cleanup, meshing retries, queue contention, and reproducibility failures across teams. We present a contract-centric blueprint for self-evolving coding agents that discover executable surrogate pipelines for predicting drag coefficient $C_d$ under industrial constraints. The method formulates surrogate discovery as constrained optimization over programs, not static model instances, and combines Famou-Agent-style evaluator feedback with population-based island evolution, structured mutations (data, model, loss, and split policies), and multi-objective selection balancing ranking quality, stability, and cost. A hard evaluation contract enforces leakage prevention, deterministic replay, multi-seed robustness, and resource budgets before any candidate is admitted. Across eight anonymized evolutionary operators, the best system reaches a Combined Score of 0.9335 with sign-accuracy 0.9180, while trajectory and ablation analyses show that adaptive sampling and island migration are primary drivers of convergence quality. The deployment model is explicitly ``screen-and-escalate'': surrogates provide high-throughput ranking for design exploration, but low-confidence or out-of-distribution cases are automatically escalated to high-fidelity CFD. The resulting contribution is an auditable, reusable workflow for accelerating aerodynamic design iteration while preserving decision-grade reliability, governance traceability, and safety boundaries.", "date": "2026-03-23", "categories": []}
{"id": "2603.20014", "title": "AgenticRS-EnsNAS: Ensemble-Decoupled Self-Evolving Architecture Search", "abstract": "Neural Architecture Search (NAS) deployment in industrial production systems faces a fundamental validation bottleneck: verifying a single candidate architecture pi requires evaluating the deployed ensemble of M models, incurring prohibitive O(M) computational cost per candidate. This cost barrier severely limits architecture iteration frequency in real-world applications where ensembles (M=50-200) are standard for robustness. This work introduces Ensemble-Decoupled Architecture Search, a framework that leverages ensemble theory to predict system-level performance from single-learner evaluation. We establish the Ensemble-Decoupled Theory with a sufficient condition for monotonic ensemble improvement under homogeneity assumptions: a candidate architecture pi yields lower ensemble error than the current baseline if rho(pi) < rho(pi_old) - (M / (M - 1)) * (Delta E(pi) / sigma^2(pi)), where Delta E, rho, and sigma^2 are estimable from lightweight dual-learner training. This decouples architecture search from full ensemble training, reducing per-candidate search cost from O(M) to O(1) while maintaining O(M) deployment cost only for validated winners. We unify solution strategies across pipeline continuity: (1) closed-form optimization for tractable continuous pi (exemplified by feature bagging in CTR prediction), (2) constrained differentiable optimization for intractable continuous pi, and (3) LLM-driven search with iterative monotonic acceptance for discrete pi. The framework reveals two orthogonal improvement mechanisms -- base diversity gain and accuracy gain -- providing actionable design principles for industrial-scale NAS. All theoretical derivations are rigorous with detailed proofs deferred to the appendix. Comprehensive empirical validation will be included in the journal extension of this work.", "date": "2026-03-20", "categories": []}
{"id": "2603.18773", "title": "Automatic Configuration of LLM Post-Training Pipelines", "abstract": "LLM post-training pipelines that combine supervised fine-tuning and reinforcement learning are difficult to configure under realistic compute budgets: the configuration space is high-dimensional and heterogeneous, stages are strongly coupled, and each end-to-end evaluation is expensive. We propose AutoPipe, a budget-aware two-stage framework for configuration selection in LLM post-training. Offline, AutoPipe learns a dataset-conditioned learning-to-rank surrogate from historical runs, capturing within-dataset preferences and providing transferable guidance toward promising regions of the configuration space. Online, for a new dataset, AutoPipe uses the offline guidance to steer Bayesian optimization and models dataset-specific deviations with a Gaussian-process residual surrogate. To reduce evaluation cost, each trial is early-stopped and scored by a learned predictor that maps early training signals to a low-cost proxy for final post-training performance. Experiments on biomedical reasoning tasks show that AutoPipe consistently outperforms offline-only baselines and achieves comparable performance with the strongest online HPO baselines while using less than 10\\% of their computational cost.", "date": "2026-03-19", "categories": []}
{"id": "2603.17863", "title": "Procedural Generation of Algorithm Discovery Tasks in Machine Learning", "abstract": "Automating the development of machine learning algorithms has the potential to unlock new breakthroughs. However, our ability to improve and evaluate algorithm discovery systems has thus far been limited by existing task suites. They suffer from many issues, such as: poor evaluation methodologies; data contamination; and containing saturated or very similar problems. Here, we introduce DiscoGen, a procedural generator of algorithm discovery tasks for machine learning, such as developing optimisers for reinforcement learning or loss functions for image classification. Motivated by the success of procedural generation in reinforcement learning, DiscoGen spans millions of tasks of varying difficulty and complexity from a range of machine learning fields. These tasks are specified by a small number of configuration parameters and can be used to optimise algorithm discovery agents (ADAs). We present DiscoBench, a benchmark consisting of a fixed, small subset of DiscoGen tasks for principled evaluation of ADAs. Finally, we propose a number of ambitious, impactful research directions enabled by DiscoGen, in addition to experiments demonstrating its use for prompt optimisation of an ADA. DiscoGen is released open-source at https://github.com/AlexGoldie/discogen.", "date": "2026-03-18", "categories": []}
{"id": "2603.15916", "title": "Auto Researching, not hyperparameter tuning: Convergence Analysis of 10,000 Experiments", "abstract": "When LLM agents autonomously design ML experiments, do they perform genuine architecture search -- or do they default to hyperparameter tuning within a narrow region of the design space? We answer this question by analyzing 10,469 experiments executed by two LLM agents (Claude Opus and Gemini 2.5 Pro) across a combinatorial configuration space of 108,000 discrete cells for dashcam collision detection over 27 days. Through ANOVA decomposition, we find that \\textbf{architectural choices explain 94\\% of performance variance} ($F = 1324$, $\\eta^2 = 0.94$), while hyperparameter variation within a fixed architecture explains only 6\\%. Cross-task validation on a second collision dataset confirms this finding (75\\% architecture-explained variance) with a \\emph{different} winning backbone, confirming genuine architecture discovery. The agents' key contribution is discovering that V-JEPA\\,2 video features with Zipformer temporal encoders achieve 0.9245 AP -- a configuration no human proposed -- and concentrating search on productive architectural regions: at $N = 50$, LLM-guided search reaches AP $= 0.985$ versus $0.965$ for from-scratch random search. Post-bugfix convergence follows a power law ($c = 0.11$, $R^2 = 0.93$); the low exponent reflects the cost of broad exploration, not inefficiency, since the LLM discovers qualitatively better regions than random or Bayesian baselines. We characterize multi-agent search dynamics via entropy cycles and Jensen--Shannon specialization, providing the first large-scale empirical framework for LLM-guided combinatorial ML experiment design.", "date": "2026-03-16", "categories": []}
{"id": "2603.15321", "title": "CASHomon Sets: Efficient Rashomon Sets Across Multiple Model Classes and their Hyperparameters", "abstract": "Rashomon sets are model sets within one model class that perform nearly as well as a reference model from the same model class. They reveal the existence of alternative well-performing models, which may support different interpretations. This enables selecting models that match domain knowledge, hidden constraints, or user preferences. However, efficient construction methods currently exist for only a few model classes. Applied machine learning usually searches many model classes, and the best class is unknown beforehand. We therefore study Rashomon sets in the combined algorithm selection and hyperparameter optimization (CASH) setting and call them CASHomon sets. We propose TruVaRImp, a model-based active learning algorithm for level set estimation with an implicit threshold, and provide convergence guarantees. On synthetic and real-world datasets, TruVaRImp reliably identifies CASHomon sets members and matches or outperforms naive sampling, Bayesian optimization, classical and implicit level set estimation methods, and other baselines. Our analyses of predictive multiplicity and feature-importance variability across model classes question the common practice of interpreting data through a single model class.", "date": "2026-03-16", "categories": []}
{"id": "2603.11603", "title": "AutoScout: Structured Optimization for Automating ML System Configuration", "abstract": "Machine learning (ML) systems expose a rapidly expanding configuration space spanning model-parallelism strategies, communication optimizations, and low-level runtime parameters. End-to-end system efficiency is highly sensitive to these choices, yet identifying high-performance configurations is challenging due to heterogeneous feature types (e.g., sparse and dense parameters), conditional dependencies (e.g., valid execution parameters only under specific upstream decisions), and the high search (profiling) cost. Existing approaches either optimize a narrow subset of configuration dimensions or rely on ad-hoc heuristics that fail to generalize as configuration spaces continue to grow. We present AutoScout, a general-purpose systems configurator for ML training, fine-tuning, and inference. It formulates the system configuration as a mixed-discrete/continuous optimization problem with hierarchical dependencies and introduces a hybrid optimization framework that jointly refines sparse structural decisions and dense execution parameters. To reduce profiling cost, AutoScout adaptively prioritizes high-impact configuration features and ensembles simulators with varying fidelity. Across diverse models, hardware platforms, and deployment objectives, AutoScout consistently identifies high-performance configurations, achieving 2.7-3.0$\\times$ training speedup over expert-tuned settings.", "date": "2026-03-12", "categories": []}
{"id": "2603.07897", "title": "LeJOT-AutoML: LLM-Driven Feature Engineering for Job Execution Time Prediction in Databricks Cost Optimization", "abstract": "Databricks job orchestration systems (e.g., LeJOT) reduce cloud costs by selecting low-priced compute configurations while meeting latency and dependency constraints. Accurate execution-time prediction under heterogeneous instance types and non-stationary runtime conditions is therefore critical. Existing pipelines rely on static, manually engineered features that under-capture runtime effects (e.g., partition pruning, data skew, and shuffle amplification), and predictive signals are scattered across logs, metadata, and job scripts-lengthening update cycles and increasing engineering overhead. We present LeJOT-AutoML, an agent-driven AutoML framework that embeds large language model agents throughout the ML lifecycle. LeJOT-AutoML combines retrieval-augmented generation over a domain knowledge base with a Model Context Protocol toolchain (log parsers, metadata queries, and a read-only SQL sandbox) to analyze job artifacts, synthesize and validate feature-extraction code via safety gates, and train/select predictors. This design materializes runtime-derived features that are difficult to obtain through static analysis alone. On enterprise Databricks workloads, LeJOT-AutoML generates over 200 features and reduces the feature-engineering and evaluation loop from weeks to 20-30 minutes, while maintaining competitive prediction accuracy. Integrated into the LeJOT pipeline, it enables automated continuous model updates and achieves 19.01% cost savings in our deployment setting through improved orchestration.", "date": "2026-03-09", "categories": []}
{"id": "2603.08181", "title": "AutoAdapt: An Automated Domain Adaptation Framework for LLMs", "abstract": "Large language models (LLMs) excel in open domains but struggle in specialized settings with limited data and evolving knowledge. Existing domain adaptation practices rely heavily on manual trial-and-error processes, incur significant hyperparameter complexity, and are highly sensitive to data and user preferences, all under the high cost of LLM training. Moreover, the interactions and transferability of hyperparameter choices across models/domains remain poorly understood, making adaptation gains uncertain even with substantial effort. To solve these challenges, we present AutoAdapt, a novel end-to-end automated framework for efficient and reliable LLM domain adaptation. AutoAdapt leverages curated knowledge bases from literature and open-source resources to reduce expert intervention. To narrow the search space, we design a novel multi-agent debating system in which proposal and critic agents iteratively interact to align user intent and incorporate data signals and best practices into the planning process. To optimize hyperparameters under tight budgets, we propose AutoRefine, a novel LLM-based surrogate that replaces costly black-box search. Across 10 tasks, AutoAdapt achieves a 25% average relative accuracy improvement over state-of-the-art Automated Machine Learning baselines with minimal overhead.", "date": "2026-03-09", "categories": []}
{"id": "2603.04873", "title": "SEA-TS: Self-Evolving Agent for Autonomous Code Generation of Time Series Forecasting Algorithms", "abstract": "Accurate time series forecasting underpins decision-making across domains, yet conventional ML development suffers from data scarcity in new deployments, poor adaptability under distribution shift, and diminishing returns from manual iteration. We propose Self-Evolving Agent for Time Series Algorithms (SEA-TS), a framework that autonomously generates, validates, and optimizes forecasting code via an iterative self-evolution loop. Our framework introduces three key innovations: (1) Metric-Advantage Monte Carlo Tree Search (MA-MCTS), which replaces fixed rewards with a normalized advantage score for discriminative search guidance; (2) Code Review with running prompt refinement, where each executed solution undergoes automated review followed by prompt updates that encode corrective patterns, preventing recurrence of similar errors; and (3) Global Steerable Reasoning, which compares each node against global best and worst solutions, enabling cross-trajectory knowledge transfer. We adopt a MAP-Elites archive for architectural diversity. On the public Solar-Energy benchmark, SEA-TS generated code achieves a 40% MAE reduction relative to TimeMixer, surpassing state-of-the-art methods. On proprietary datasets, SEA-TS generated code reduces WAPE by 8.6% on solar PV forecasting and 7.7% on residential load forecasting compared to human-engineered baselines, and achieves 26.17% MAPE on load forecasting versus 29.34% by TimeMixer. Notably, the evolved models discover novel architectural patterns--including physics-informed monotonic decay heads encoding solar irradiance constraints, per-station learned diurnal cycle profiles, and learnable hourly bias correction--demonstrating that autonomous ML engineering can generate genuinely novel algorithmic ideas beyond manual design.", "date": "2026-03-05", "categories": []}
{"id": "2603.01824", "title": "OpenAutoNLU: Open Source AutoML Library for NLU", "abstract": "OpenAutoNLU is an open-source automated machine learning library for natural language understanding (NLU) tasks, covering both text classification and named entity recognition (NER). Unlike existing solutions, we introduce data-aware training regime selection that requires no manual configuration from the user. The library also provides integrated data quality diagnostics, configurable out-of-distribution (OOD) detection, and large language model (LLM) features, all within a minimal lowcode API. The demo app is accessible here https://openautonlu.dev.", "date": "2026-03-02", "categories": []}
{"id": "2506.20640", "title": "CoMind: Towards Community-Driven Agents for Machine Learning Engineering", "abstract": "Large language model (LLM) agents show promise in automating machine learning (ML) engineering. However, existing agents typically operate in isolation on a given research problem, without engaging with the broader research community, where human researchers often gain insights and contribute by sharing knowledge. To bridge this gap, we introduce MLE-Live, a live evaluation framework designed to assess an agent's ability to communicate with and leverage collective knowledge from a simulated Kaggle research community. Building on this framework, we propose CoMind, a multi-agent system designed to systematically leverage external knowledge. CoMind employs an iterative parallel exploration mechanism, developing multiple solutions simultaneously to balance exploratory breadth with implementation depth. On 75 past Kaggle competitions within our MLE-Live framework, CoMind achieves a 36% medal rate, establishing a new state of the art. Critically, when deployed in eight live, ongoing competitions, CoMind outperforms 92.6% of human competitors on average, placing in the top 5% on three official leaderboards and the top 1% on one.", "date": "2025-06-25", "categories": []}
{"id": "2602.23630", "title": "BTTackler: A Diagnosis-based Framework for Efficient Deep Learning Hyperparameter Optimization", "abstract": "Hyperparameter optimization (HPO) is known to be costly in deep learning, especially when leveraging automated approaches. Most of the existing automated HPO methods are accuracy-based, i.e., accuracy metrics are used to guide the trials of different hyperparameter configurations amongst a specific search space. However, many trials may encounter severe training problems, such as vanishing gradients and insufficient convergence, which can hardly be reflected by accuracy metrics in the early stages of the training and often result in poor performance. This leads to an inefficient optimization trajectory because the bad trials occupy considerable computation resources and reduce the probability of finding excellent hyperparameter configurations within a time limitation. In this paper, we propose \\textbf{Bad Trial Tackler (BTTackler)}, a novel HPO framework that introduces training diagnosis to identify training problems automatically and hence tackles bad trials. BTTackler diagnoses each trial by calculating a set of carefully designed quantified indicators and triggers early termination if any training problems are detected. Evaluations are performed on representative HPO tasks consisting of three classical deep neural networks (DNN) and four widely used HPO methods. To better quantify the effectiveness of an automated HPO method, we propose two new measurements based on accuracy and time consumption. Results show the advantage of BTTackler on two-fold: (1) it reduces 40.33\\% of time consumption to achieve the same accuracy comparable to baseline methods on average and (2) it conducts 44.5\\% more top-10 trials than baseline methods on average within a given time budget. We also released an open-source Python library that allows users to easily apply BTTackler to automated HPO processes with minimal code changes.", "date": "2026-02-27", "categories": []}
{"id": "2602.21634", "title": "AgentLTV: An Agent-Based Unified Search-and-Evolution Framework for Automated Lifetime Value Prediction", "abstract": "Lifetime Value (LTV) prediction is critical in advertising, recommender systems, and e-commerce. In practice, LTV data patterns vary across decision scenarios. As a result, practitioners often build complex, scenario-specific pipelines and iterate over feature processing, objective design, and tuning. This process is expensive and hard to transfer. We propose AgentLTV, an agent-based unified search-and-evolution framework for automated LTV modeling. AgentLTV treats each candidate solution as an {executable pipeline program}. LLM-driven agents generate code, run and repair pipelines, and analyze execution feedback. Two decision agents coordinate a two-stage search. The Monte Carlo Tree Search (MCTS) stage explores a broad space of modeling choices under a fixed budget, guided by the Polynomial Upper Confidence bounds for Trees criterion and a Pareto-aware multi-metric value function. The Evolutionary Algorithm (EA) stage refines the best MCTS program via island-based evolution with crossover, mutation, and migration. Experiments on a large-scale proprietary dataset and a public benchmark show that AgentLTV consistently discovers strong models across ranking and error metrics. Online bucket-level analysis further indicates improved ranking consistency and value calibration, especially for high-value and negative-LTV segments. We summarize practitioner-oriented takeaways: use MCTS for rapid adaptation to new data patterns, use EA for stable refinement, and validate deployment readiness with bucket-level ranking and calibration diagnostics. The proposed AgentLTV has been successfully deployed online.", "date": "2026-02-25", "categories": []}
{"id": "2512.00672", "title": "ML-Tool-Bench: Tool-Augmented Planning for ML Tasks", "abstract": "The development of autonomous machine learning (ML) agents capable of end-to-end data science workflows represents a significant frontier in artificial intelligence. These agents must orchestrate complex sequences of data analysis, feature engineering, model selection, and hyperparameter optimization, tasks that require sophisticated planning and iteration. While recent work on building ML agents has explored using large language models (LLMs) for direct code generation, tool-augmented approaches offer greater modularity and reliability. However, existing tool-use benchmarks focus primarily on task-specific tool selection or argument extraction for tool invocation, failing to evaluate the sophisticated planning capabilities required for ML Agents. In this work, we introduce a comprehensive benchmark for evaluating tool-augmented ML agents using a curated set of 61 specialized tools and 15 tabular ML challenges from Kaggle. Our benchmark goes beyond traditional tool-use evaluation by incorporating an in-memory named object management, allowing agents to flexibly name, save, and retrieve intermediate results throughout the workflows. We demonstrate that standard ReAct-style approaches struggle to generate valid tool sequences for complex ML pipelines, and that tree search methods with LLM-based evaluation underperform due to inconsistent state scoring. To address these limitations, we propose two simple approaches: 1) using shaped deterministic rewards with structured textual feedback, and 2) decomposing the original problem into a sequence of sub-tasks, which significantly improves trajectory validity and task performance. Using GPT-4o, our approach improves over ReAct by 16.52 percentile positions, taking the median across all Kaggle challenges. We believe our work provides a foundation for developing more capable tool-augmented planning ML agents.", "date": "2025-11-29", "categories": []}
{"id": "2602.18348", "title": "Explaining AutoClustering: Uncovering Meta-Feature Contribution in AutoML for Clustering", "abstract": "AutoClustering methods aim to automate unsupervised learning tasks, including algorithm selection (AS), hyperparameter optimization (HPO), and pipeline synthesis (PS), by often leveraging meta-learning over dataset meta-features. While these systems often achieve strong performance, their recommendations are often difficult to justify: the influence of dataset meta-features on algorithm and hyperparameter choices is typically not exposed, limiting reliability, bias diagnostics, and efficient meta-feature engineering. This limits reliability and diagnostic insight for further improvements. In this work, we investigate the explainability of the meta-models in AutoClustering. We first review 22 existing methods and organize their meta-features into a structured taxonomy. We then apply a global explainability technique (i.e., Decision Predicate Graphs) to assess feature importance within meta-models from selected frameworks. Finally, we use local explainability tools such as SHAP (SHapley Additive exPlanations) to analyse specific clustering decisions. Our findings highlight consistent patterns in meta-feature relevance, identify structural weaknesses in current meta-learning strategies that can distort recommendations, and provide actionable guidance for more interpretable Automated Machine Learning (AutoML) design. This study therefore offers a practical foundation for increasing decision transparency in unsupervised learning automation.", "date": "2026-02-20", "categories": []}
{"id": "2602.08351", "title": "The Chicken and Egg Dilemma: Co-optimizing Data and Model Configurations for LLMs", "abstract": "Co-optimizing data and model configurations for training LLMs presents a classic chicken-and-egg dilemma: The best training data configuration (e.g., data mixture) for a downstream task depends on the chosen model configuration (e.g., model architecture), and vice versa. However, jointly optimizing both data and model configurations is often deemed intractable, and existing methods focus on either data or model optimization without considering their interaction. We introduce JoBS, an approach that uses a scaling-law-inspired performance predictor to aid Bayesian optimization (BO) in jointly optimizing LLM training data and model configurations efficiently. JoBS allocates a portion of the optimization budget to learn an LLM performance predictor that predicts how promising a training configuration is from a small number of training steps. The remaining budget is used to perform BO entirely with the predictor, effectively amortizing the cost of running full-training runs. We study JoBS's average regret and devise the optimal budget allocation to minimize regret. JoBS outperforms existing multi-fidelity BO baselines, as well as data and model optimization approaches across diverse LLM tasks under the same optimization budget.", "date": "2026-02-09", "categories": []}
{"id": "2602.17641", "title": "FAMOSE: A ReAct Approach to Automated Feature Discovery", "abstract": "Feature engineering remains a critical yet challenging bottleneck in machine learning, particularly for tabular data, as identifying optimal features from an exponentially large feature space traditionally demands substantial domain expertise. To address this challenge, we introduce FAMOSE (Feature AugMentation and Optimal Selection agEnt), a novel framework that leverages the ReAct paradigm to autonomously explore, generate, and refine features while integrating feature selection and evaluation tools within an agent architecture. To our knowledge, FAMOSE represents the first application of an agentic ReAct framework to automated feature engineering, especially for both regression and classification tasks. Extensive experiments demonstrate that FAMOSE is at or near the state-of-the-art on classification tasks (especially tasks with more than 10K instances, where ROC-AUC increases 0.23% on average), and achieves the state-of-the-art for regression tasks by reducing RMSE by 2.0% on average, while remaining more robust to errors than other algorithms. We hypothesize that FAMOSE's strong performance is because ReAct allows the LLM context window to record (via iterative feature discovery and evaluation steps) what features did or did not work. This is similar to a few-shot prompt and guides the LLM to invent better, more innovative features. Our work offers evidence that AI agents are remarkably effective in solving problems that require highly inventive solutions, such as feature engineering.", "date": "2026-02-19", "categories": []}
{"id": "2602.16435", "title": "Causally-Guided Automated Feature Engineering with Multi-Agent Reinforcement Learning", "abstract": "Automated feature engineering (AFE) enables AI systems to autonomously construct high-utility representations from raw tabular data. However, existing AFE methods rely on statistical heuristics, yielding brittle features that fail under distribution shift. We introduce CAFE, a framework that reformulates AFE as a causally-guided sequential decision process, bridging causal discovery with reinforcement learning-driven feature construction. Phase I learns a sparse directed acyclic graph over features and the target to obtain soft causal priors, grouping features as direct, indirect, or other based on their causal influence with respect to the target. Phase II uses a cascading multi-agent deep Q-learning architecture to select causal groups and transformation operators, with hierarchical reward shaping and causal group-level exploration strategies that favor causally plausible transformations while controlling feature complexity. Across 15 public benchmarks (classification with macro-F1; regression with inverse relative absolute error), CAFE achieves up to 7% improvement over strong AFE baselines, reduces episodes-to-convergence, and delivers competitive time-to-target. Under controlled covariate shifts, CAFE reduces performance drop by ~4x relative to a non-causal multi-agent baseline, and produces more compact feature sets with more stable post-hoc attributions. These findings underscore that causal structure, used as a soft inductive prior rather than a rigid constraint, can substantially improve the robustness and efficiency of automated feature engineering.", "date": "2026-02-18", "categories": []}
{"id": "2510.26510", "title": "LLMs as In-Context Meta-Learners for Model and Hyperparameter Selection", "abstract": "Model and hyperparameter selection are critical but challenging in machine learning, typically requiring expert intuition or expensive automated search. We investigate whether large language models (LLMs) can act as in-context meta-learners for this task. By converting each dataset into interpretable metadata, we prompt an LLM to recommend both model families and hyperparameters. We study two prompting strategies: (1) a zero-shot mode relying solely on pretrained knowledge, and (2) a meta-informed mode augmented with examples of models and their performance on past tasks. Across synthetic and real-world benchmarks, we show that LLMs can exploit dataset metadata to recommend competitive models and hyperparameters without search, and that improvements from meta-informed prompting demonstrate their capacity for in-context meta-learning. These results highlight a promising new role for LLMs as lightweight, general-purpose assistants for model selection and hyperparameter optimization.", "date": "2025-10-30", "categories": []}
{"id": "2408.06717", "title": "Proficient Graph Neural Network Design by Accumulating Knowledge on Large Language Models", "abstract": "High-level automation is increasingly critical in AI, driven by rapid advances in large language models (LLMs) and AI agents. However, LLMs, despite their general reasoning power, struggle significantly in specialized, data-sensitive tasks such as designing Graph Neural Networks (GNNs). This difficulty arises from (1) the inherent knowledge gaps in modeling the intricate, varying relationships between graph properties and suitable architectures and (2) the external noise from misleading descriptive inputs, often resulting in generic or even misleading model suggestions. Achieving proficiency in designing data-aware models -- defined as the meta-level capability to systematically accumulate, interpret, and apply data-specific design knowledge -- remains challenging for existing automated approaches, due to their inefficient construction and application of meta-knowledge. To achieve meta-level proficiency, we propose DesiGNN, a knowledge-centered framework that systematically converts past model design experience into structured, fine-grained knowledge priors well-suited for meta-learning with LLMs. To account for the inherent variability and external noise, DesiGNN aligns empirical property filtering from extensive benchmarks with adaptive elicitation of literature insights via LLMs. By constructing a solid meta-knowledge between unseen graph understanding and known effective architecture patterns, DesiGNN can deliver top-5.77% initial model proposals for unseen datasets within seconds and achieve consistently superior performance with minimal search cost compared to baselines.", "date": "2024-08-13", "categories": []}
{"id": "2602.10226", "title": "Self-Evolving Recommendation System: End-To-End Autonomous Model Optimization With LLM Agents", "abstract": "Optimizing large-scale machine learning systems, such as recommendation models for global video platforms, requires navigating a massive hyperparameter search space and, more critically, designing sophisticated optimizers, architectures, and reward functions to capture nuanced user behaviors. Achieving substantial improvements in these areas is a non-trivial task, traditionally relying on extensive manual iterations to test new hypotheses. We propose a self-evolving system that leverages Large Language Models (LLMs), specifically those from Google's Gemini family, to autonomously generate, train, and deploy high-performing, complex model changes within an end-to-end automated workflow. The self-evolving system is comprised of an Offline Agent (Inner Loop) that performs high-throughput hypothesis generation using proxy metrics, and an Online Agent (Outer Loop) that validates candidates against delayed north star business metrics in live production. Our agents act as specialized Machine Learning Engineers (MLEs): they exhibit deep reasoning capabilities, discovering novel improvements in optimization algorithms and model architecture, and formulating innovative reward functions that target long-term user engagement. The effectiveness of this approach is demonstrated through several successful production launches at YouTube, confirming that autonomous, LLM-driven evolution can surpass traditional engineering workflows in both development velocity and model performance.", "date": "2026-02-10", "categories": []}
{"id": "2602.08774", "title": "Default Machine Learning Hyperparameters Do Not Provide Informative Initialization for Bayesian Optimization", "abstract": "Bayesian Optimization (BO) is a standard tool for hyperparameter tuning thanks to its sample efficiency on expensive black-box functions. While most BO pipelines begin with uniform random initialization, default hyperparameter values shipped with popular ML libraries such as scikit-learn encode implicit expert knowledge and could serve as informative starting points that accelerate convergence. This hypothesis, despite its intuitive appeal, has remained largely unexamined. We formalize the idea by initializing BO with points drawn from truncated Gaussian distributions centered at library defaults and compare the resulting trajectories against a uniform-random baseline. We conduct an extensive empirical evaluation spanning three BO back-ends (BoTorch, Optuna, Scikit-Optimize), three model families (Random Forests, Support Vector Machines, Multilayer Perceptrons), and five benchmark datasets covering classification and regression tasks. Performance is assessed through convergence speed and final predictive quality, and statistical significance is determined via one-sided binomial tests. Across all conditions, default-informed initialization yields no statistically significant advantage over purely random sampling, with p-values ranging from 0.141 to 0.908. A sensitivity analysis on the prior variance confirms that, while tighter concentration around the defaults improves early evaluations, this transient benefit vanishes as optimization progresses, leaving final performance unchanged. Our results provide no evidence that default hyperparameters encode useful directional information for optimization. We therefore recommend that practitioners treat hyperparameter tuning as an integral part of model development and favor principled, data-driven search strategies over heuristic reliance on library defaults.", "date": "2026-02-09", "categories": []}
{"id": "2602.00412", "title": "Robustness of AutoML on Dirty Categorical Data", "abstract": "The goal of automated machine learning (AutoML) is to reduce trial and error when doing machine learning (ML). Although AutoML methods for classification are able to deal with data imperfections, such as outliers, multiple scales and missing data, their behavior is less known on dirty categorical datasets. These datasets often have several categorical features with high cardinality arising from issues such as lack of curation and automated collection. Recent research has shown that ML models can benefit from morphological encoders for dirty categorical data, leading to significantly superior predictive performance. However the effects of using such encoders in AutoML methods are not known at the moment. In this paper, we propose a pipeline that transforms categorical data into numerical data so that an AutoML can handle categorical data transformed by more advanced encoding schemes. We benchmark the current robustness of AutoML methods on a set of dirty datasets and compare it with the proposed pipeline. This allows us to get insight on differences in predictive performance. We also look at the ML pipelines built by AutoMLs in order to gain insight beyond the best model as typically returned by these methods.", "date": "2026-01-31", "categories": []}
{"id": "2602.01078", "title": "AutoHealth: An Uncertainty-Aware Multi-Agent System for Autonomous Health Data Modeling", "abstract": "LLM-based agents have demonstrated strong potential for autonomous machine learning, yet their applicability to health data remains limited. Existing systems often struggle to generalize across heterogeneous health data modalities, rely heavily on predefined solution templates with insufficient adaptation to task-specific objectives, and largely overlook uncertainty estimation, which is essential for reliable decision-making in healthcare. To address these challenges, we propose \\textit{AutoHealth}, a novel uncertainty-aware multi-agent system that autonomously models health data and assesses model reliability. \\textit{AutoHealth} employs closed-loop coordination among five specialized agents to perform data exploration, task-conditioned model construction, training, and optimization, while jointly prioritizing predictive performance and uncertainty quantification. Beyond producing ready-to-use models, the system generates comprehensive reports to support trustworthy interpretation and risk-aware decision-making. To rigorously evaluate its effectiveness, we curate a challenging real-world benchmark comprising 17 tasks across diverse data modalities and learning settings. \\textit{AutoHealth} completes all tasks and outperforms state-of-the-art baselines by 29.2\\% in prediction performance and 50.2\\% in uncertainty estimation.", "date": "2026-02-01", "categories": []}
{"id": "2502.14693", "title": "I-MCTS: Enhancing Agentic AutoML via Introspective Monte Carlo Tree Search", "abstract": "Recent advancements in large language models (LLMs) have shown remarkable potential in automating machine learning tasks. However, existing LLM-based agents often struggle with low-diversity and suboptimal code generation. While recent work has introduced Monte Carlo Tree Search (MCTS) to address these issues, limitations persist in the quality and diversity of thoughts generated, as well as in the scalar value feedback mechanisms used for node selection. In this study, we introduce Introspective Monte Carlo Tree Search (I-MCTS), a novel approach that iteratively expands tree nodes through an introspective process that meticulously analyzes solutions and results from parent and sibling nodes. This facilitates a continuous refinement of the node in the search tree, thereby enhancing the overall decision-making process. Furthermore, we integrate a Large Language Model (LLM)-based value model to facilitate direct evaluation of each node's solution prior to conducting comprehensive computational rollouts. A hybrid rewarding mechanism is implemented to seamlessly transition the Q-value from LLM-estimated scores to actual performance scores. This allows higher-quality nodes to be traversed earlier. Applied to the various ML tasks, our approach demonstrates a 4% absolute improvement in performance compared to the strong open-source AutoML agents, showcasing its effectiveness in enhancing agentic AutoML systems. Resource available at https://github.com/jokieleung/I-MCTS", "date": "2025-02-20", "categories": []}
{"id": "2507.21184", "title": "Can Language Models Discover Scaling Laws?", "abstract": "Discovering scaling laws for predicting model performance at scale is a fundamental and open-ended challenge, mostly reliant on slow, case specific human experimentation. To investigate the potential for LLMs to automate this process, we collect over 5,000 experiments from existing literature and curate eight diverse scaling law discovery tasks. While existing agents struggle to produce accurate law formulas, this paper introduces SLDAgent, an evolution-based agent that co-optimize the scaling law model and the parameters, enabling it to autonomously explore complex relationships between variables. For the first time, we demonstrates that SLDAgent can automatically discover laws that exhibit consistently more accurate extrapolation than their established, human-derived counterparts across all tasks. Through comprehensive analysis, we elucidate why these discovered laws are superior and verify their practical utility in both pretraining and finetuning applications. This work establishes a new paradigm for agentic scientific discovery, showing that AI systems can understand their own scaling behavior, and can contribute novel and practical knowledge back to the research community.", "date": "2025-07-27", "categories": []}
{"id": "2601.14525", "title": "Towards Execution-Grounded Automated AI Research", "abstract": "Automated AI research holds great potential to accelerate scientific discovery. However, current LLMs often generate plausible-looking but ineffective ideas. Execution grounding may help, but it is unclear whether automated execution is feasible and whether LLMs can learn from the execution feedback. To investigate these, we first build an automated executor to implement ideas and launch large-scale parallel GPU experiments to verify their effectiveness. We then convert two realistic research problems - LLM pre-training and post-training - into execution environments and demonstrate that our automated executor can implement a large fraction of the ideas sampled from frontier LLMs. We analyze two methods to learn from the execution feedback: evolutionary search and reinforcement learning. Execution-guided evolutionary search is sample-efficient: it finds a method that significantly outperforms the GRPO baseline (69.4% vs 48.0%) on post-training, and finds a pre-training recipe that outperforms the nanoGPT baseline (19.7 minutes vs 35.9 minutes) on pre-training, all within just ten search epochs. Frontier LLMs often generate meaningful algorithmic ideas during search, but they tend to saturate early and only occasionally exhibit scaling trends. Reinforcement learning from execution reward, on the other hand, suffers from mode collapse. It successfully improves the average reward of the ideator model but not the upper-bound, due to models converging on simple ideas. We thoroughly analyze the executed ideas and training dynamics to facilitate future efforts towards execution-grounded automated AI research.", "date": "2026-01-20", "categories": []}
{"id": "2601.11389", "title": "Hyperparameter Optimization of Constraint Programming Solvers", "abstract": "The performance of constraint programming solvers is highly sensitive to the choice of their hyperparameters. Manually finding the best solver configuration is a difficult, time-consuming task that typically requires expert knowledge. In this paper, we introduce probe and solve algorithm, a novel two-phase framework for automated hyperparameter optimization integrated into the CPMpy library. This approach partitions the available time budget into two phases: a probing phase that explores different sets of hyperparameters using configurable hyperparameter optimization methods, followed by a solving phase where the best configuration found is used to tackle the problem within the remaining time.\n  We implement and compare two hyperparameter optimization methods within the probe and solve algorithm: Bayesian optimization and Hamming distance search. We evaluate the algorithm on two different constraint programming solvers, ACE and Choco, across 114 combinatorial problem instances, comparing their performance against the solver's default configurations.\n  Results show that using Bayesian optimization, the algorithm outperforms the solver's default configurations, improving solution quality for ACE in 25.4% of instances and matching the default performance in 57.9%, and for Choco, achieving superior results in 38.6% of instances. It also consistently surpasses Hamming distance search within the same framework, confirming the advantage of model-based exploration over simple local search. Overall, the probe and solve algorithm offers a practical, resource-aware approach for tuning constraint solvers that yields robust improvements across diverse problem types.", "date": "2026-01-16", "categories": []}
{"id": "2509.21138", "title": "AutoIntent: AutoML for Text Classification", "abstract": "AutoIntent is an automated machine learning tool for text classification tasks. Unlike existing solutions, AutoIntent offers end-to-end automation with embedding model selection, classifier optimization, and decision threshold tuning, all within a modular, sklearn-like interface. The framework is designed to support multi-label classification and out-of-scope detection. AutoIntent demonstrates superior performance compared to existing AutoML tools on standard intent classification datasets and enables users to balance effectiveness and resource consumption.", "date": "2025-09-25", "categories": []}
{"id": "2503.23314", "title": "SPIO: Ensemble and Selective Strategies via LLM-Based Multi-Agent Planning in Automated Data Science", "abstract": "Large Language Models (LLMs) have enabled dynamic reasoning in automated data analytics, yet recent multi-agent systems remain limited by rigid, single-path workflows that restrict strategic exploration and often lead to suboptimal outcomes. To overcome these limitations, we propose SPIO (Sequential Plan Integration and Optimization), a framework that replaces rigid workflows with adaptive, multi-path planning across four core modules: data preprocessing, feature engineering, model selection, and hyperparameter tuning. In each module, specialized agents generate diverse candidate strategies, which are cascaded and refined by an optimization agent. SPIO offers two operating modes: SPIO-S for selecting a single optimal pipeline, and SPIO-E for ensembling top-k pipelines to maximize robustness. Extensive evaluations on Kaggle and OpenML benchmarks show that SPIO consistently outperforms state-of-the-art baselines, achieving an average performance gain of 5.6%. By explicitly exploring and integrating multiple solution paths, SPIO delivers a more flexible, accurate, and reliable foundation for automated data science.", "date": "2025-03-30", "categories": []}
{"id": "2512.19246", "title": "From Black-Box Tuning to Guided Optimization via Hyperparameters Interaction Analysis", "abstract": "Hyperparameters tuning is a fundamental, yet computationally expensive, step in optimizing machine learning models. Beyond optimization, understanding the relative importance and interaction of hyperparameters is critical to efficient model development. In this paper, we introduce MetaSHAP, a scalable semi-automated eXplainable AI (XAI) method, that uses meta-learning and Shapley values analysis to provide actionable and dataset-aware tuning insights. MetaSHAP operates over a vast benchmark of over 09 millions evaluated machine learning pipelines, allowing it to produce interpretable importance scores and actionable tuning insights that reveal how much each hyperparameter matters, how it interacts with others and in which value ranges its influence is concentrated. For a given algorithm and dataset, MetaSHAP learns a surrogate performance model from historical configurations, computes hyperparameters interactions using SHAP-based analysis, and derives interpretable tuning ranges from the most influential hyperparameters. This allows practitioners not only to prioritize which hyperparameters to tune, but also to understand their directionality and interactions. We empirically validate MetaSHAP on a diverse benchmark of 164 classification datasets and 14 classifiers, demonstrating that it produces reliable importance rankings and competitive performance when used to guide Bayesian optimization.", "date": "2025-12-22", "categories": []}
{"id": "2512.17517", "title": "PathBench-MIL: A Comprehensive AutoML and Benchmarking Framework for Multiple Instance Learning in Histopathology", "abstract": "We introduce PathBench-MIL, an open-source AutoML and benchmarking framework for multiple instance learning (MIL) in histopathology. The system automates end-to-end MIL pipeline construction, including preprocessing, feature extraction, and MIL-aggregation, and provides reproducible benchmarking of dozens of MIL models and feature extractors. PathBench-MIL integrates visualization tooling, a unified configuration system, and modular extensibility, enabling rapid experimentation and standardization across datasets and tasks. PathBench-MIL is publicly available at https://github.com/Sbrussee/PathBench-MIL", "date": "2025-12-19", "categories": []}
{"id": "2404.16560", "title": "Automated Model Selection for Generalized Linear Models", "abstract": "In this paper, we show how mixed-integer conic optimization can be used to combine feature subset selection with holistic generalized linear models to fully automate the model selection process. Concretely, we directly optimize for the Akaike and Bayesian information criteria while imposing constraints designed to deal with multicollinearity in the feature selection task. Specifically, we propose a novel pairwise correlation constraint that combines the sign coherence constraint with ideas from classical statistical models like Ridge regression and the OSCAR model.", "date": "2024-04-25", "categories": []}
{"id": "2410.09596", "title": "Mastering AI: Big Data, Deep Learning, and the Evolution of Large Language Models -- AutoML from Basics to State-of-the-Art Techniques", "abstract": "A comprehensive guide to Automated Machine Learning (AutoML) is presented, covering fundamental principles, practical implementations, and future trends. The paper is structured to assist both beginners and experienced practitioners, with detailed discussions on popular AutoML tools such as TPOT, AutoGluon, and Auto-Keras. Emerging topics like Neural Architecture Search (NAS) and AutoML's applications in deep learning are also addressed. It is anticipated that this work will contribute to ongoing research and development in the field of AI and machine learning.", "date": "2024-10-12", "categories": []}
{"id": "2512.01810", "title": "DeepCAVE: A Visualization and Analysis Tool for Automated Machine Learning", "abstract": "Hyperparameter optimization (HPO), as a central paradigm of AutoML, is crucial for leveraging the full potential of machine learning (ML) models; yet its complexity poses challenges in understanding and debugging the optimization process. We present DeepCAVE, a tool for interactive visualization and analysis, providing insights into HPO. Through an interactive dashboard, researchers, data scientists, and ML engineers can explore various aspects of the HPO process and identify issues, untouched potentials, and new insights about the ML model being tuned. By empowering users with actionable insights, DeepCAVE contributes to the interpretability of HPO and ML on a design level and aims to foster the development of more robust and efficient methodologies in the future.", "date": "2025-12-01", "categories": []}
{"id": "2508.04231", "title": "Empowering Time Series Forecasting with LLM-Agents", "abstract": "Large Language Model (LLM) powered agents have emerged as effective planners for Automated Machine Learning (AutoML) systems. While most existing AutoML approaches focus on automating feature engineering and model architecture search, recent studies in time series forecasting suggest that lightweight models can often achieve state-of-the-art performance. This observation led us to explore improving data quality, rather than model architecture, as a potentially fruitful direction for AutoML on time series data. We propose DCATS, a Data-Centric Agent for Time Series. DCATS leverages metadata accompanying time series to clean data while optimizing forecasting performance. We evaluated DCATS using four time series forecasting models on a large-scale traffic volume forecasting dataset. Results demonstrate that DCATS achieves an average 6% error reduction across all tested models and time horizons, highlighting the potential of data-centric approaches in AutoML for time series forecasting.", "date": "2025-08-06", "categories": []}
{"id": "2511.18835", "title": "Auto-ML Graph Neural Network Hypermodels for Outcome Prediction in Event-Sequence Data", "abstract": "This paper introduces HGNN(O), an AutoML GNN hypermodel framework for outcome prediction on event-sequence data. Building on our earlier work on graph convolutional network hypermodels, HGNN(O) extends four architectures-One Level, Two Level, Two Level Pseudo Embedding, and Two Level Embedding-across six canonical GNN operators. A self-tuning mechanism based on Bayesian optimization with pruning and early stopping enables efficient adaptation over architectures and hyperparameters without manual configuration. Empirical evaluation on both balanced and imbalanced event logs shows that HGNN(O) achieves accuracy exceeding 0.98 on the Traffic Fines dataset and weighted F1 scores up to 0.86 on the Patients dataset without explicit imbalance handling. These results demonstrate that the proposed AutoML-GNN approach provides a robust and generalizable benchmark for outcome prediction in complex event-sequence data.", "date": "2025-11-24", "categories": []}
{"id": "2502.21025", "title": "AutoQML: A Framework for Automated Quantum Machine Learning", "abstract": "Automated Machine Learning (AutoML) has significantly advanced the efficiency of ML-focused software development by automating hyperparameter optimization and pipeline construction, reducing the need for manual intervention. Quantum Machine Learning (QML) offers the potential to surpass classical machine learning (ML) capabilities by utilizing quantum computing. However, the complexity of QML presents substantial entry barriers. We introduce \\emph{AutoQML}, a novel framework that adapts the AutoML approach to QML, providing a modular and unified programming interface to facilitate the development of QML pipelines. AutoQML leverages the QML library sQUlearn to support a variety of QML algorithms. The framework is capable of constructing end-to-end pipelines for supervised learning tasks, ensuring accessibility and efficacy. We evaluate AutoQML across four industrial use cases, demonstrating its ability to generate high-performing QML pipelines that are competitive with both classical ML models and manually crafted quantum solutions.", "date": "2025-02-28", "categories": []}
{"id": "2505.05226", "title": "Put CASH on Bandits: A Max K-Armed Problem for Automated Machine Learning", "abstract": "The Combined Algorithm Selection and Hyperparameter optimization (CASH) is a challenging resource allocation problem in the field of AutoML. We propose MaxUCB, a max k-armed bandit method to trade off exploring different model classes and conducting hyperparameter optimization. MaxUCB is specifically designed for the light-tailed and bounded reward distributions arising in this setting and, thus, provides an efficient alternative compared to classic max k-armed bandit methods assuming heavy-tailed reward distributions. We theoretically and empirically evaluate our method on four standard AutoML benchmarks, demonstrating superior performance over prior approaches. We make our code and data available at https://github.com/amirbalef/CASH_with_Bandits", "date": "2025-05-08", "categories": []}
{"id": "2511.08371", "title": "Multi-objective Hyperparameter Optimization in the Age of Deep Learning", "abstract": "While Deep Learning (DL) experts often have prior knowledge about which hyperparameter settings yield strong performance, only few Hyperparameter Optimization (HPO) algorithms can leverage such prior knowledge and none incorporate priors over multiple objectives. As DL practitioners often need to optimize not just one but many objectives, this is a blind spot in the algorithmic landscape of HPO. To address this shortcoming, we introduce PriMO, the first HPO algorithm that can integrate multi-objective user beliefs. We show PriMO achieves state-of-the-art performance across 8 DL benchmarks in the multi-objective and single-objective setting, clearly positioning itself as the new go-to HPO algorithm for DL practitioners.", "date": "2025-11-11", "categories": []}
{"id": "2511.08491", "title": "Toward Autonomous and Efficient Cybersecurity: A Multi-Objective AutoML-based Intrusion Detection System", "abstract": "With increasingly sophisticated cybersecurity threats and rising demand for network automation, autonomous cybersecurity mechanisms are becoming critical for securing modern networks. The rapid expansion of Internet of Things (IoT) systems amplifies these challenges, as resource-constrained IoT devices demand scalable and efficient security solutions. In this work, an innovative Intrusion Detection System (IDS) utilizing Automated Machine Learning (AutoML) and Multi-Objective Optimization (MOO) is proposed for autonomous and optimized cyber-attack detection in modern networking environments. The proposed IDS framework integrates two primary innovative techniques: Optimized Importance and Percentage-based Automated Feature Selection (OIP-AutoFS) and Optimized Performance, Confidence, and Efficiency-based Combined Algorithm Selection and Hyperparameter Optimization (OPCE-CASH). These components optimize feature selection and model learning processes to strike a balance between intrusion detection effectiveness and computational efficiency. This work presents the first IDS framework that integrates all four AutoML stages and employs multi-objective optimization to jointly optimize detection effectiveness, efficiency, and confidence for deployment in resource-constrained systems. Experimental evaluations over two benchmark cybersecurity datasets demonstrate that the proposed MOO-AutoML IDS outperforms state-of-the-art IDSs, establishing a new benchmark for autonomous, efficient, and optimized security for networks. Designed to support IoT and edge environments with resource constraints, the proposed framework is applicable to a variety of autonomous cybersecurity applications across diverse networked environments.", "date": "2025-11-11", "categories": []}
{"id": "2511.03985", "title": "ArchPilot: A Proxy-Guided Multi-Agent Approach for Machine Learning Engineering", "abstract": "Recent LLM-based agents have demonstrated strong capabilities in automated ML engineering. However, they heavily rely on repeated full training runs to evaluate candidate solutions, resulting in significant computational overhead, limited scalability to large search spaces, and slow iteration cycles. To address these challenges, we introduce ArchPilot, a multi-agent system that integrates architecture generation, proxy-based evaluation, and adaptive search into a unified framework. ArchPilot consists of three specialized agents: an orchestration agent that coordinates the search process using a Monte Carlo Tree Search (MCTS)-inspired novel algorithm with a restart mechanism and manages memory of previous candidates; a generation agent that iteratively generates, improves, and debugs candidate architectures; and an evaluation agent that executes proxy training runs, generates and optimizes proxy functions, and aggregates the proxy scores into a fidelity-aware performance metric. This multi-agent collaboration allows ArchPilot to prioritize high-potential candidates with minimal reliance on expensive full training runs, facilitating efficient ML engineering under limited budgets. Experiments on MLE-Bench demonstrate that ArchPilot outperforms SOTA baselines such as AIDE and ML-Master, validating the effectiveness of our multi-agent system.", "date": "2025-11-06", "categories": []}
{"id": "2504.04365", "title": "AutoPDL: Automatic Prompt Optimization for LLM Agents", "abstract": "The performance of large language models (LLMs) depends on how they are prompted, with choices spanning both the high-level prompting pattern (e.g., Zero-Shot, CoT, ReAct, ReWOO) and the specific prompt content (instructions and few-shot demonstrations). Manually tuning this combination is tedious, error-prone, and specific to a given LLM and task. Therefore, this paper proposes AutoPDL, an automated approach to discovering good LLM agent configurations. Our approach frames this as a structured AutoML problem over a combinatorial space of agentic and non-agentic prompting patterns and demonstrations, using successive halving to efficiently navigate this space. We introduce a library implementing common prompting patterns using the PDL prompt programming language. AutoPDL solutions are human-readable, editable, and executable PDL programs that use this library. This approach also enables source-to-source optimization, allowing human-in-the-loop refinement and reuse. Evaluations across three tasks and seven LLMs (ranging from 3B to 70B parameters) show consistent accuracy gains ($9.21\\pm15.46$ percentage points), up to 67.5pp, and reveal that selected prompting strategies vary across models and tasks.", "date": "2025-04-06", "categories": []}
{"id": "2507.02554", "title": "AI Research Agents for Machine Learning: Search, Exploration, and Generalization in MLE-bench", "abstract": "AI research agents are demonstrating great potential to accelerate scientific progress by automating the design, implementation, and training of machine learning models. We focus on methods for improving agents' performance on MLE-bench, a challenging benchmark where agents compete in Kaggle competitions to solve real-world machine learning problems. We formalize AI research agents as search policies that navigate a space of candidate solutions, iteratively modifying them using operators. By designing and systematically varying different operator sets and search policies (Greedy, MCTS, Evolutionary), we show that their interplay is critical for achieving high performance. Our best pairing of search strategy and operator set achieves a state-of-the-art result on MLE-bench lite, increasing the success rate of achieving a Kaggle medal from 39.6% to 47.7%. Our investigation underscores the importance of jointly considering the search strategy, operator design, and evaluation methodology in advancing automated machine learning.", "date": "2025-07-03", "categories": []}
{"id": "2508.20519", "title": "Khiops: An End-to-End, Frugal AutoML and XAI Machine Learning Solution for Large, Multi-Table Databases", "abstract": "Khiops is an open source machine learning tool designed for mining large multi-table databases. Khiops is based on a unique Bayesian approach that has attracted academic interest with more than 20 publications on topics such as variable selection, classification, decision trees and co-clustering. It provides a predictive measure of variable importance using discretisation models for numerical data and value clustering for categorical data. The proposed classification/regression model is a naive Bayesian classifier incorporating variable selection and weight learning. In the case of multi-table databases, it provides propositionalisation by automatically constructing aggregates. Khiops is adapted to the analysis of large databases with millions of individuals, tens of thousands of variables and hundreds of millions of records in secondary tables. It is available on many environments, both from a Python library and via a user interface.", "date": "2025-08-28", "categories": []}
{"id": "2510.24339", "title": "VDSAgents: A PCS-Guided Multi-Agent System for Veridical Data Science Automation", "abstract": "Large language models (LLMs) become increasingly integrated into data science workflows for automated system design. However, these LLM-driven data science systems rely solely on the internal reasoning of LLMs, lacking guidance from scientific and theoretical principles. This limits their trustworthiness and robustness, especially when dealing with noisy and complex real-world datasets. This paper provides VDSAgents, a multi-agent system grounded in the Predictability-Computability-Stability (PCS) principles proposed in the Veridical Data Science (VDS) framework. Guided by PCS principles, the system implements a modular workflow for data cleaning, feature engineering, modeling, and evaluation. Each phase is handled by an elegant agent, incorporating perturbation analysis, unit testing, and model validation to ensure both functionality and scientific auditability. We evaluate VDSAgents on nine datasets with diverse characteristics, comparing it with state-of-the-art end-to-end data science systems, such as AutoKaggle and DataInterpreter, using DeepSeek-V3 and GPT-4o as backends. VDSAgents consistently outperforms the results of AutoKaggle and DataInterpreter, which validates the feasibility of embedding PCS principles into LLM-driven data science automation.", "date": "2025-10-28", "categories": []}
{"id": "2510.23535", "title": "Sequential Multi-Agent Dynamic Algorithm Configuration", "abstract": "Dynamic algorithm configuration (DAC) is a recent trend in automated machine learning, which can dynamically adjust the algorithm's configuration during the execution process and relieve users from tedious trial-and-error tuning tasks. Recently, multi-agent reinforcement learning (MARL) approaches have improved the configuration of multiple heterogeneous hyperparameters, making various parameter configurations for complex algorithms possible. However, many complex algorithms have inherent inter-dependencies among multiple parameters (e.g., determining the operator type first and then the operator's parameter), which are, however, not considered in previous approaches, thus leading to sub-optimal results. In this paper, we propose the sequential multi-agent DAC (Seq-MADAC) framework to address this issue by considering the inherent inter-dependencies of multiple parameters. Specifically, we propose a sequential advantage decomposition network, which can leverage action-order information through sequential advantage decomposition. Experiments from synthetic functions to the configuration of multi-objective optimization algorithms demonstrate Seq-MADAC's superior performance over state-of-the-art MARL methods and show strong generalization across problem classes. Seq-MADAC establishes a new paradigm for the widespread dependency-aware automated algorithm configuration. Our code is available at https://github.com/lamda-bbo/seq-madac.", "date": "2025-10-27", "categories": []}
{"id": "2510.17845", "title": "MAT-Agent: Adaptive Multi-Agent Training Optimization", "abstract": "Multi-label image classification demands adaptive training strategies to navigate complex, evolving visual-semantic landscapes, yet conventional methods rely on static configurations that falter in dynamic settings. We propose MAT-Agent, a novel multi-agent framework that reimagines training as a collaborative, real-time optimization process. By deploying autonomous agents to dynamically tune data augmentation, optimizers, learning rates, and loss functions, MAT-Agent leverages non-stationary multi-armed bandit algorithms to balance exploration and exploitation, guided by a composite reward harmonizing accuracy, rare-class performance, and training stability. Enhanced with dual-rate exponential moving average smoothing and mixed-precision training, it ensures robustness and efficiency. Extensive experiments across Pascal VOC, COCO, and VG-256 demonstrate MAT-Agent's superiority: it achieves an mAP of 97.4 (vs. 96.2 for PAT-T), OF1 of 92.3, and CF1 of 91.4 on Pascal VOC; an mAP of 92.8 (vs. 92.0 for HSQ-CvN), OF1 of 88.2, and CF1 of 87.1 on COCO; and an mAP of 60.9, OF1 of 70.8, and CF1 of 61.1 on VG-256. With accelerated convergence and robust cross-domain generalization, MAT-Agent offers a scalable, intelligent solution for optimizing complex visual models, paving the way for adaptive deep learning advancements.", "date": "2025-10-10", "categories": []}
{"id": "2502.12172", "title": "Application-oriented automatic hyperparameter optimization for spiking neural network prototyping", "abstract": "Hyperparameter optimization (HPO) is of paramount importance in the development of high-performance, specialized artificial intelligence (AI) models, ranging from well-established machine learning (ML) solutions to the deep learning (DL) domain and the field of spiking neural networks (SNNs). The latter introduce further complexity due to the neuronal computational units and their additional hyperparameters, whose inadequate setting can dramatically impact the final model performance. At the cost of possible reduced generalization capabilities, the most suitable strategy to fully disclose the power of SNNs is to adopt an application-oriented approach and perform extensive HPO experiments. To facilitate these operations, automatic pipelines are fundamental, and their configuration is crucial. In this document, the Neural Network Intelligence (NNI) toolkit is used as reference framework to present one such solution, with a use case example providing evidence of the corresponding results. In addition, a summary of published works employing the presented pipeline is reported as a potential source of insights into application-oriented HPO experiments for SNN prototyping.", "date": "2025-02-13", "categories": []}
{"id": "2510.07569", "title": "Automated Machine Learning for Unsupervised Tabular Tasks", "abstract": "In this work, we present LOTUS (Learning to Learn with Optimal Transport for Unsupervised Scenarios), a simple yet effective method to perform model selection for multiple unsupervised machine learning(ML) tasks such as outlier detection and clustering. Our intuition behind this work is that a machine learning pipeline will perform well in a new dataset if it previously worked well on datasets with a similar underlying data distribution. We use Optimal Transport distances to find this similarity between unlabeled tabular datasets and recommend machine learning pipelines with one unified single method on two downstream unsupervised tasks: outlier detection and clustering. We present the effectiveness of our approach with experiments against strong baselines and show that LOTUS is a very promising first step toward model selection for multiple unsupervised ML tasks.", "date": "2025-10-08", "categories": []}
{"id": "2510.10232", "title": "SGM: A Statistical Godel Machine for Risk-Controlled Recursive Self-Modification", "abstract": "Recursive self-modification is increasingly central in AutoML, neural architecture search, and adaptive optimization, yet no existing framework ensures that such changes are made safely. Godel machines offer a principled safeguard by requiring formal proofs of improvement before rewriting code; however, such proofs are unattainable in stochastic, high-dimensional settings. We introduce the Statistical Godel Machine (SGM), the first statistical safety layer for recursive edits. SGM replaces proof-based requirements with statistical confidence tests (e-values, Hoeffding bounds), admitting a modification only when superiority is certified at a chosen confidence level, while allocating a global error budget to bound cumulative risk across rounds.We also propose Confirm-Triggered Harmonic Spending (CTHS), which indexes spending by confirmation events rather than rounds, concentrating the error budget on promising edits while preserving familywise validity.Experiments across supervised learning, reinforcement learning, and black-box optimization validate this role: SGM certifies genuine gains on CIFAR-100, rejects spurious improvement on ImageNet-100, and demonstrates robustness on RL and optimization benchmarks.Together, these results position SGM as foundational infrastructure for continual, risk-aware self-modification in learning systems.Code is available at: https://github.com/gravitywavelet/sgm-anon.", "date": "2025-10-11", "categories": []}
{"id": "2510.11143", "title": "Spec-Driven AI for Science: The ARIA Framework for Automated and Reproducible Data Analysis", "abstract": "The rapid expansion of scientific data has widened the gap between analytical capability and research intent. Existing AI-based analysis tools, ranging from AutoML frameworks to agentic research assistants, either favor automation over transparency or depend on manual scripting that hinders scalability and reproducibility. We present ARIA (Automated Research Intelligence Assistant), a spec-driven, human-in-the-loop framework for automated and interpretable data analysis. ARIA integrates six interoperable layers, namely Command, Context, Code, Data, Orchestration, and AI Module, within a document-centric workflow that unifies human reasoning and machine execution. Through natural-language specifications, researchers define analytical goals while ARIA autonomously generates executable code, validates computations, and produces transparent documentation. Beyond achieving high predictive accuracy, ARIA can rapidly identify optimal feature sets and select suitable models, minimizing redundant tuning and repetitive experimentation. In the Boston Housing case, ARIA discovered 25 key features and determined XGBoost as the best performing model (R square = 0.93) with minimal overfitting. Evaluations across heterogeneous domains demonstrate ARIA's strong performance, interpretability, and efficiency compared with state-of-the-art systems. By combining AI for research and AI for science principles within a spec-driven architecture, ARIA establishes a new paradigm for transparent, collaborative, and reproducible scientific discovery.", "date": "2025-10-13", "categories": []}
{"id": "2510.09566", "title": "Automated Evolutionary Optimization for Resource-Efficient Neural Network Training", "abstract": "There are many critical challenges in optimizing neural network models, including distributed computing, compression techniques, and efficient training, regardless of their application to specific tasks. Solving such problems is crucial because the need for scalable and resource-efficient models is increasing. To address these challenges, we have developed a new automated machine learning (AutoML) framework, Parameter Efficient Training with Robust Automation (PETRA). It applies evolutionary optimization to model architecture and training strategy. PETRA includes pruning, quantization, and loss regularization. Experimental studies on real-world data with financial event sequences, as well as image and time-series -- benchmarks, demonstrate PETRA's ability to improve neural model performance and scalability -- namely, a significant decrease in model size (up to 75%) and latency (up to 33%), and an increase in throughput (by 13%) without noticeable degradation in the target metric.", "date": "2025-10-10", "categories": []}
{"id": "2510.08511", "title": "AutoMLGen: Navigating Fine-Grained Optimization for Coding Agents", "abstract": "Large language models (LLMs) have shown impressive performance in general programming tasks. However, in Machine Learning Engineering (MLE) scenarios such as AutoML and Kaggle competitions, achieving high performance depends heavily on expert intervention and repeated adjustments rather than simply generating correct code. When applied directly to these tasks, LLMs often lack fine-grained domain priors, and existing MLE approaches that use linear or tree-structured searches limit knowledge transfer to adjacent hierarchical links. As a result, they cannot leverage past full trajectories or share information across branches, limiting self-evolving ability and search space diversity. To address these limitations, we introduce AutoMLGen, an LLM-based coding agent that integrates a domain knowledge base for high-quality prior guidance and Monte Carlo Graph Search (MCGS) for efficient exploration. MCGS retains the tree-guided exploration of MCTS while embedding a graph structure into the expansion stage to enable dynamic path reorganization, historical trajectory reuse, and multi-solution fusion to support both self-evolution and collaborative learning. Combined with fine-grained operator sets, this design improves stability and accelerates convergence. Evaluation on the MLE-Bench shows that AutoMLGen achieves state-of-the-art performance in numerous dimensions, such as the average medal rate and the valid submission rate, under a 12-hour budget (half the standard runtime). The code is available at https://github.com/Alpha-Innovator/InternAgent.", "date": "2025-10-09", "categories": []}
{"id": "2506.10974", "title": "AutoMind: Adaptive Knowledgeable Agent for Automated Data Science", "abstract": "Large Language Model (LLM) agents have shown great potential in addressing real-world data science problems. LLM-driven data science agents promise to automate the entire machine learning pipeline, yet their real-world effectiveness remains limited. Existing frameworks depend on rigid, pre-defined workflows and inflexible coding strategies; consequently, they excel only on relatively simple, classical problems and fail to capture the empirical expertise that human practitioners bring to complex, innovative tasks. In this work, we introduce AutoMind, an adaptive, knowledgeable LLM-agent framework that overcomes these deficiencies through three key advances: (1) a curated expert knowledge base that grounds the agent in domain expert knowledge, (2) an agentic knowledgeable tree search algorithm that strategically explores possible solutions, and (3) a self-adaptive coding strategy that dynamically tailors code generation to task complexity. Evaluations on two automated data science benchmarks demonstrate that AutoMind delivers superior performance versus state-of-the-art baselines. Additional analyses confirm favorable effectiveness, efficiency, and qualitative solution quality, highlighting AutoMind as an efficient and robust step toward fully automated data science. Code is at https://github.com/innovatingAI/AutoMind.", "date": "2025-06-12", "categories": []}
{"id": "2510.07307", "title": "MLE-Smith: Scaling MLE Tasks with Automated Multi-Agent Pipeline", "abstract": "While Language Models (LMs) have made significant progress in automating machine learning engineering (MLE), the acquisition of high-quality MLE training data is significantly constrained. Current MLE benchmarks suffer from low scalability and limited applicability because they rely on static, manually curated tasks, demanding extensive time and manual effort to produce. We introduce MLE-Smith, a fully automated multi-agent pipeline, to transform raw datasets into competition-style MLE challenges through an efficient generate-verify-execute paradigm for scaling MLE tasks with verifiable quality, real-world usability, and rich diversity. The proposed multi-agent pipeline in MLE-Smith drives structured task design and standardized refactoring, coupled with a hybrid verification mechanism that enforces strict structural rules and high-level semantic soundness. It further validates empirical solvability and real-world fidelity through interactive execution. We apply MLE-Smith to 224 of real-world datasets and generate 606 tasks spanning multiple categories, objectives, and modalities, demonstrating that MLE-Smith can work effectively across a wide range of real-world datasets. Evaluation on the generated tasks shows that the performance of eight mainstream and cutting-edge LLMs on MLE-Smith tasks is strongly correlated with their performance on carefully human-designed tasks, highlighting the effectiveness of the MLE-Smith to scaling up MLE tasks, while maintaining task quality.", "date": "2025-10-08", "categories": []}
{"id": "2509.09387", "title": "MetaLLMix : An XAI Aided LLM-Meta-learning Based Approach for Hyper-parameters Optimization", "abstract": "Effective model and hyperparameter selection remains a major challenge in deep learning, often requiring extensive expertise and computation. While AutoML and large language models (LLMs) promise automation, current LLM-based approaches rely on trial and error and expensive APIs, which provide limited interpretability and generalizability. We propose MetaLLMiX, a zero-shot hyperparameter optimization framework combining meta-learning, explainable AI, and efficient LLM reasoning. By leveraging historical experiment outcomes with SHAP explanations, MetaLLMiX recommends optimal hyperparameters and pretrained models without additional trials. We further employ an LLM-as-judge evaluation to control output format, accuracy, and completeness. Experiments on eight medical imaging datasets using nine open-source lightweight LLMs show that MetaLLMiX achieves competitive or superior performance to traditional HPO methods while drastically reducing computational cost. Our local deployment outperforms prior API-based approaches, achieving optimal results on 5 of 8 tasks, response time reductions of 99.6-99.9%, and the fastest training times on 6 datasets (2.4-15.7x faster), maintaining accuracy within 1-5% of best-performing baselines.", "date": "2025-09-11", "categories": []}
{"id": "2508.00924", "title": "XAutoLM: Efficient Fine-Tuning of Language Models via Meta-Learning and AutoML", "abstract": "Experts in machine learning leverage domain knowledge to navigate decisions in model selection, hyperparameter optimization, and resource allocation. This is particularly critical for fine-tuning language models (LMs), where repeated trials incur substantial computational overhead and environmental impact. However, no existing automated framework simultaneously tackles the entire model selection and hyperparameter optimization (HPO) task for resource-efficient LM fine-tuning. We introduce XAutoLM, a meta-learning-augmented AutoML framework that reuses past experiences to optimize discriminative and generative LM fine-tuning pipelines efficiently. XAutoLM learns from stored successes and failures by extracting task- and system-level meta-features to bias its sampling toward valuable configurations and away from costly dead ends. On four text classification and two question-answering benchmarks, XAutoLM surpasses zero-shot optimizer's peak F1 on five of six tasks, cuts mean evaluation time of pipelines by up to 4.5x, reduces search error ratios by up to sevenfold, and uncovers up to 50% more pipelines above the zero-shot Pareto front. In contrast, simpler memory-based baselines suffer negative transfer. We release XAutoLM and our experience store to catalyze resource-efficient, Green AI fine-tuning in the NLP community.", "date": "2025-07-30", "categories": []}
{"id": "2510.01842", "title": "Pre-Hoc Predictions in AutoML: Leveraging LLMs to Enhance Model Selection and Benchmarking for Tabular datasets", "abstract": "The field of AutoML has made remarkable progress in post-hoc model selection, with libraries capable of automatically identifying the most performing models for a given dataset. Nevertheless, these methods often rely on exhaustive hyperparameter searches, where methods automatically train and test different types of models on the target dataset. Contrastingly, pre-hoc prediction emerges as a promising alternative, capable of bypassing exhaustive search through intelligent pre-selection of models. Despite its potential, pre-hoc prediction remains under-explored in the literature. This paper explores the intersection of AutoML and pre-hoc model selection by leveraging traditional models and Large Language Model (LLM) agents to reduce the search space of AutoML libraries. By relying on dataset descriptions and statistical information, we reduce the AutoML search space. Our methodology is applied to the AWS AutoGluon portfolio dataset, a state-of-the-art AutoML benchmark containing 175 tabular classification datasets available on OpenML. The proposed approach offers a shift in AutoML workflows, significantly reducing computational overhead, while still selecting the best model for the given dataset.", "date": "2025-10-02", "categories": []}
{"id": "2505.14738", "title": "R&D-Agent: An LLM-Agent Framework Towards Autonomous Data Science", "abstract": "Recent advances in AI and ML have transformed data science, yet increasing complexity and expertise requirements continue to hinder progress. Although crowd-sourcing platforms alleviate some challenges, high-level machine learning engineering (MLE) tasks remain labor-intensive and iterative. We introduce R&D-Agent, a comprehensive, decoupled, and extensible framework that formalizes the MLE process. R&D-Agent defines the MLE workflow into two phases and six components, turning agent design for MLE from ad-hoc craftsmanship into a principled, testable process. Although several existing agents report promising gains on their chosen components, they can mostly be summarized as a partial optimization from our framework's simple baseline. Inspired by human experts, we designed efficient and effective agents within this framework that achieve state-of-the-art performance. Evaluated on MLE-Bench, the agent built on R&D-Agent ranks as the top-performing machine learning engineering agent, achieving 35.1% any medal rate, demonstrating the ability of the framework to speed up innovation and improve accuracy across a wide range of data science applications. We have open-sourced R&D-Agent on GitHub: https://github.com/microsoft/RD-Agent.", "date": "2025-05-20", "categories": []}
{"id": "2509.25946", "title": "Automated Model Discovery via Multi-modal & Multi-step Pipeline", "abstract": "Automated model discovery is the process of automatically searching and identifying the most appropriate model for a given dataset over a large combinatorial search space. Existing approaches, however, often face challenges in balancing the capture of fine-grained details with ensuring generalizability beyond training data regimes with a reasonable model complexity. In this paper, we present a multi-modal \\& multi-step pipeline for effective automated model discovery. Our approach leverages two vision-language-based modules (VLM), AnalyzerVLM and EvaluatorVLM, for effective model proposal and evaluation in an agentic way. AnalyzerVLM autonomously plans and executes multi-step analyses to propose effective candidate models. EvaluatorVLM assesses the candidate models both quantitatively and perceptually, regarding the fitness for local details and the generalibility for overall trends. Our results demonstrate that our pipeline effectively discovers models that capture fine details and ensure strong generalizability. Additionally, extensive ablation studies show that both multi-modality and multi-step reasoning play crucial roles in discovering favorable models.", "date": "2025-09-30", "categories": []}
{"id": "2506.02752", "title": "BenLOC: A Benchmark for Learning to Configure MIP Optimizers", "abstract": "The automatic configuration of Mixed-Integer Programming (MIP) optimizers has become increasingly critical as the large number of configurations can significantly affect solver performance. Yet the lack of standardized evaluation frameworks has led to data leakage and over-optimistic claims, as prior studies often rely on homogeneous datasets and inconsistent experimental setups. To promote a fair evaluation process, we present BenLOC, a comprehensive benchmark and open-source toolkit, which not only offers an end-to-end pipeline for learning instance-wise MIP optimizer configurations, but also standardizes dataset selection, train-test splits, feature engineering and baseline choice for unbiased and comprehensive evaluations. Leveraging this framework, we conduct an empirical analysis on five well-established MIP datasets and compare classical machine learning models with handcrafted features against state-of-the-art deep-learning techniques. The results demonstrate the importance of datasets, features and baseline criteria proposed by BenLOC and the effectiveness of BenLOC in providing unbiased and comprehensive evaluations.", "date": "2025-06-03", "categories": []}
{"id": "2509.15561", "title": "Small LLMs with Expert Blocks Are Good Enough for Hyperparamter Tuning", "abstract": "Hyper-parameter Tuning (HPT) is a necessary step in machine learning (ML) pipelines but becomes computationally expensive and opaque with larger models. Recently, Large Language Models (LLMs) have been explored for HPT, yet most rely on models exceeding 100 billion parameters. We propose an Expert Block Framework for HPT using Small LLMs. At its core is the Trajectory Context Summarizer (TCS), a deterministic block that transforms raw training trajectories into structured context, enabling small LLMs to analyze optimization progress with reliability comparable to larger models. Using two locally-run LLMs (phi4:reasoning14B and qwen2.5-coder:32B) and a 10-trial budget, our TCS-enabled HPT pipeline achieves average performance within ~0.9 percentage points of GPT-4 across six diverse tasks.", "date": "2025-09-19", "categories": []}
{"id": "2509.17734", "title": "An AutoML Framework using AutoGluonTS for Forecasting Seasonal Extreme Temperatures", "abstract": "In recent years, great progress has been made in the field of forecasting meteorological variables. Recently, deep learning architectures have made a major breakthrough in forecasting the daily average temperature over a ten-day horizon. However, advances in forecasting events related to the maximum temperature over short horizons remain a challenge for the community. A problem that is even more complex consists in making predictions of the maximum daily temperatures in the short, medium, and long term. In this work, we focus on forecasting events related to the maximum daily temperature over medium-term periods (90 days). Therefore, instead of addressing the problem from a meteorological point of view, this article tackles it from a climatological point of view. Due to the complexity of this problem, a common approach is to frame the study as a temporal classification problem with the classes: maximum temperature \"above normal\", \"normal\" or \"below normal\". From a practical point of view, we created a large historical dataset (from 1981 to 2018) collecting information from weather stations located in South America. In addition, we also integrated exogenous information from the Pacific, Atlantic, and Indian Ocean basins. We applied the AutoGluonTS platform to solve the above-mentioned problem. This AutoML tool shows competitive forecasting performance with respect to large operational platforms dedicated to tackling this climatological problem; but with a \"relatively\" low computational cost in terms of time and resources.", "date": "2025-09-22", "categories": []}
{"id": "2506.06143", "title": "carps: A Framework for Comparing N Hyperparameter Optimizers on M Benchmarks", "abstract": "Hyperparameter Optimization (HPO) is crucial to develop well-performing machine learning models. In order to ease prototyping and benchmarking of HPO methods, we propose carps, a benchmark framework for Comprehensive Automated Research Performance Studies allowing to evaluate N optimizers on M benchmark tasks. In this first release of carps, we focus on the four most important types of HPO task types: blackbox, multi-fidelity, multi-objective and multi-fidelity-multi-objective. With 3 336 tasks from 5 community benchmark collections and 28 variants of 9 optimizer families, we offer the biggest go-to library to date to evaluate and compare HPO methods. The carps framework relies on a purpose-built, lightweight interface, gluing together optimizers and benchmark tasks. It also features an analysis pipeline, facilitating the evaluation of optimizers on benchmarks. However, navigating a huge number of tasks while developing and comparing methods can be computationally infeasible. To address this, we obtain a subset of representative tasks by minimizing the star discrepancy of the subset, in the space spanned by the full set. As a result, we propose an initial subset of 10 to 30 diverse tasks for each task type, and include functionality to re-compute subsets as more benchmarks become available, enabling efficient evaluations. We also establish a first set of baseline results on these tasks as a measure for future comparisons. With carps (https://www.github.com/automl/CARP-S), we make an important step in the standardization of HPO evaluation.", "date": "2025-06-06", "categories": []}
{"id": "2502.18530", "title": "IMPROVE: Iterative Model Pipeline Refinement and Optimization Leveraging LLM Experts", "abstract": "Large language model (LLM) agents have emerged as a promising solution to automate the workflow of machine learning, but most existing methods share a common limitation: they attempt to optimize entire pipelines in a single step before evaluation, making it difficult to attribute improvements to specific changes. This lack of granularity leads to unstable optimization and slower convergence, limiting their effectiveness. To address this, we introduce Iterative Refinement, a novel strategy for LLM-driven ML pipeline design inspired by how human ML experts iteratively refine models, focusing on one component at a time rather than making sweeping changes all at once. By systematically updating individual components based on real training feedback, Iterative Refinement improves overall model performance. We also provide some theoretical edvience of the superior properties of this Iterative Refinement. Further, we implement this strategy in IMPROVE, an end-to-end LLM agent framework for automating and optimizing object classification pipelines. Through extensive evaluations across datasets of varying sizes and domains, we demonstrate that Iterative Refinement enables IMPROVE to consistently achieve better performance over existing zero-shot LLM-based approaches.", "date": "2025-02-25", "categories": []}
{"id": "2509.12019", "title": "AMQ: Enabling AutoML for Mixed-precision Weight-Only Quantization of Large Language Models", "abstract": "To enable broader deployment of Large Language Models (LLMs), it is essential to identify the best-performing model under strict memory constraints. We present AMQ, Automated Mixed-Precision Weight-Only Quantization, a framework that assigns layer-wise quantization bit-widths to optimally balance model quality and memory usage. However, the combinatorial search space, with over 10^{100} possible configurations, makes conventional black-box optimization infeasible. AMQ overcomes this challenge through four key innovations:(1) search space pruning using prior knowledge to exclude unpromising configurations, (2) quantization proxy to bypass costly format conversions during search, (3) quality predictor to minimize evaluation overhead, and (4) iterative search-and-update strategy for fast and stable convergence. By integrating these components, AMQ efficiently explores the quality-efficiency landscape, reaching the Pareto frontier and yielding LLMs that are both compact and high-performing. Our code is available at https://github.com/dlwns147/amq.", "date": "2025-09-15", "categories": []}
{"id": "2509.09321", "title": "Towards Adaptive ML Benchmarks: Web-Agent-Driven Construction, Domain Expansion, and Metric Optimization", "abstract": "Recent advances in large language models (LLMs) have enabled the emergence of general-purpose agents for automating end-to-end machine learning (ML) workflows, including data analysis, feature engineering, model training, and competition solving. However, existing benchmarks remain limited in task coverage, domain diversity, difficulty modeling, and evaluation rigor, failing to capture the full capabilities of such agents in realistic settings. We present TAM Bench, a diverse, realistic, and structured benchmark for evaluating LLM-based agents on end-to-end ML tasks. TAM Bench features three key innovations: (1) A browser automation and LLM-based task acquisition system that automatically collects and structures ML challenges from platforms such as Kaggle, AIcrowd, and Biendata, spanning multiple task types and data modalities (e.g., tabular, text, image, graph, audio); (2) A leaderboard-driven difficulty modeling mechanism that estimates task complexity using participant counts and score dispersion, enabling scalable and objective task calibration; (3) A multi-dimensional evaluation framework incorporating performance, format compliance, constraint adherence, and task generalization. Based on 150 curated AutoML tasks, we construct three benchmark subsets of different sizes -- Lite, Medium, and Full -- designed for varying evaluation scenarios. The Lite version, with 18 tasks and balanced coverage across modalities and difficulty levels, serves as a practical testbed for daily benchmarking and comparative studies.", "date": "2025-09-11", "categories": []}
{"id": "2506.07940", "title": "Gradients: When Markets Meet Fine-tuning -- A Distributed Approach to Model Optimisation", "abstract": "Current AutoML platforms leave substantial performance untapped. Testing 180 fine-tuning tasks across models from 70M to 70B parameters, we found that HuggingFace AutoTrain, TogetherAI, Databricks, and Google Cloud consistently produce suboptimal configurations. Gradients, built on the Bittensor network, attacks this problem through competition. Independent miners race to find optimal hyperparameters, earning rewards proportional to their models' performance. This tournament drives exploration of configuration spaces that single-strategy methods never examine. In our experiments, Gradients achieved a 100\\% win rate against TogetherAI, Databricks, and Google Cloud, and beat HuggingFace AutoTrain in 82.8\\% of experiments. Mean improvements reached 42.1\\% against commercial platforms. Retrieval-augmented generation tasks saw 30-40\\% gains; diffusion models improved 23.4\\% on person-specific generation. When miners compete for rewards, they develop optimization strategies that centralized approaches overlook. These findings demonstrate that decentralized systems with economic incentives can systematically outperform traditional AutoML, suggesting market dynamics may be key to achieving superior fine-tuning results. Code is available at https://github.com/rayonlabs/G.O.D.", "date": "2025-06-09", "categories": []}
{"id": "2503.05227", "title": "MOHPER: Multi-objective Hyperparameter Optimization Framework for E-commerce Retrieval System", "abstract": "E-commerce search optimization has evolved to include a wider range of metrics that reflect user engagement and business objectives. Modern search frameworks now incorporate advanced quality features, such as sales counts and document-query relevance, to better align search results with these goals. Traditional methods typically focus on click-through rate (CTR) as a measure of engagement or relevance, but this can miss true purchase intent, creating a gap between user interest and actual conversions. Joint training with the click-through conversion rate (CTCVR) has become essential for understanding buying behavior, although its sparsity poses challenges for reliable optimization. This study presents MOHPER, a Multi-Objective Hyperparameter Optimization framework for E-commerce Retrieval systems. Utilizing Bayesian optimization and sampling, it jointly optimizes both CTR, CTCVR, and relevant objectives, focusing on engagement and conversion of the users. In addition, to improve the selection of the best configuration from multi-objective optimization, we suggest advanced methods for hyperparameter selection, including a meta-configuration voting strategy and a cumulative training approach that leverages prior optimal configurations, to improve speeds of training and efficiency. Currently deployed in a live setting, our proposed framework substantiates its practical efficacy in achieving a balanced optimization that aligns with both user satisfaction and revenue goals.", "date": "2025-03-07", "categories": []}
{"id": "2506.15692", "title": "MLE-STAR: Machine Learning Engineering Agent via Search and Targeted Refinement", "abstract": "Agents based on large language models (LLMs) for machine learning engineering (MLE) can automatically implement ML models via code generation. However, existing approaches to build such agents often rely heavily on inherent LLM knowledge and employ coarse exploration strategies that modify the entire code structure at once. This limits their ability to select effective task-specific models and perform deep exploration within specific components, such as experimenting extensively with feature engineering options. To overcome these, we propose MLE-STAR, a novel approach to build MLE agents. MLE-STAR first leverages external knowledge by using a search engine to retrieve effective models from the web, forming an initial solution, then iteratively refines it by exploring various strategies targeting specific ML components. This exploration is guided by ablation studies analyzing the impact of individual code blocks. Furthermore, we introduce a novel ensembling method using an effective strategy suggested by MLE-STAR. Our experimental results show that MLE-STAR achieves medals in 64% of the Kaggle competitions on the MLE-bench Lite, significantly outperforming the best alternative.", "date": "2025-05-27", "categories": []}
{"id": "2508.20550", "title": "Theoretical foundations of the integral indicator application in hyperparametric optimization", "abstract": "The article discusses the concept of hyperparametric optimization of recommendation algorithms using an integral assessment that combines various performance indicators into a single consolidated criterion. This approach is opposed to traditional methods of setting up a single metric and allows you to achieve a balance between accuracy, ranking quality, variety of output and the resource intensity of algorithms. The theoretical significance of the research lies in the development of a universal multi-criteria optimization tool that is applicable not only in recommendation systems, but also in a wide range of machine learning and data analysis tasks.", "date": "2025-08-28", "categories": []}
{"id": "2508.20824", "title": "GPT-FT: An Efficient Automated Feature Transformation Using GPT for Sequence Reconstruction and Performance Enhancement", "abstract": "Feature transformation plays a critical role in enhancing machine learning model performance by optimizing data representations. Recent state-of-the-art approaches address this task as a continuous embedding optimization problem, converting discrete search into a learnable process. Although effective, these methods often rely on sequential encoder-decoder structures that cause high computational costs and parameter requirements, limiting scalability and efficiency. To address these limitations, we propose a novel framework that accomplishes automated feature transformation through four steps: transformation records collection, embedding space construction with a revised Generative Pre-trained Transformer (GPT) model, gradient-ascent search, and autoregressive reconstruction. In our approach, the revised GPT model serves two primary functions: (a) feature transformation sequence reconstruction and (b) model performance estimation and enhancement for downstream tasks by constructing the embedding space. Such a multi-objective optimization framework reduces parameter size and accelerates transformation processes. Experimental results on benchmark datasets show that the proposed framework matches or exceeds baseline performance, with significant gains in computational efficiency. This work highlights the potential of transformer-based architectures for scalable, high-performance automated feature transformation.", "date": "2025-08-28", "categories": []}
{"id": "2508.20986", "title": "Graph-Based Feature Augmentation for Predictive Tasks on Relational Datasets", "abstract": "Data has become a foundational asset driving innovation across domains such as finance, healthcare, and e-commerce. In these areas, predictive modeling over relational tables is commonly employed, with increasing emphasis on reducing manual effort through automated machine learning (AutoML) techniques. This raises an interesting question: can feature augmentation itself be automated and identify and utilize task-related relational signals?\n  To address this challenge, we propose an end-to-end automated feature augmentation framework, ReCoGNN, which enhances initial datasets using features extracted from multiple relational tables to support predictive tasks. ReCoGNN first captures semantic dependencies within each table by modeling intra-table attribute relationships, enabling it to partition tables into structured, semantically coherent segments. It then constructs a heterogeneous weighted graph that represents inter-row relationships across all segments. Finally, ReCoGNN leverages message-passing graph neural networks to propagate information through the graph, guiding feature selection and augmenting the original dataset. Extensive experiments conducted on ten real-life and synthetic datasets demonstrate that ReCoGNN consistently outperforms existing methods on both classification and regression tasks.", "date": "2025-08-28", "categories": []}
{"id": "2508.17283", "title": "Quickly Tuning Foundation Models for Image Segmentation", "abstract": "Foundation models like SAM (Segment Anything Model) exhibit strong zero-shot image segmentation performance, but often fall short on domain-specific tasks. Fine-tuning these models typically requires significant manual effort and domain expertise. In this work, we introduce QTT-SEG, a meta-learning-driven approach for automating and accelerating the fine-tuning of SAM for image segmentation. Built on the Quick-Tune hyperparameter optimization framework, QTT-SEG predicts high-performing configurations using meta-learned cost and performance models, efficiently navigating a search space of over 200 million possibilities. We evaluate QTT-SEG on eight binary and five multiclass segmentation datasets under tight time constraints. Our results show that QTT-SEG consistently improves upon SAM's zero-shot performance and surpasses AutoGluon Multimodal, a strong AutoML baseline, on most binary tasks within three minutes. On multiclass datasets, QTT-SEG delivers consistent gains as well. These findings highlight the promise of meta-learning in automating model adaptation for specialized segmentation tasks. Code available at: https://github.com/ds-brx/QTT-SEG/", "date": "2025-08-24", "categories": []}
{"id": "2508.15757", "title": "Language-Guided Tuning: Enhancing Numeric Optimization with Textual Feedback", "abstract": "Configuration optimization remains a critical bottleneck in machine learning, requiring coordinated tuning across model architecture, training strategy, feature engineering, and hyperparameters. Traditional approaches treat these dimensions independently and lack interpretability, while recent automated methods struggle with dynamic adaptability and semantic reasoning about optimization decisions. We introduce Language-Guided Tuning (LGT), a novel framework that employs multi-agent Large Language Models to intelligently optimize configurations through natural language reasoning. We apply textual gradients - qualitative feedback signals that complement numerical optimization by providing semantic understanding of training dynamics and configuration interdependencies. LGT coordinates three specialized agents: an Advisor that proposes configuration changes, an Evaluator that assesses progress, and an Optimizer that refines the decision-making process, creating a self-improving feedback loop. Through comprehensive evaluation on six diverse datasets, LGT demonstrates substantial improvements over traditional optimization methods, achieving performance gains while maintaining high interpretability.", "date": "2025-08-21", "categories": []}
{"id": "2508.13436", "title": "Dynamic Design of Machine Learning Pipelines via Metalearning", "abstract": "Automated machine learning (AutoML) has democratized the design of machine learning based systems, by automating model selection, hyperparameter tuning and feature engineering. However, the high computational cost associated with traditional search and optimization strategies, such as Random Search, Particle Swarm Optimization and Bayesian Optimization, remains a significant challenge. Moreover, AutoML systems typically explore a large search space, which can lead to overfitting. This paper introduces a metalearning method for dynamically designing search spaces for AutoML system. The proposed method uses historical metaknowledge to select promising regions of the search space, accelerating the optimization process. According to experiments conducted for this study, the proposed method can reduce runtime by 89\\% in Random Search and search space by (1.8/13 preprocessor and 4.3/16 classifier), without compromising significant predictive performance. Moreover, the proposed method showed competitive performance when adapted to Auto-Sklearn, reducing its search space. Furthermore, this study encompasses insights into meta-feature selection, meta-model explainability, and the trade-offs inherent in search space reduction strategies.", "date": "2025-08-19", "categories": []}
{"id": "2508.13932", "title": "How Usable is Automated Feature Engineering for Tabular Data?", "abstract": "Tabular data, consisting of rows and columns, is omnipresent across various machine learning applications. Each column represents a feature, and features can be combined or transformed to create new, more informative features. Such feature engineering is essential to achieve peak performance in machine learning. Since manual feature engineering is expensive and time-consuming, a substantial effort has been put into automating it. Yet, existing automated feature engineering (AutoFE) methods have never been investigated regarding their usability for practitioners. Thus, we investigated 53 AutoFE methods. We found that these methods are, in general, hard to use, lack documentation, and have no active communities. Furthermore, no method allows users to set time and memory constraints, which we see as a necessity for usable automation. Our survey highlights the need for future work on usable, well-engineered AutoFE methods.", "date": "2025-08-19", "categories": []}
{"id": "2409.18338", "title": "AQMLator -- An Auto Quantum Machine Learning E-Platform", "abstract": "A successful Machine Learning (ML) model implementation requires three main components: training dataset, suitable model architecture and training procedure. Given dataset and task, finding an appropriate model might be challenging. AutoML, a branch of ML, focuses on automatic architecture search -- a meta method that aims at moving human from ML system design process. The success of ML and the development of quantum computing (QC) in recent years led to a birth of new fascinating field called Quantum Machine Learning (QML) that, amongst others, incorporates quantum computers into ML models. In this paper we present AQMLator, an Auto Quantum Machine Learning platform that aims to automatically propose and train the quantum layers of an ML model with minimal input from the user. This way, data scientists can bypass the entry barrier for QC and use QML. AQMLator uses standard ML libraries, making it easy to introduce into existing ML pipelines.", "date": "2024-09-26", "categories": []}
{"id": "2504.11301", "title": "Learning to Be A Doctor: Searching for Effective Medical Agent Architectures", "abstract": "Large Language Model (LLM)-based agents have demonstrated strong capabilities across a wide range of tasks, and their application in the medical domain holds particular promise due to the demand for high generalizability and reliance on interdisciplinary knowledge. However, existing medical agent systems often rely on static, manually crafted workflows that lack the flexibility to accommodate diverse diagnostic requirements and adapt to emerging clinical scenarios. Motivated by the success of automated machine learning (AutoML), this paper introduces a novel framework for the automated design of medical agent architectures. Specifically, we define a hierarchical and expressive agent search space that enables dynamic workflow adaptation through structured modifications at the node, structural, and framework levels. Our framework conceptualizes medical agents as graph-based architectures composed of diverse, functional node types and supports iterative self-improvement guided by diagnostic feedback. Experimental results on skin disease diagnosis tasks demonstrate that the proposed method effectively evolves workflow structures and significantly enhances diagnostic accuracy over time. This work represents the first fully automated framework for medical agent architecture design and offers a scalable, adaptable foundation for deploying intelligent agents in real-world clinical environments.", "date": "2025-04-15", "categories": []}
{"id": "2507.20446", "title": "BOASF: A Unified Framework for Speeding up Automatic Machine Learning via Adaptive Successive Filtering", "abstract": "Machine learning has been making great success in many application areas. However, for the non-expert practitioners, it is always very challenging to address a machine learning task successfully and efficiently. Finding the optimal machine learning model or the hyperparameter combination set from a large number of possible alternatives usually requires considerable expert knowledge and experience. To tackle this problem, we propose a combined Bayesian Optimization and Adaptive Successive Filtering algorithm (BOASF) under a unified multi-armed bandit framework to automate the model selection or the hyperparameter optimization. Specifically, BOASF consists of multiple evaluation rounds in each of which we select promising configurations for each arm using the Bayesian optimization. Then, ASF can early discard the poor-performed arms adaptively using a Gaussian UCB-based probabilistic model. Furthermore, a Softmax model is employed to adaptively allocate available resources for each promising arm that advances to the next round. The arm with a higher probability of advancing will be allocated more resources. Experimental results show that BOASF is effective for speeding up the model selection and hyperparameter optimization processes while achieving robust and better prediction performance than the existing state-of-the-art automatic machine learning methods. Moreover, BOASF achieves better anytime performance under various time budgets.", "date": "2025-07-28", "categories": []}
{"id": "2508.02625", "title": "AutoML-Med: A Framework for Automated Machine Learning in Medical Tabular Data", "abstract": "Medical datasets are typically affected by issues such as missing values, class imbalance, a heterogeneous feature types, and a high number of features versus a relatively small number of samples, preventing machine learning models from obtaining proper results in classification and regression tasks. This paper introduces AutoML-Med, an Automated Machine Learning tool specifically designed to address these challenges, minimizing user intervention and identifying the optimal combination of preprocessing techniques and predictive models. AutoML-Med's architecture incorporates Latin Hypercube Sampling (LHS) for exploring preprocessing methods, trains models using selected metrics, and utilizes Partial Rank Correlation Coefficient (PRCC) for fine-tuned optimization of the most influential preprocessing steps. Experimental results demonstrate AutoML-Med's effectiveness in two different clinical settings, achieving higher balanced accuracy and sensitivity, which are crucial for identifying at-risk patients, compared to other state-of-the-art tools. AutoML-Med's ability to improve prediction results, especially in medical datasets with sparse data and class imbalance, highlights its potential to streamline Machine Learning applications in healthcare.", "date": "2025-08-04", "categories": []}
{"id": "2507.13710", "title": "SoftPipe: A Soft-Guided Reinforcement Learning Framework for Automated Data Preparation", "abstract": "Data preparation is a foundational yet notoriously challenging component of the machine learning lifecycle, characterized by a vast combinatorial search space. While reinforcement learning (RL) offers a promising direction, state-of-the-art methods suffer from a critical limitation: to manage the search space, they rely on rigid ``hard constraints'' that prematurely prune the search space and often preclude optimal solutions. To address this, we introduce SoftPipe, a novel RL framework that replaces these constraints with a flexible ``soft guidance'' paradigm. SoftPipe formulates action selection as a Bayesian inference problem. A high-level strategic prior, generated by a Large Language Model (LLM), probabilistically guides exploration. This prior is combined with empirical estimators from two sources through a collaborative process: a fine-grained quality score from a supervised Learning-to-Rank (LTR) model and a long-term value estimate from the agent's Q-function. Through extensive experiments on 18 diverse datasets, we demonstrate that SoftPipe achieves up to a 13.9\\% improvement in pipeline quality and 2.8$\\times$ faster convergence compared to existing methods.", "date": "2025-07-18", "categories": []}
{"id": "2507.18115", "title": "Agentic AI framework for End-to-End Medical Data Inference", "abstract": "Building and deploying machine learning solutions in healthcare remains expensive and labor-intensive due to fragmented preprocessing workflows, model compatibility issues, and stringent data privacy constraints. In this work, we introduce an Agentic AI framework that automates the entire clinical data pipeline, from ingestion to inference, through a system of modular, task-specific agents. These agents handle both structured and unstructured data, enabling automatic feature selection, model selection, and preprocessing recommendation without manual intervention. We evaluate the system on publicly available datasets from geriatrics, palliative care, and colonoscopy imaging. For example, in the case of structured data (anxiety data) and unstructured data (colonoscopy polyps data), the pipeline begins with file-type detection by the Ingestion Identifier Agent, followed by the Data Anonymizer Agent ensuring privacy compliance, where we first identify the data type and then anonymize it. The Feature Extraction Agent identifies features using an embedding-based approach for tabular data, extracting all column names, and a multi-stage MedGemma-based approach for image data, which infers modality and disease name. These features guide the Model-Data Feature Matcher Agent in selecting the best-fit model from a curated repository. The Preprocessing Recommender Agent and Preprocessing Implementor Agent then apply tailored preprocessing based on data type and model requirements. Finally, the ``Model Inference Agent\" runs the selected model on the uploaded data and generates interpretable outputs using tools like SHAP, LIME, and DETR attention maps. By automating these high-friction stages of the ML lifecycle, the proposed framework reduces the need for repeated expert intervention, offering a scalable, cost-efficient pathway for operationalizing AI in clinical environments.", "date": "2025-07-24", "categories": []}
{"id": "2507.15545", "title": "Data Aware Differentiable Neural Architecture Search for Tiny Keyword Spotting Applications", "abstract": "The success of Machine Learning is increasingly tempered by its significant resource footprint, driving interest in efficient paradigms like TinyML. However, the inherent complexity of designing TinyML systems hampers their broad adoption. To reduce this complexity, we introduce \"Data Aware Differentiable Neural Architecture Search\". Unlike conventional Differentiable Neural Architecture Search, our approach expands the search space to include data configuration parameters alongside architectural choices. This enables Data Aware Differentiable Neural Architecture Search to co-optimize model architecture and input data characteristics, effectively balancing resource usage and system performance for TinyML applications. Initial results on keyword spotting demonstrate that this novel approach to TinyML system design can generate lean but highly accurate systems.", "date": "2025-07-21", "categories": []}
{"id": "2507.13413", "title": "LightAutoDS-Tab: Multi-AutoML Agentic System for Tabular Data", "abstract": "AutoML has advanced in handling complex tasks using the integration of LLMs, yet its efficiency remains limited by dependence on specific underlying tools. In this paper, we introduce LightAutoDS-Tab, a multi-AutoML agentic system for tasks with tabular data, which combines an LLM-based code generation with several AutoML tools. Our approach improves the flexibility and robustness of pipeline design, outperforming state-of-the-art open-source solutions on several data science tasks from Kaggle. The code of LightAutoDS-Tab is available in the open repository https://github.com/sb-ai-lab/LADS", "date": "2025-07-17", "categories": []}
{"id": "2507.13712", "title": "LLaPipe: LLM-Guided Reinforcement Learning for Automated Data Preparation Pipeline Construction", "abstract": "Automated data preparation is crucial for democratizing machine learning, yet existing reinforcement learning (RL) based approaches suffer from inefficient exploration in the vast space of possible preprocessing pipelines. We present LLaPipe, a novel framework that addresses this exploration bottleneck by integrating Large Language Models (LLMs) as intelligent policy advisors. Unlike traditional methods that rely solely on statistical features and blind trial-and-error, LLaPipe leverages the semantic understanding capabilities of LLMs to provide contextually relevant exploration guidance. Our framework introduces three key innovations: (1) an LLM Policy Advisor that analyzes dataset semantics and pipeline history to suggest promising preprocessing operations, (2) an Experience Distillation mechanism that mines successful patterns from past pipelines and transfers this knowledge to guide future exploration, and (3) an Adaptive Advisor Triggering strategy (Advisor\\textsuperscript{+}) that dynamically determines when LLM intervention is most beneficial, balancing exploration effectiveness with computational cost. Through extensive experiments on 18 diverse datasets spanning multiple domains, we demonstrate that LLaPipe achieves up to 22.4\\% improvement in pipeline quality and 2.3$\\times$ faster convergence compared to state-of-the-art RL-based methods, while maintaining computational efficiency through selective LLM usage (averaging only 19.0\\% of total exploration steps).", "date": "2025-07-18", "categories": []}
{"id": "2507.11901", "title": "Imbalanced Regression Pipeline Recommendation", "abstract": "Imbalanced problems are prevalent in various real-world scenarios and are extensively explored in classification tasks. However, they also present challenges for regression tasks due to the rarity of certain target values. A common alternative is to employ balancing algorithms in preprocessing to address dataset imbalance. However, due to the variety of resampling methods and learning models, determining the optimal solution requires testing many combinations. Furthermore, the learning model, dataset, and evaluation metric affect the best strategies. This work proposes the Meta-learning for Imbalanced Regression (Meta-IR) framework, which diverges from existing literature by training meta-classifiers to recommend the best pipeline composed of the resampling strategy and learning model per task in a zero-shot fashion. The meta-classifiers are trained using a set of meta-features to learn how to map the meta-features to the classes indicating the best pipeline. We propose two formulations: Independent and Chained. Independent trains the meta-classifiers to separately indicate the best learning algorithm and resampling strategy. Chained involves a sequential procedure where the output of one meta-classifier is used as input for another to model intrinsic relationship factors. The Chained scenario showed superior performance, suggesting a relationship between the learning algorithm and the resampling strategy per task. Compared with AutoML frameworks, Meta-IR obtained better results. Moreover, compared with baselines of six learning algorithms and six resampling algorithms plus no resampling, totaling 42 (6 X 7) configurations, Meta-IR outperformed all of them. The code, data, and further information of the experiments can be found on GitHub: https://github.com/JusciAvelino/Meta-IR.", "date": "2025-07-16", "categories": []}
{"id": "2507.03498", "title": "Reinforcement Learning-based Feature Generation Algorithm for Scientific Data", "abstract": "Feature generation (FG) aims to enhance the prediction potential of original data by constructing high-order feature combinations and removing redundant features. It is a key preprocessing step for tabular scientific data to improve downstream machine-learning model performance. Traditional methods face the following two challenges when dealing with the feature generation of scientific data: First, the effective construction of high-order feature combinations in scientific data necessitates profound and extensive domain-specific expertise. Secondly, as the order of feature combinations increases, the search space expands exponentially, imposing prohibitive human labor consumption. Advancements in the Data-Centric Artificial Intelligence (DCAI) paradigm have opened novel avenues for automating feature generation processes. Inspired by that, this paper revisits the conventional feature generation workflow and proposes the Multi-agent Feature Generation (MAFG) framework. Specifically, in the iterative exploration stage, multi-agents will construct mathematical transformation equations collaboratively, synthesize and identify feature combinations ex-hibiting high information content, and leverage a reinforcement learning mechanism to evolve their strategies. Upon completing the exploration phase, MAFG integrates the large language models (LLMs) to interpreta-tively evaluate the generated features of each significant model performance breakthrough. Experimental results and case studies consistently demonstrate that the MAFG framework effectively automates the feature generation process and significantly enhances various downstream scientific data mining tasks.", "date": "2025-07-04", "categories": []}
{"id": "2506.22419", "title": "The Automated LLM Speedrunning Benchmark: Reproducing NanoGPT Improvements", "abstract": "Rapid advancements in large language models (LLMs) have the potential to assist in scientific progress. A critical capability toward this endeavor is the ability to reproduce existing work. To evaluate the ability of AI agents to reproduce results in an active research area, we introduce the Automated LLM Speedrunning Benchmark, leveraging the research community contributions on the NanoGPT speedrun, a competition to train a GPT-2 model in the shortest time. Each of the 19 speedrun tasks provides the agent with the previous records training script, optionally paired with one of three hint formats, ranging from pseudocode to paper-like descriptions of the new records improvements. Records execute quickly by design and speedrun improvements encompass diverse code-level changes, ranging from high-level algorithmic advancements to hardware-aware optimizations. These features make the benchmark both accessible and realistic for the frontier problem of improving LLM training. We find that recent reasoning LLMs combined with SoTA scaffolds struggle to reimplement already-known innovations in our benchmark, even when given detailed hints. Our benchmark thus provides a simple, non-saturated measure of an LLMs ability to automate scientific reproduction, a necessary (but not sufficient) skill for an autonomous research agent.", "date": "2025-06-27", "categories": []}
{"id": "2506.20249", "title": "Language Modeling by Language Models", "abstract": "Can we leverage LLMs to model the process of discovering novel language model (LM) architectures? Inspired by real research, we propose a multi-agent LLM approach that simulates the conventional stages of research, from ideation and literature search (proposal stage) to design implementation (code generation), generative pre-training, and downstream evaluation (verification). Using ideas from scaling laws, our system, Genesys, employs a Ladder of Scales approach; new designs are proposed, adversarially reviewed, implemented, and selectively verified at increasingly larger model scales (14M$\\sim$350M parameters) with a narrowing budget (the number of models we can train at each scale). To help make discovery efficient and factorizable, Genesys uses a novel genetic programming backbone, which we show has empirical advantages over commonly used direct prompt generation workflows (e.g., $\\sim$86\\% percentage point improvement in successful design generation, a key bottleneck). We report experiments involving 1,162 newly discovered designs (1,062 fully verified through pre-training) and find the best designs to be highly competitive with known architectures (e.g., outperform GPT2, Mamba2, etc., on 6/9 common benchmarks). We couple these results with comprehensive system-level ablations and formal results, which give broader insights into the design of effective autonomous discovery systems.", "date": "2025-06-25", "categories": []}
{"id": "2506.19540", "title": "Overtuning in Hyperparameter Optimization", "abstract": "Hyperparameter optimization (HPO) aims to identify an optimal hyperparameter configuration (HPC) such that the resulting model generalizes well to unseen data. As the expected generalization error cannot be optimized directly, it is estimated with a resampling strategy, such as holdout or cross-validation. This approach implicitly assumes that minimizing the validation error leads to improved generalization. However, since validation error estimates are inherently stochastic and depend on the resampling strategy, a natural question arises: Can excessive optimization of the validation error lead to overfitting at the HPO level, akin to overfitting in model training based on empirical risk minimization? In this paper, we investigate this phenomenon, which we term overtuning, a form of overfitting specific to HPO. Despite its practical relevance, overtuning has received limited attention in the HPO and AutoML literature. We provide a formal definition of overtuning and distinguish it from related concepts such as meta-overfitting. We then conduct a large-scale reanalysis of HPO benchmark data to assess the prevalence and severity of overtuning. Our results show that overtuning is more common than previously assumed, typically mild but occasionally severe. In approximately 10% of cases, overtuning leads to the selection of a seemingly optimal HPC with worse generalization error than the default or first configuration tried. We further analyze how factors such as performance metric, resampling strategy, dataset size, learning algorithm, and HPO method affect overtuning and discuss mitigation strategies. Our results highlight the need to raise awareness of overtuning, particularly in the small-data regime, indicating that further mitigation strategies should be studied.", "date": "2025-06-24", "categories": []}
{"id": "2506.16499", "title": "ML-Master: Towards AI-for-AI via Integration of Exploration and Reasoning", "abstract": "As AI capabilities advance toward and potentially beyond human-level performance, a natural transition emerges where AI-driven development becomes more efficient than human-centric approaches. A promising pathway toward this transition lies in AI-for-AI (AI4AI), which leverages AI techniques to automate and optimize the design, training, and deployment of AI systems themselves. While LLM-based agents have shown the potential to realize AI4AI, they are often unable to fully leverage the experience accumulated by agents during the exploration of solutions in the reasoning process, leading to inefficiencies and suboptimal performance. To address this limitation, we propose ML-Master, a novel AI4AI agent that seamlessly integrates exploration and reasoning by employing a selectively scoped memory mechanism. This approach allows ML-Master to efficiently combine diverse insights from parallel solution trajectories with analytical reasoning, guiding further exploration without overwhelming the agent with excessive context. We evaluate ML-Master on the MLE-Bench, where it achieves a 29.3% average medal rate, significantly surpassing existing methods, particularly in medium-complexity tasks, while accomplishing this superior performance within a strict 12-hour time constraint-half the 24-hour limit used by previous baselines. These results demonstrate ML-Master's potential as a powerful tool for advancing AI4AI.", "date": "2025-06-19", "categories": []}
{"id": "2506.12161", "title": "Meta-Learning and Synthetic Data for Automated Pretraining and Finetuning", "abstract": "The growing number of pretrained models in Machine Learning (ML) presents significant challenges for practitioners. Given a new dataset, they need to determine the most suitable deep learning (DL) pipeline, consisting of the pretrained model and the hyperparameters for finetuning to it. Moreover, as models grow in scale, the increasing reliance on real-world data poses a bottleneck for training and requires leveraging data more effectively. Addressing the first challenge often involves manual model selection and hyperparameter tuning. At the same time, as models grow larger and more and more of the available human-generated data is being used for training, data augmentation and synthetic data become critical elements. Automated machine learning offers a path to address these challenges but is traditionally designed for tabular data and classical ML methods. This dissertation adopts meta-learning to extend automated machine learning to the deep learning domain. We propose empirical approaches to automate DL pipeline selection for Computer Vision tasks using prior task knowledge to learn surrogate models for pipeline ranking. Extending these methods to the language domain, we learn to finetune large language models. As a result, we show that our approach can outperform finetuning foundation models. Additionally, we meta-learn data augmentation and synthetic data to enhance performance in up-stream and down-stream tasks. We empirically show the underestimated importance of data augmentation when using Self-Supervised Learning and meta-learn advanced data augmentation strategies. Leveraging synthetic data, we also propose to meta-learn neural synthetic data generators as proxies for Reinforcement Learning (RL) environments. Additionally, we learn a multiple-environment world model in an in-context learning fashion by purely using synthetic, randomly sampled data.", "date": "2025-06-11", "categories": []}
{"id": "2405.11059", "title": "Frugal Algorithm Selection", "abstract": "When solving decision and optimisation problems, many competing algorithms (model and solver choices) have complementary strengths. Typically, there is no single algorithm that works well for all instances of a problem. Automated algorithm selection has been shown to work very well for choosing a suitable algorithm for a given instance. However, the cost of training can be prohibitively large due to running candidate algorithms on a representative set of training instances. In this work, we explore reducing this cost by choosing a subset of the training instances on which to train. We approach this problem in three ways: using active learning to decide based on prediction uncertainty, augmenting the algorithm predictors with a timeout predictor, and collecting training data using a progressively increasing timeout. We evaluate combinations of these approaches on six datasets from ASLib and present the reduction in labelling cost achieved by each option.", "date": "2024-05-17", "categories": []}
{"id": "2410.02958", "title": "AutoML-Agent: A Multi-Agent LLM Framework for Full-Pipeline AutoML", "abstract": "Automated machine learning (AutoML) accelerates AI development by automating tasks in the development pipeline, such as optimal model search and hyperparameter tuning. Existing AutoML systems often require technical expertise to set up complex tools, which is in general time-consuming and requires a large amount of human effort. Therefore, recent works have started exploiting large language models (LLM) to lessen such burden and increase the usability of AutoML frameworks via a natural language interface, allowing non-expert users to build their data-driven solutions. These methods, however, are usually designed only for a particular process in the AI development pipeline and do not efficiently use the inherent capacity of the LLMs. This paper proposes AutoML-Agent, a novel multi-agent framework tailored for full-pipeline AutoML, i.e., from data retrieval to model deployment. AutoML-Agent takes user's task descriptions, facilitates collaboration between specialized LLM agents, and delivers deployment-ready models. Unlike existing work, instead of devising a single plan, we introduce a retrieval-augmented planning strategy to enhance exploration to search for more optimal plans. We also decompose each plan into sub-tasks (e.g., data preprocessing and neural network design) each of which is solved by a specialized agent we build via prompting executing in parallel, making the search process more efficient. Moreover, we propose a multi-stage verification to verify executed results and guide the code generation LLM in implementing successful solutions. Extensive experiments on seven downstream tasks using fourteen datasets show that AutoML-Agent achieves a higher success rate in automating the full AutoML process, yielding systems with good performance throughout the diverse domains.", "date": "2024-10-03", "categories": []}
{"id": "2506.05542", "title": "Agentomics-ML: Autonomous Machine Learning Experimentation Agent for Genomic and Transcriptomic Data", "abstract": "The adoption of machine learning (ML) and deep learning methods has revolutionized molecular medicine by driving breakthroughs in genomics, transcriptomics, drug discovery, and biological systems modeling. The increasing quantity, multimodality, and heterogeneity of biological datasets demand automated methods that can produce generalizable predictive models. Recent developments in large language model-based agents have shown promise for automating end-to-end ML experimentation on structured benchmarks. However, when applied to heterogeneous computational biology datasets, these methods struggle with generalization and success rates. Here, we introduce Agentomics-ML, a fully autonomous agent-based system designed to produce a classification model and the necessary files for reproducible training and inference. Our method follows predefined steps of an ML experimentation process, repeatedly interacting with the file system through Bash to complete individual steps. Once an ML model is produced, training and validation metrics provide scalar feedback to a reflection step to identify issues such as overfitting. This step then creates verbal feedback for future iterations, suggesting adjustments to steps such as data representation, model architecture, and hyperparameter choices. We have evaluated Agentomics-ML on several established genomic and transcriptomic benchmark datasets and show that it outperforms existing state-of-the-art agent-based methods in both generalization and success rates. While state-of-the-art models built by domain experts still lead in absolute performance on the majority of the computational biology datasets used in this work, Agentomics-ML narrows the gap for fully autonomous systems and achieves state-of-the-art performance on one of the used benchmark datasets. The code is available at https://github.com/BioGeMT/Agentomics-ML.", "date": "2025-06-05", "categories": []}
{"id": "2506.01584", "title": "VirnyFlow: A Design Space for Responsible Model Development", "abstract": "Developing machine learning (ML) models requires a deep understanding of real-world problems, which are inherently multi-objective. In this paper, we present VirnyFlow, the first design space for responsible model development, designed to assist data scientists in building ML pipelines that are tailored to the specific context of their problem. Unlike conventional AutoML frameworks, VirnyFlow enables users to define customized optimization criteria, perform comprehensive experimentation across pipeline stages, and iteratively refine models in alignment with real-world constraints. Our system integrates evaluation protocol definition, multi-objective Bayesian optimization, cost-aware multi-armed bandits, query optimization, and distributed parallelism into a unified architecture. We show that VirnyFlow significantly outperforms state-of-the-art AutoML systems in both optimization quality and scalability across five real-world benchmarks, offering a flexible, efficient, and responsible alternative to black-box automation in ML development.", "date": "2025-06-02", "categories": []}
{"id": "2505.19205", "title": "OptiMindTune: A Multi-Agent Framework for Intelligent Hyperparameter Optimization", "abstract": "Hyperparameter optimization (HPO) is a critical yet challenging aspect of machine learning model development, significantly impacting model performance and generalization. Traditional HPO methods often struggle with high dimensionality, complex interdependencies, and computational expense. This paper introduces OptiMindTune, a novel multi-agent framework designed to intelligently and efficiently optimize hyperparameters. OptiMindTune leverages the collaborative intelligence of three specialized AI agents -- a Recommender Agent, an Evaluator Agent, and a Decision Agent -- each powered by Google's Gemini models. These agents address distinct facets of the HPO problem, from model selection and hyperparameter suggestion to robust evaluation and strategic decision-making. By fostering dynamic interactions and knowledge sharing, OptiMindTune aims to converge to optimal hyperparameter configurations more rapidly and robustly than existing single-agent or monolithic approaches. Our framework integrates principles from advanced large language models, and adaptive search to achieve scalable and intelligent AutoML. We posit that this multi-agent paradigm offers a promising avenue for tackling the increasing complexity of modern machine learning model tuning.", "date": "2025-05-25", "categories": []}
{"id": "2505.21963", "title": "LaMDAgent: An Autonomous Framework for Post-Training Pipeline Optimization via LLM Agents", "abstract": "Large Language Models (LLMs) have demonstrated exceptional performance across a wide range of tasks. To further tailor LLMs to specific domains or applications, post-training techniques such as Supervised Fine-Tuning (SFT), Preference Learning, and model merging are commonly employed. While each of these methods has been extensively studied in isolation, the automated construction of complete post-training pipelines remains an underexplored area. Existing approaches typically rely on manual design or focus narrowly on optimizing individual components, such as data ordering or merging strategies. In this work, we introduce LaMDAgent (short for Language Model Developing Agent), a novel framework that autonomously constructs and optimizes full post-training pipelines through the use of LLM-based agents. LaMDAgent systematically explores diverse model generation techniques, datasets, and hyperparameter configurations, leveraging task-based feedback to discover high-performing pipelines with minimal human intervention. Our experiments show that LaMDAgent improves tool-use accuracy by 9.0 points while preserving instruction-following capabilities. Moreover, it uncovers effective post-training strategies that are often overlooked by conventional human-driven exploration. We further analyze the impact of data and model size scaling to reduce computational costs on the exploration, finding that model size scalings introduces new challenges, whereas scaling data size enables cost-effective pipeline discovery.", "date": "2025-05-28", "categories": []}
{"id": "2505.21978", "title": "Two-Stage Feature Generation with Transformer and Reinforcement Learning", "abstract": "Feature generation is a critical step in machine learning, aiming to enhance model performance by capturing complex relationships within the data and generating meaningful new features. Traditional feature generation methods heavily rely on domain expertise and manual intervention, making the process labor-intensive and challenging to adapt to different scenarios. Although automated feature generation techniques address these issues to some extent, they often face challenges such as feature redundancy, inefficiency in feature space exploration, and limited adaptability to diverse datasets and tasks. To address these problems, we propose a Two-Stage Feature Generation (TSFG) framework, which integrates a Transformer-based encoder-decoder architecture with Proximal Policy Optimization (PPO). The encoder-decoder model in TSFG leverages the Transformer's self-attention mechanism to efficiently represent and transform features, capturing complex dependencies within the data. PPO further enhances TSFG by dynamically adjusting the feature generation strategy based on task-specific feedback, optimizing the process for improved performance and adaptability. TSFG dynamically generates high-quality feature sets, significantly improving the predictive performance of machine learning models. Experimental results demonstrate that TSFG outperforms existing state-of-the-art methods in terms of feature quality and adaptability.", "date": "2025-05-28", "categories": []}
{"id": "2505.18243", "title": "ZeroML: A Next Generation AutoML Language", "abstract": "ZeroML is a new generation programming language for AutoML to drive the ML pipeline in a compiled and multi-paradigm way, with a pure functional core. Meeting the shortcomings introduced by Python, R, or Julia such as slow-running time, brittle pipelines or high dependency cost ZeroML brings the Microservices-based architecture adding the modular, reusable pieces such as DataCleaner, FeatureEngineer or ModelSelector. As a native multithread and memory-aware search optimized toolkit, and with one command deployability ability, ZeroML ensures non-coders and ML professionals to create high-accuracy models super fast and in a more reproducible way. The verbosity of the language ensures that when it comes to dropping into the backend, the code we will be creating is extremely clear but the level of repetition and boilerplate required when developing on the front end is now removed.", "date": "2025-05-23", "categories": []}
{"id": "2505.13941", "title": "MLZero: A Multi-Agent System for End-to-end Machine Learning Automation", "abstract": "Existing AutoML systems have advanced the automation of machine learning (ML); however, they still require substantial manual configuration and expert input, particularly when handling multimodal data. We introduce MLZero, a novel multi-agent framework powered by Large Language Models (LLMs) that enables end-to-end ML automation across diverse data modalities with minimal human intervention. A cognitive perception module is first employed, transforming raw multimodal inputs into perceptual context that effectively guides the subsequent workflow. To address key limitations of LLMs, such as hallucinated code generation and outdated API knowledge, we enhance the iterative code generation process with semantic and episodic memory. MLZero demonstrates superior performance on MLE-Bench Lite, outperforming all competitors in both success rate and solution quality, securing six gold medals. Additionally, when evaluated on our Multimodal AutoML Agent Benchmark, which includes 25 more challenging tasks spanning diverse data modalities, MLZero outperforms the competing methods by a large margin with a success rate of 0.92 (+263.6\\%) and an average rank of 2.28. Our approach maintains its robust effectiveness even with a compact 8B LLM, outperforming full-size systems from existing solutions.", "date": "2025-05-20", "categories": []}
{"id": "2504.20117", "title": "ResearchCodeAgent: An LLM Multi-Agent System for Automated Codification\n  of Research Methodologies", "abstract": "  In this paper we introduce ResearchCodeAgent, a novel multi-agent system\nleveraging large language models (LLMs) agents to automate the codification of\nresearch methodologies described in machine learning literature. The system\nbridges the gap between high-level research concepts and their practical\nimplementation, allowing researchers auto-generating code of existing research\npapers for benchmarking or building on top-of existing methods specified in the\nliterature with availability of partial or complete starter code.\nResearchCodeAgent employs a flexible agent architecture with a comprehensive\naction suite, enabling context-aware interactions with the research\nenvironment. The system incorporates a dynamic planning mechanism, utilizing\nboth short and long-term memory to adapt its approach iteratively. We evaluate\nResearchCodeAgent on three distinct machine learning tasks with distinct task\ncomplexity and representing different parts of the ML pipeline: data\naugmentation, optimization, and data batching. Our results demonstrate the\nsystem's effectiveness and generalizability, with 46.9% of generated code being\nhigh-quality and error-free, and 25% showing performance improvements over\nbaseline implementations. Empirical analysis shows an average reduction of\n57.9% in coding time compared to manual implementation. We observe higher gains\nfor more complex tasks. ResearchCodeAgent represents a significant step towards\nautomating the research implementation process, potentially accelerating the\npace of machine learning research.\n", "date": "2025-04-28", "categories": []}
{"id": "2504.17355", "title": "Collaborative Multi-Agent Reinforcement Learning for Automated Feature\n  Transformation with Graph-Driven Path Optimization", "abstract": "  Feature transformation methods aim to find an optimal mathematical\nfeature-feature crossing process that generates high-value features and\nimproves the performance of downstream machine learning tasks. Existing\nframeworks, though designed to mitigate manual costs, often treat feature\ntransformations as isolated operations, ignoring dynamic dependencies between\ntransformation steps. To address the limitations, we propose TCTO, a\ncollaborative multi-agent reinforcement learning framework that automates\nfeature engineering through graph-driven path optimization. The framework's\ncore innovation lies in an evolving interaction graph that models features as\nnodes and transformations as edges. Through graph pruning and backtracking, it\ndynamically eliminates low-impact edges, reduces redundant operations, and\nenhances exploration stability. This graph also provides full traceability to\nempower TCTO to reuse high-utility subgraphs from historical transformations.\nTo demonstrate the efficacy and adaptability of our approach, we conduct\ncomprehensive experiments and case studies, which show superior performance\nacross a range of datasets.\n", "date": "2025-04-24", "categories": []}
{"id": "2504.01222", "title": "AutoML Benchmark with shorter time constraints and early stopping", "abstract": "  Automated Machine Learning (AutoML) automatically builds machine learning\n(ML) models on data. The de facto standard for evaluating new AutoML frameworks\nfor tabular data is the AutoML Benchmark (AMLB). AMLB proposed to evaluate\nAutoML frameworks using 1- and 4-hour time budgets across 104 tasks. We argue\nthat shorter time constraints should be considered for the benchmark because of\ntheir practical value, such as when models need to be retrained with high\nfrequency, and to make AMLB more accessible. This work considers two ways in\nwhich to reduce the overall computation used in the benchmark: smaller time\nconstraints and the use of early stopping. We conduct evaluations of 11 AutoML\nframeworks on 104 tasks with different time constraints and find the relative\nranking of AutoML frameworks is fairly consistent across time constraints, but\nthat using early-stopping leads to a greater variety in model performance.\n", "date": "2025-04-01", "categories": []}
{"id": "2503.04350", "title": "EDCA - An Evolutionary Data-Centric AutoML Framework for Efficient\n  Pipelines", "abstract": "  Automated Machine Learning (AutoML) gained popularity due to the increased\ndemand for Machine Learning (ML) specialists, allowing them to apply ML\ntechniques effortlessly and quickly. AutoML implementations use optimisation\nmethods to identify the most effective ML solution for a given dataset, aiming\nto improve one or more predefined metrics. However, most implementations focus\non model selection and hyperparameter tuning. Despite being an important factor\nin obtaining high-performance ML systems, data quality is usually an overlooked\npart of AutoML and continues to be a manual and time-consuming task. This work\npresents EDCA, an Evolutionary Data Centric AutoML framework. In addition to\nthe traditional tasks such as selecting the best models and hyperparameters,\nEDCA enhances the given data by optimising data processing tasks such as data\nreduction and cleaning according to the problems' needs. All these steps create\nan ML pipeline that is optimised by an evolutionary algorithm. To assess its\neffectiveness, EDCA was compared to FLAML and TPOT, two frameworks at the top\nof the AutoML benchmarks. The frameworks were evaluated in the same conditions\nusing datasets from AMLB classification benchmarks. EDCA achieved statistically\nsimilar results in performance to FLAML and TPOT but used significantly less\ndata to train the final solutions. Moreover, EDCA experimental results reveal\nthat a good performance can be achieved using less data and efficient ML\nalgorithm aspects that align with Green AutoML guidelines\n", "date": "2025-03-06", "categories": []}
{"id": "2501.03916", "title": "Dolphin: Moving Towards Closed-loop Auto-research through Thinking,\n  Practice, and Feedback", "abstract": "  The scientific research paradigm is undergoing a profound transformation\nowing to the development of Artificial Intelligence (AI). Recent works\ndemonstrate that various AI-assisted research methods can largely improve\nresearch efficiency by improving data analysis, accelerating computation, and\nfostering novel idea generation. To further move towards the ultimate goal\n(i.e., automatic scientific research), in this paper, we introduce Dolphin, a\nclosed-loop LLM-driven framework to enhance the automation level of scientific\nresearch. Dolphin first generates novel ideas based on feedback from previous\nexperiments and relevant papers ranked by the topic and task attributes. Then,\nthe generated ideas can be implemented using a code template refined and\ndebugged with the designed exception-traceback-guided local code structure.\nFinally, Dolphin automatically analyzes the results of each idea and feeds the\nresults back to the next round of idea generation. Experiments are conducted on\nthe benchmark datasets of different topics and a subset of MLE-bench. Results\nshow that Dolphin can continuously improve the performance of the input topic\nin a loop. We highlight that Dolphin can automatically propose methods that are\ncomparable to the state-of-the-art in some tasks such as 3D point\nclassification.\n", "date": "2025-01-07", "categories": []}
{"id": "2504.06207", "title": "An experimental survey and Perspective View on Meta-Learning for\n  Automated Algorithms Selection and Parametrization", "abstract": "  Considerable progress has been made in the recent literature studies to\ntackle the Algorithms Selection and Parametrization (ASP) problem, which is\ndiversified in multiple meta-learning setups. Yet there is a lack of surveys\nand comparative evaluations that critically analyze, summarize and assess the\nperformance of existing methods. In this paper, we provide an overview of the\nstate of the art in this continuously evolving field. The survey sheds light on\nthe motivational reasons for pursuing classifiers selection through\nmeta-learning. In this regard, Automated Machine Learning (AutoML) is usually\ntreated as an ASP problem under the umbrella of the democratization of machine\nlearning. Accordingly, AutoML makes machine learning techniques accessible to\ndomain scientists who are interested in applying advanced analytics but lack\nthe required expertise. It can ease the task of manually selecting ML\nalgorithms and tuning related hyperparameters. We comprehensively discuss the\ndifferent phases of classifiers selection based on a generic framework that is\nformed as an outcome of reviewing prior works. Subsequently, we propose a\nbenchmark knowledge base of 4 millions previously learned models and present\nextensive comparative evaluations of the prominent methods for classifiers\nselection based on 08 classification algorithms and 400 benchmark datasets. The\ncomparative study quantitatively assesses the performance of algorithms\nselection methods along while emphasizing the strengths and limitations of\nexisting studies.\n", "date": "2025-04-08", "categories": []}
{"id": "2503.24019", "title": "AutoML Algorithms for Online Generalized Additive Model Selection:\n  Application to Electricity Demand Forecasting", "abstract": "  Electricity demand forecasting is key to ensuring that supply meets demand\nlest the grid would blackout. Reliable short-term forecasts may be obtained by\ncombining a Generalized Additive Models (GAM) with a State-Space model (Obst et\nal., 2021), leading to an adaptive (or online) model. A GAM is an\nover-parameterized linear model defined by a formula and a state-space model\ninvolves hyperparameters. Both the formula and adaptation parameters have to be\nfixed before model training and have a huge impact on the model's predictive\nperformance. We propose optimizing them using the DRAGON package of Keisler\n(2025), originally designed for neural architecture search. This work\ngeneralizes it for automated online generalized additive model selection by\ndefining an efficient modeling of the search space (namely, the space of the\nGAM formulae and adaptation parameters). Its application to short-term French\nelectricity demand forecasting demonstrates the relevance of the approach\n", "date": "2025-03-31", "categories": []}
{"id": "2503.20394", "title": "FastFT: Accelerating Reinforced Feature Transformation via Advanced\n  Exploration Strategies", "abstract": "  Feature Transformation is crucial for classic machine learning that aims to\ngenerate feature combinations to enhance the performance of downstream tasks\nfrom a data-centric perspective. Current methodologies, such as manual\nexpert-driven processes, iterative-feedback techniques, and\nexploration-generative tactics, have shown promise in automating such data\nengineering workflow by minimizing human involvement. However, three challenges\nremain in those frameworks: (1) It predominantly depends on downstream task\nperformance metrics, as assessment is time-consuming, especially for large\ndatasets. (2) The diversity of feature combinations will hardly be guaranteed\nafter random exploration ends. (3) Rare significant transformations lead to\nsparse valuable feedback that hinders the learning processes or leads to less\neffective results. In response to these challenges, we introduce FastFT, an\ninnovative framework that leverages a trio of advanced strategies.We first\ndecouple the feature transformation evaluation from the outcomes of the\ngenerated datasets via the performance predictor. To address the issue of\nreward sparsity, we developed a method to evaluate the novelty of generated\ntransformation sequences. Incorporating this novelty into the reward function\naccelerates the model's exploration of effective transformations, thereby\nimproving the search productivity. Additionally, we combine novelty and\nperformance to create a prioritized memory buffer, ensuring that essential\nexperiences are effectively revisited during exploration. Our extensive\nexperimental evaluations validate the performance, efficiency, and traceability\nof our proposed framework, showcasing its superiority in handling complex\nfeature transformation tasks.\n", "date": "2025-03-26", "categories": []}
{"id": "2503.05106", "title": "Grouped Sequential Optimization Strategy -- the Application of\n  Hyperparameter Importance Assessment in Deep Learning", "abstract": "  Hyperparameter optimization (HPO) is a critical component of machine learning\npipelines, significantly affecting model robustness, stability, and\ngeneralization. However, HPO is often a time-consuming and computationally\nintensive task. Traditional HPO methods, such as grid search and random search,\noften suffer from inefficiency. Bayesian optimization, while more efficient,\nstill struggles with high-dimensional search spaces. In this paper, we\ncontribute to the field by exploring how insights gained from hyperparameter\nimportance assessment (HIA) can be leveraged to accelerate HPO, reducing both\ntime and computational resources. Building on prior work that quantified\nhyperparameter importance by evaluating 10 hyperparameters on CNNs using 10\ncommon image classification datasets, we implement a novel HPO strategy called\n'Sequential Grouping.' That prior work assessed the importance weights of the\ninvestigated hyperparameters based on their influence on model performance,\nproviding valuable insights that we leverage to optimize our HPO process. Our\nexperiments, validated across six additional image classification datasets,\ndemonstrate that incorporating hyperparameter importance assessment (HIA) can\nsignificantly accelerate HPO without compromising model performance, reducing\noptimization time by an average of 31.9\\% compared to the conventional\nsimultaneous strategy.\n", "date": "2025-03-07", "categories": []}
{"id": "2403.08352", "title": "Data augmentation with automated machine learning: approaches and\n  performance comparison with classical data augmentation methods", "abstract": "  Data augmentation is arguably the most important regularization technique\ncommonly used to improve generalization performance of machine learning models.\nIt primarily involves the application of appropriate data transformation\noperations to create new data samples with desired properties. Despite its\neffectiveness, the process is often challenging because of the time-consuming\ntrial and error procedures for creating and testing different candidate\naugmentations and their hyperparameters manually. State-of-the-art approaches\nare increasingly relying on automated machine learning (AutoML) principles.\nThis work presents a comprehensive survey of AutoML-based data augmentation\ntechniques. We discuss various approaches for accomplishing data augmentation\nwith AutoML, including data manipulation, data integration and data synthesis\ntechniques. The focus of this work is on image data augmentation methods.\nNonetheless, we cover other data modalities, especially in cases where the\nspecific data augmentations techniques being discussed are more suitable for\nthese other modalities. For instance, since automated data integration methods\nare more suitable for tabular data, we cover tabular data in the discussion of\ndata integration methods. The work also presents extensive discussion of\ntechniques for accomplishing each of the major subtasks of the image data\naugmentation process: search space design, hyperparameter optimization and\nmodel evaluation. Finally, we carried out an extensive comparison and analysis\nof the performance of automated data augmentation techniques and\nstate-of-the-art methods based on classical augmentation approaches. The\nresults show that AutoML methods for data augmentation currently outperform\nstate-of-the-art techniques based on conventional approaches.\n", "date": "2024-03-13", "categories": []}
{"id": "2408.08435", "title": "Automated Design of Agentic Systems", "abstract": "  Researchers are investing substantial effort in developing powerful\ngeneral-purpose agents, wherein Foundation Models are used as modules within\nagentic systems (e.g. Chain-of-Thought, Self-Reflection, Toolformer). However,\nthe history of machine learning teaches us that hand-designed solutions are\neventually replaced by learned solutions. We describe a newly forming research\narea, Automated Design of Agentic Systems (ADAS), which aims to automatically\ncreate powerful agentic system designs, including inventing novel building\nblocks and/or combining them in new ways. We further demonstrate that there is\nan unexplored yet promising approach within ADAS where agents can be defined in\ncode and new agents can be automatically discovered by a meta agent programming\never better ones in code. Given that programming languages are Turing Complete,\nthis approach theoretically enables the learning of any possible agentic\nsystem: including novel prompts, tool use, workflows, and combinations thereof.\nWe present a simple yet effective algorithm named Meta Agent Search to\ndemonstrate this idea, where a meta agent iteratively programs interesting new\nagents based on an ever-growing archive of previous discoveries. Through\nextensive experiments across multiple domains including coding, science, and\nmath, we show that our algorithm can progressively invent agents with novel\ndesigns that greatly outperform state-of-the-art hand-designed agents.\nImportantly, we consistently observe the surprising result that agents invented\nby Meta Agent Search maintain superior performance even when transferred across\ndomains and models, demonstrating their robustness and generality. Provided we\ndevelop it safely, our work illustrates the potential of an exciting new\nresearch direction toward automatically designing ever-more powerful agentic\nsystems to benefit humanity.\n", "date": "2024-08-15", "categories": []}
{"id": "2409.04665", "title": "IIFE: Interaction Information Based Automated Feature Engineering", "abstract": "  Automated feature engineering (AutoFE) is the process of automatically\nbuilding and selecting new features that help improve downstream predictive\nperformance. While traditional feature engineering requires significant domain\nexpertise and time-consuming iterative testing, AutoFE strives to make feature\nengineering easy and accessible to all data science practitioners. We introduce\na new AutoFE algorithm, IIFE, based on determining which feature pairs\nsynergize well through an information-theoretic perspective called interaction\ninformation. We demonstrate the superior performance of IIFE over existing\nalgorithms. We also show how interaction information can be used to improve\nexisting AutoFE algorithms. Finally, we highlight several critical experimental\nsetup issues in the existing AutoFE literature and their effects on\nperformance.\n", "date": "2024-09-07", "categories": []}
{"id": "2502.20279", "title": "Online Meta-learning for AutoML in Real-time (OnMAR)", "abstract": "  Automated machine learning (AutoML) is a research area focusing on using\noptimisation techniques to design machine learning (ML) algorithms, alleviating\nthe need for a human to perform manual algorithm design. Real-time AutoML\nenables the design process to happen while the ML algorithm is being applied to\na task. Real-time AutoML is an emerging research area, as such existing\nreal-time AutoML techniques need improvement with respect to the quality of\ndesigns and time taken to create designs. To address these issues, this study\nproposes an Online Meta-learning for AutoML in Real-time (OnMAR) approach.\nMeta-learning gathers information about the optimisation process undertaken by\nthe ML algorithm in the form of meta-features. Meta-features are used in\nconjunction with a meta-learner to optimise the optimisation process. The OnMAR\napproach uses a meta-learner to predict the accuracy of an ML design. If the\naccuracy predicted by the meta-learner is sufficient, the design is used, and\nif the predicted accuracy is low, an optimisation technique creates a new\ndesign. A genetic algorithm (GA) is the optimisation technique used as part of\nthe OnMAR approach. Different meta-learners (k-nearest neighbours, random\nforest and XGBoost) are tested. The OnMAR approach is model-agnostic (i.e. not\nspecific to a single real-time AutoML application) and therefore evaluated on\nthree different real-time AutoML applications, namely: composing an image\nclustering algorithm, configuring the hyper-parameters of a convolutional\nneural network, and configuring a video classification pipeline. The OnMAR\napproach is effective, matching or outperforming existing real-time AutoML\napproaches, with the added benefit of a faster runtime.\n", "date": "2025-02-27", "categories": []}
{"id": "2502.20301", "title": "M^3Builder: A Multi-Agent System for Automated Machine Learning in\n  Medical Imaging", "abstract": "  Agentic AI systems have gained significant attention for their ability to\nautonomously perform complex tasks. However, their reliance on well-prepared\ntools limits their applicability in the medical domain, which requires to train\nspecialized models. In this paper, we make three contributions: (i) We present\nM3Builder, a novel multi-agent system designed to automate machine learning\n(ML) in medical imaging. At its core, M3Builder employs four specialized agents\nthat collaborate to tackle complex, multi-step medical ML workflows, from\nautomated data processing and environment configuration to self-contained auto\ndebugging and model training. These agents operate within a medical imaging ML\nworkspace, a structured environment designed to provide agents with free-text\ndescriptions of datasets, training codes, and interaction tools, enabling\nseamless communication and task execution. (ii) To evaluate progress in\nautomated medical imaging ML, we propose M3Bench, a benchmark comprising four\ngeneral tasks on 14 training datasets, across five anatomies and three imaging\nmodalities, covering both 2D and 3D data. (iii) We experiment with seven\nstate-of-the-art large language models serving as agent cores for our system,\nsuch as Claude series, GPT-4o, and DeepSeek-V3. Compared to existing ML agentic\ndesigns, M3Builder shows superior performance on completing ML tasks in medical\nimaging, achieving a 94.29% success rate using Claude-3.7-Sonnet as the agent\ncore, showing huge potential towards fully automated machine learning in\nmedical imaging.\n", "date": "2025-02-27", "categories": []}
{"id": "2402.01881", "title": "Large Language Model Agent for Hyper-Parameter Optimization", "abstract": "  Hyperparameter optimization is critical in modern machine learning, requiring\nexpert knowledge, numerous trials, and high computational and human resources.\nDespite the advancements in Automated Machine Learning (AutoML), challenges in\nterms of trial efficiency, setup complexity, and interoperability still\npersist. To address these issues, we introduce a novel paradigm leveraging\nLarge Language Models (LLMs) to automate hyperparameter optimization across\ndiverse machine learning tasks, which is named AgentHPO (short for LLM\nAgent-based Hyperparameter Optimization). Specifically, AgentHPO processes the\ntask information autonomously, conducts experiments with specific\nhyperparameters (HPs), and iteratively optimizes them based on historical\ntrials. This human-like optimization process largely reduces the number of\nrequired trials, simplifies the setup process, and enhances interpretability\nand user trust, compared to traditional AutoML methods. Extensive empirical\nexperiments conducted on 12 representative machine-learning tasks indicate that\nAgentHPO not only matches but also often surpasses the best human trials in\nterms of performance while simultaneously providing explainable results.\nFurther analysis sheds light on the strategies employed by the LLM in\noptimizing these tasks, highlighting its effectiveness and adaptability in\nvarious scenarios.\n", "date": "2024-02-02", "categories": []}
{"id": "2502.19180", "title": "AutoML for Multi-Class Anomaly Compensation of Sensor Drift", "abstract": "  Addressing sensor drift is essential in industrial measurement systems, where\nprecise data output is necessary for maintaining accuracy and reliability in\nmonitoring processes, as it progressively degrades the performance of machine\nlearning models over time. Our findings indicate that the standard\ncross-validation method used in existing model training overestimates\nperformance by inadequately accounting for drift. This is primarily because\ntypical cross-validation techniques allow data instances to appear in both\ntraining and testing sets, thereby distorting the accuracy of the predictive\nevaluation. As a result, these models are unable to precisely predict future\ndrift effects, compromising their ability to generalize and adapt to evolving\ndata conditions. This paper presents two solutions: (1) a novel sensor drift\ncompensation learning paradigm for validating models, and (2) automated machine\nlearning (AutoML) techniques to enhance classification performance and\ncompensate sensor drift. By employing strategies such as data balancing,\nmeta-learning, automated ensemble learning, hyperparameter optimization,\nfeature selection, and boosting, our AutoML-DC (Drift Compensation) model\nsignificantly improves classification performance against sensor drift.\nAutoML-DC further adapts effectively to varying drift severities.\n", "date": "2025-02-26", "categories": []}
{"id": "2502.13138", "title": "AIDE: AI-Driven Exploration in the Space of Code", "abstract": "  Machine learning, the foundation of modern artificial intelligence, has\ndriven innovations that have fundamentally transformed the world. Yet, behind\nadvancements lies a complex and often tedious process requiring labor and\ncompute intensive iteration and experimentation. Engineers and scientists\ndeveloping machine learning models spend much of their time on trial-and-error\ntasks instead of conceptualizing innovative solutions or research hypotheses.\nTo address this challenge, we introduce AI-Driven Exploration (AIDE), a machine\nlearning engineering agent powered by large language models (LLMs). AIDE frames\nmachine learning engineering as a code optimization problem, and formulates\ntrial-and-error as a tree search in the space of potential solutions. By\nstrategically reusing and refining promising solutions, AIDE effectively trades\ncomputational resources for enhanced performance, achieving state-of-the-art\nresults on multiple machine learning engineering benchmarks, including our\nKaggle evaluations, OpenAI MLE-Bench and METRs RE-Bench.\n", "date": "2025-02-18", "categories": []}
{"id": "2405.18246", "title": "Utilitarian Algorithm Configuration for Infinite Parameter Spaces", "abstract": "  Utilitarian algorithm configuration is a general-purpose technique for\nautomatically searching the parameter space of a given algorithm to optimize\nits performance, as measured by a given utility function, on a given set of\ninputs. Recently introduced utilitarian configuration procedures offer\noptimality guarantees about the returned parameterization while provably\nadapting to the hardness of the underlying problem. However, the applicability\nof these approaches is severely limited by the fact that they only search a\nfinite, relatively small set of parameters. They cannot effectively search the\nconfiguration space of algorithms with continuous or uncountable parameters. In\nthis paper we introduce a new procedure, which we dub COUP (Continuous,\nOptimistic Utilitarian Procrastination). COUP is designed to search infinite\nparameter spaces efficiently to find good configurations quickly. Furthermore,\nCOUP maintains the theoretical benefits of previous utilitarian configuration\nprocedures when applied to finite parameter spaces but is significantly faster,\nboth provably and experimentally.\n", "date": "2024-05-28", "categories": []}
{"id": "2502.09329", "title": "Bayesian Optimization for Simultaneous Selection of Machine Learning\n  Algorithms and Hyperparameters on Shared Latent Space", "abstract": "  Selecting the optimal combination of a machine learning (ML) algorithm and\nits hyper-parameters is crucial for the development of high-performance ML\nsystems. However, since the combination of ML algorithms and hyper-parameters\nis enormous, the exhaustive validation requires a significant amount of time.\nMany existing studies use Bayesian optimization (BO) for accelerating the\nsearch. On the other hand, a significant difficulty is that, in general, there\nexists a different hyper-parameter space for each one of candidate ML\nalgorithms. BO-based approaches typically build a surrogate model independently\nfor each hyper-parameter space, by which sufficient observations are required\nfor all candidate ML algorithms. In this study, our proposed method embeds\ndifferent hyper-parameter spaces into a shared latent space, in which a\nsurrogate multi-task model for BO is estimated. This approach can share\ninformation of observations from different ML algorithms by which efficient\noptimization is expected with a smaller number of total observations. We\nfurther propose the pre-training of the latent space embedding with an\nadversarial regularization, and a ranking model for selecting an effective\npre-trained embedding for a given target dataset. Our empirical study\ndemonstrates effectiveness of the proposed method through datasets from OpenML.\n", "date": "2025-02-13", "categories": []}
{"id": "2502.08056", "title": "Cognify: Supercharging Gen-AI Workflows With Hierarchical Autotuning", "abstract": "  Today's gen-AI workflows that involve multiple ML model calls, tool/API\ncalls, data retrieval, or generic code execution are often tuned manually in an\nad-hoc way that is both time-consuming and error-prone. In this paper, we\npropose a systematic approach for automatically tuning gen-AI workflows. Our\nkey insight is that gen-AI workflows can benefit from structure, operator, and\nprompt changes, but unique properties of gen-AI workflows require new\noptimization techniques. We propose AdaSeek, an adaptive hierarchical search\nalgorithm for autotuning gen-AI workflows. AdaSeek organizes workflow tuning\nmethods into different layers based on the user-specified total search budget\nand distributes the budget across different layers based on the complexity of\neach layer. During its hierarchical search, AdaSeek redistributes the search\nbudget from less useful to more promising tuning configurations based on\nworkflow-level evaluation results. We implement AdaSeek in a workflow\nautotuning framework called Cognify and evaluate Cognify using six types of\nworkflows such as RAG-based QA and text-to-SQL transformation. Overall, Cognify\nimproves these workflows' generation quality by up to 2.8x, reduces execution\nmonetary cost by up to 10x, and reduces end-to-end latency by 2.7x.\n", "date": "2025-02-12", "categories": []}
{"id": "2502.00871", "title": "Modified Adaptive Tree-Structured Parzen Estimator for Hyperparameter\n  Optimization", "abstract": "  In this paper, we review hyperparameter optimization methods for machine\nlearning models, with a particular focus on the Adaptive Tree-Structured Parzen\nEstimator (ATPE) algorithm. We propose several modifications to ATPE and assess\ntheir efficacy on a diverse set of standard benchmark functions. Experimental\nresults demonstrate that the proposed modifications significantly improve the\neffectiveness of ATPE hyperparameter optimization on selected benchmarks, a\nfinding that holds practical relevance for their application in real-world\nmachine learning / optimization tasks.\n", "date": "2025-02-02", "categories": []}
{"id": "2409.15844", "title": "Adaptive Learn-then-Test: Statistically Valid and Efficient\n  Hyperparameter Selection", "abstract": "  We introduce adaptive learn-then-test (aLTT), an efficient hyperparameter\nselection procedure that provides finite-sample statistical guarantees on the\npopulation risk of AI models. Unlike the existing learn-then-test (LTT)\ntechnique, which relies on conventional p-value-based multiple hypothesis\ntesting (MHT), aLTT implements sequential data-dependent MHT with early\ntermination by leveraging e-processes. As a result, aLTT can reduce the number\nof testing rounds, making it particularly well-suited for scenarios in which\ntesting is costly or presents safety risks. Apart from maintaining statistical\nvalidity, in applications such as online policy selection for offline\nreinforcement learning and prompt engineering, aLTT is shown to achieve the\nsame performance as LTT while requiring only a fraction of the testing rounds.\n", "date": "2024-09-24", "categories": []}
{"id": "2501.18126", "title": "HyperZero: A Customized End-to-End Auto-Tuning System for Recommendation\n  with Hourly Feedback", "abstract": "  Modern recommendation systems can be broadly divided into two key stages: the\nranking stage, where the system predicts various user engagements (e.g.,\nclick-through rate, like rate, follow rate, watch time), and the value model\nstage, which aggregates these predictive scores through a function (e.g., a\nlinear combination defined by a weight vector) to measure the value of each\ncontent by a single numerical score. Both stages play roughly equally important\nroles in real industrial systems; however, how to optimize the model weights\nfor the second stage still lacks systematic study. This paper focuses on\noptimizing the second stage through auto-tuning technology. Although general\nauto-tuning systems and solutions - both from established production practices\nand open-source solutions - can address this problem, they typically require\nweeks or even months to identify a feasible solution. Such prolonged tuning\nprocesses are unacceptable in production environments for recommendation\nsystems, as suboptimal value models can severely degrade user experience. An\neffective auto-tuning solution is required to identify a viable model within\n2-3 days, rather than the extended timelines typically associated with existing\napproaches. In this paper, we introduce a practical auto-tuning system named\nHyperZero that addresses these time constraints while effectively solving the\nunique challenges inherent in modern recommendation systems. Moreover, this\nframework has the potential to be expanded to broader tuning tasks within\nrecommendation systems.\n", "date": "2025-01-30", "categories": []}
{"id": "2501.17411", "title": "A Genetic Algorithm-Based Approach for Automated Optimization of\n  Kolmogorov-Arnold Networks in Classification Tasks", "abstract": "  To address the issue of interpretability in multilayer perceptrons (MLPs),\nKolmogorov-Arnold Networks (KANs) are introduced in 2024. However, optimizing\nKAN structures is labor-intensive, typically requiring manual intervention and\nparameter tuning. This paper proposes GA-KAN, a genetic algorithm-based\napproach that automates the optimization of KANs, requiring no human\nintervention in the design process. To the best of our knowledge, this is the\nfirst time that evolutionary computation is explored to optimize KANs\nautomatically. Furthermore, inspired by the use of sparse connectivity in MLPs\nin effectively reducing the number of parameters, GA-KAN further explores\nsparse connectivity to tackle the challenge of extensive parameter spaces in\nKANs. GA-KAN is validated on two toy datasets, achieving optimal results\nwithout the manual tuning required by the original KAN. Additionally, GA-KAN\ndemonstrates superior performance across five classification datasets,\noutperforming traditional methods on all datasets and providing interpretable\nsymbolic formulae for the Wine and Iris datasets, thereby enhancing model\ntransparency. Furthermore, GA-KAN significantly reduces the number of\nparameters over the standard KAN across all the five datasets. The core\ncontributions of GA-KAN include automated optimization, a new encoding\nstrategy, and a new decoding process, which together improve the accuracy and\ninterpretability, and reduce the number of parameters.\n", "date": "2025-01-29", "categories": []}
{"id": "2501.04260", "title": "Modeling All Response Surfaces in One for Conditional Search Spaces", "abstract": "  Bayesian Optimization (BO) is a sample-efficient black-box optimizer commonly\nused in search spaces where hyperparameters are independent. However, in many\npractical AutoML scenarios, there will be dependencies among hyperparameters,\nforming a conditional search space, which can be partitioned into structurally\ndistinct subspaces. The structure and dimensionality of hyperparameter\nconfigurations vary across these subspaces, challenging the application of BO.\nSome previous BO works have proposed solutions to develop multiple Gaussian\nProcess models in these subspaces. However, these approaches tend to be\ninefficient as they require a substantial number of observations to guarantee\neach GP's performance and cannot capture relationships between hyperparameters\nacross different subspaces. To address these issues, this paper proposes a\nnovel approach to model the response surfaces of all subspaces in one, which\ncan model the relationships between hyperparameters elegantly via a\nself-attention mechanism. Concretely, we design a structure-aware\nhyperparameter embedding to preserve the structural information. Then, we\nintroduce an attention-based deep feature extractor, capable of projecting\nconfigurations with different structures from various subspaces into a unified\nfeature space, where the response surfaces can be formulated using a single\nstandard Gaussian Process. The empirical results on a simulation function,\nvarious real-world tasks, and HPO-B benchmark demonstrate that our proposed\napproach improves the efficacy and efficiency of BO within conditional search\nspaces.\n", "date": "2025-01-08", "categories": []}
{"id": "2501.14995", "title": "GreenAuto: An Automated Platform for Sustainable AI Model Design on Edge\n  Devices", "abstract": "  We present GreenAuto, an end-to-end automated platform designed for\nsustainable AI model exploration, generation, deployment, and evaluation.\nGreenAuto employs a Pareto front-based search method within an expanded neural\narchitecture search (NAS) space, guided by gradient descent to optimize model\nexploration. Pre-trained kernel-level energy predictors estimate energy\nconsumption across all models, providing a global view that directs the search\ntoward more sustainable solutions. By automating performance measurements and\niteratively refining the search process, GreenAuto demonstrates the efficient\nidentification of sustainable AI models without the need for human\nintervention.\n", "date": "2025-01-25", "categories": []}
{"id": "2307.10774", "title": "Assessing the Use of AutoML for Data-Driven Software Engineering", "abstract": "  Background. Due to the widespread adoption of Artificial Intelligence (AI)\nand Machine Learning (ML) for building software applications, companies are\nstruggling to recruit employees with a deep understanding of such technologies.\nIn this scenario, AutoML is soaring as a promising solution to fill the AI/ML\nskills gap since it promises to automate the building of end-to-end AI/ML\npipelines that would normally be engineered by specialized team members. Aims.\nDespite the growing interest and high expectations, there is a dearth of\ninformation about the extent to which AutoML is currently adopted by teams\ndeveloping AI/ML-enabled systems and how it is perceived by practitioners and\nresearchers. Method. To fill these gaps, in this paper, we present a\nmixed-method study comprising a benchmark of 12 end-to-end AutoML tools on two\nSE datasets and a user survey with follow-up interviews to further our\nunderstanding of AutoML adoption and perception. Results. We found that AutoML\nsolutions can generate models that outperform those trained and optimized by\nresearchers to perform classification tasks in the SE domain. Also, our\nfindings show that the currently available AutoML solutions do not live up to\ntheir names as they do not equally support automation across the stages of the\nML development workflow and for all the team members. Conclusions. We derive\ninsights to inform the SE research community on how AutoML can facilitate their\nactivities and tool builders on how to design the next generation of AutoML\ntechnologies.\n", "date": "2023-07-20", "categories": []}
{"id": "2306.12215", "title": "Automated Machine Learning for Remaining Useful Life Predictions", "abstract": "  Being able to predict the remaining useful life (RUL) of an engineering\nsystem is an important task in prognostics and health management. Recently,\ndata-driven approaches to RUL predictions are becoming prevalent over\nmodel-based approaches since no underlying physical knowledge of the\nengineering system is required. Yet, this just replaces required expertise of\nthe underlying physics with machine learning (ML) expertise, which is often\nalso not available. Automated machine learning (AutoML) promises to build\nend-to-end ML pipelines automatically enabling domain experts without ML\nexpertise to create their own models. This paper introduces AutoRUL, an\nAutoML-driven end-to-end approach for automatic RUL predictions. AutoRUL\ncombines fine-tuned standard regression methods to an ensemble with high\npredictive power. By evaluating the proposed method on eight real-world and\nsynthetic datasets against state-of-the-art hand-crafted models, we show that\nAutoML provides a viable alternative to hand-crafted data-driven RUL\npredictions. Consequently, creating RUL predictions can be made more accessible\nfor domain experts using AutoML by eliminating ML expertise from data-driven\nmodel construction.\n", "date": "2023-06-21", "categories": []}
{"id": "2412.11526", "title": "Probabilities-Informed Machine Learning", "abstract": "  Machine learning (ML) has emerged as a powerful tool for tackling complex\nregression and classification tasks, yet its success often hinges on the\nquality of training data. This study introduces an ML paradigm inspired by\ndomain knowledge of the structure of output function, akin to physics-informed\nML, but rooted in probabilistic principles rather than physical laws. The\nproposed approach integrates the probabilistic structure of the target variable\n(such as its cumulative distribution function) into the training process. This\nprobabilistic information is obtained from historical data or estimated using\nstructural reliability methods during experimental design. By embedding\ndomain-specific probabilistic insights into the learning process, the technique\nenhances model accuracy and mitigates risks of overfitting and underfitting.\nApplications in regression, image denoising, and classification demonstrate the\napproach's effectiveness in addressing real-world problems.\n", "date": "2024-12-16", "categories": []}
{"id": "2405.07640", "title": "Hyperparameter Importance Analysis for Multi-Objective AutoML", "abstract": "  Hyperparameter optimization plays a pivotal role in enhancing the predictive\nperformance and generalization capabilities of ML models. However, in many\napplications, we do not only care about predictive performance but also about\nadditional objectives such as inference time, memory, or energy consumption. In\nsuch multi-objective scenarios, determining the importance of hyperparameters\nposes a significant challenge due to the complex interplay between the\nconflicting objectives. In this paper, we propose the first method for\nassessing the importance of hyperparameters in multi-objective hyperparameter\noptimization. Our approach leverages surrogate-based hyperparameter importance\nmeasures, i.e., fANOVA and ablation paths, to provide insights into the impact\nof hyperparameters on the optimization objectives. Specifically, we compute the\na-priori scalarization of the objectives and determine the importance of the\nhyperparameters for different objective tradeoffs. Through extensive empirical\nevaluations on diverse benchmark datasets with three different objective pairs,\neach combined with accuracy, namely time, demographic parity loss, and energy\nconsumption, we demonstrate the effectiveness and robustness of our proposed\nmethod. Our findings not only offer valuable guidance for hyperparameter tuning\nin multi-objective optimization tasks but also contribute to advancing the\nunderstanding of hyperparameter importance in complex optimization scenarios.\n", "date": "2024-05-13", "categories": []}
{"id": "2501.00138", "title": "NiaAutoARM: Automated generation and evaluation of Association Rule\n  Mining pipelines", "abstract": "  The Numerical Association Rule Mining paradigm that includes concurrent\ndealing with numerical and categorical attributes is beneficial for discovering\nassociations from datasets consisting of both features. The process is not\nconsidered as easy since it incorporates several processing steps running\nsequentially that form an entire pipeline, e.g., preprocessing, algorithm\nselection, hyper-parameter optimization, and the definition of metrics\nevaluating the quality of the association rule. In this paper, we proposed a\nnovel Automated Machine Learning method, NiaAutoARM, for constructing the full\nassociation rule mining pipelines based on stochastic population-based\nmeta-heuristics automatically. Along with the theoretical representation of the\nproposed method, we also present a comprehensive experimental evaluation of the\nproposed method.\n", "date": "2024-12-30", "categories": []}
{"id": "2206.03070", "title": "SubStrat: A Subset-Based Strategy for Faster AutoML", "abstract": "  Automated machine learning (AutoML) frameworks have become important tools in\nthe data scientists' arsenal, as they dramatically reduce the manual work\ndevoted to the construction of ML pipelines. Such frameworks intelligently\nsearch among millions of possible ML pipelines - typically containing feature\nengineering, model selection and hyper parameters tuning steps - and finally\noutput an optimal pipeline in terms of predictive accuracy. However, when the\ndataset is large, each individual configuration takes longer to execute,\ntherefore the overall AutoML running times become increasingly high. To this\nend, we present SubStrat, an AutoML optimization strategy that tackles the data\nsize, rather than configuration space. It wraps existing AutoML tools, and\ninstead of executing them directly on the entire dataset, SubStrat uses a\ngenetic-based algorithm to find a small yet representative data subset which\npreserves a particular characteristic of the full data. It then employs the\nAutoML tool on the small subset, and finally, it refines the resulted pipeline\nby executing a restricted, much shorter, AutoML process on the large dataset.\nOur experimental results, performed on two popular AutoML frameworks,\nAuto-Sklearn and TPOT, show that SubStrat reduces their running times by 79%\n(on average), with less than 2% average loss in the accuracy of the resulted ML\npipeline.\n", "date": "2022-06-07", "categories": []}
{"id": "2402.15351", "title": "AutoMMLab: Automatically Generating Deployable Models from Language\n  Instructions for Computer Vision Tasks", "abstract": "  Automated machine learning (AutoML) is a collection of techniques designed to\nautomate the machine learning development process. While traditional AutoML\napproaches have been successfully applied in several critical steps of model\ndevelopment (e.g. hyperparameter optimization), there lacks a AutoML system\nthat automates the entire end-to-end model production workflow for computer\nvision. To fill this blank, we propose a novel request-to-model task, which\ninvolves understanding the user's natural language request and execute the\nentire workflow to output production-ready models. This empowers non-expert\nindividuals to easily build task-specific models via a user-friendly language\ninterface. To facilitate development and evaluation, we develop a new\nexperimental platform called AutoMMLab and a new benchmark called LAMP for\nstudying key components in the end-to-end request-to-model pipeline.\nHyperparameter optimization (HPO) is one of the most important components for\nAutoML. Traditional approaches mostly rely on trial-and-error, leading to\ninefficient parameter search. To solve this problem, we propose a novel\nLLM-based HPO algorithm, called HPO-LLaMA. Equipped with extensive knowledge\nand experience in model hyperparameter tuning, HPO-LLaMA achieves significant\nimprovement of HPO efficiency. Dataset and code are available at\nhttps://github.com/yang-ze-kang/AutoMMLab.\n", "date": "2024-02-23", "categories": []}
{"id": "2411.10478", "title": "Large Language Models for Constructing and Optimizing Machine Learning\n  Workflows: A Survey", "abstract": "  Building effective machine learning (ML) workflows to address complex tasks\nis a primary focus of the Automatic ML (AutoML) community and a critical step\ntoward achieving artificial general intelligence (AGI). Recently, the\nintegration of Large Language Models (LLMs) into ML workflows has shown great\npotential for automating and enhancing various stages of the ML pipeline. This\nsurvey provides a comprehensive and up-to-date review of recent advancements in\nusing LLMs to construct and optimize ML workflows, focusing on key components\nencompassing data and feature engineering, model selection and hyperparameter\noptimization, and workflow evaluation. We discuss both the advantages and\nlimitations of LLM-driven approaches, emphasizing their capacity to streamline\nand enhance ML workflow modeling process through language understanding,\nreasoning, interaction, and generation. Finally, we highlight open challenges\nand propose future research directions to advance the effective application of\nLLMs in ML workflows.\n", "date": "2024-11-11", "categories": []}
{"id": "2412.16243", "title": "Bag of Tricks for Multimodal AutoML with Image, Text, and Tabular Data", "abstract": "  This paper studies the best practices for automatic machine learning\n(AutoML). While previous AutoML efforts have predominantly focused on unimodal\ndata, the multimodal aspect remains under-explored. Our study delves into\nclassification and regression problems involving flexible combinations of\nimage, text, and tabular data. We curate a benchmark comprising 22 multimodal\ndatasets from diverse real-world applications, encompassing all 4 combinations\nof the 3 modalities. Across this benchmark, we scrutinize design choices\nrelated to multimodal fusion strategies, multimodal data augmentation,\nconverting tabular data into text, cross-modal alignment, and handling missing\nmodalities. Through extensive experimentation and analysis, we distill a\ncollection of effective strategies and consolidate them into a unified\npipeline, achieving robust performance on diverse datasets.\n", "date": "2024-12-19", "categories": []}
{"id": "2405.16203", "title": "Evolutionary Large Language Model for Automated Feature Transformation", "abstract": "  Feature transformation aims to reconstruct the feature space of raw features\nto enhance the performance of downstream models. However, the exponential\ngrowth in the combinations of features and operations poses a challenge, making\nit difficult for existing methods to efficiently explore a wide space.\nAdditionally, their optimization is solely driven by the accuracy of downstream\nmodels in specific domains, neglecting the acquisition of general feature\nknowledge. To fill this research gap, we propose an evolutionary LLM framework\nfor automated feature transformation. This framework consists of two parts: 1)\nconstructing a multi-population database through an RL data collector while\nutilizing evolutionary algorithm strategies for database maintenance, and 2)\nutilizing the ability of Large Language Model (LLM) in sequence understanding,\nwe employ few-shot prompts to guide LLM in generating superior samples based on\nfeature transformation sequence distinction. Leveraging the multi-population\ndatabase initially provides a wide search scope to discover excellent\npopulations. Through culling and evolution, the high-quality populations are\nafforded greater opportunities, thereby furthering the pursuit of optimal\nindividuals. Through the integration of LLMs with evolutionary algorithms, we\nachieve efficient exploration within a vast space, while harnessing feature\nknowledge to propel optimization, thus realizing a more adaptable search\nparadigm. Finally, we empirically demonstrate the effectiveness and generality\nof our proposed method.\n", "date": "2024-05-25", "categories": []}
{"id": "2412.07000", "title": "Extreme AutoML: Analysis of Classification, Regression, and NLP\n  Performance", "abstract": "  Utilizing machine learning techniques has always required choosing\nhyperparameters. This is true whether one uses a classical technique such as a\nKNN or very modern neural networks such as Deep Learning. Though in many\napplications, hyperparameters are chosen by hand, automated methods have become\nincreasingly more common. These automated methods have become collectively\nknown as automated machine learning, or AutoML. Several automated selection\nalgorithms have shown similar or improved performance over state-of-the-art\nmethods. This breakthrough has led to the development of cloud-based services\nlike Google AutoML, which is based on Deep Learning and is widely considered to\nbe the industry leader in AutoML services. Extreme Learning Machines (ELMs) use\na fundamentally different type of neural architecture, producing better results\nat a significantly discounted computational cost. We benchmark the Extreme\nAutoML technology against Google's AutoML using several popular classification\ndata sets from the University of California at Irvine's (UCI) repository, and\nseveral other data sets, observing significant advantages for Extreme AutoML in\naccuracy, Jaccard Indices, the variance of Jaccard Indices across classes (i.e.\nclass variance) and training times.\n", "date": "2024-12-09", "categories": []}
{"id": "2412.03666", "title": "Hyperparameter Tuning Through Pessimistic Bilevel Optimization", "abstract": "  Automated hyperparameter search in machine learning, especially for deep\nlearning models, is typically formulated as a bilevel optimization problem,\nwith hyperparameter values determined by the upper level and the model learning\nachieved by the lower-level problem. Most of the existing bilevel optimization\nsolutions either assume the uniqueness of the optimal training model given\nhyperparameters or adopt an optimistic view when the non-uniqueness issue\nemerges. Potential model uncertainty may arise when training complex models\nwith limited data, especially when the uniqueness assumption is violated. Thus,\nthe suitability of the optimistic view underlying current bilevel\nhyperparameter optimization solutions is questionable. In this paper, we\npropose pessimistic bilevel hyperparameter optimization to assure appropriate\nouter-level hyperparameters to better generalize the inner-level learned\nmodels, by explicitly incorporating potential uncertainty of the inner-level\nsolution set. To solve the resulting computationally challenging pessimistic\nbilevel optimization problem, we develop a novel relaxation-based approximation\nmethod. It derives pessimistic solutions with more robust prediction models. In\nour empirical studies of automated hyperparameter search for binary linear\nclassifiers, pessimistic solutions have demonstrated better prediction\nperformances than optimistic counterparts when we have limited training data or\nperturbed testing data, showing the necessity of considering pessimistic\nsolutions besides existing optimistic ones.\n", "date": "2024-12-04", "categories": []}
{"id": "2410.03736", "title": "CliMB: An AI-enabled Partner for Clinical Predictive Modeling", "abstract": "  Despite its significant promise and continuous technical advances, real-world\napplications of artificial intelligence (AI) remain limited. We attribute this\nto the \"domain expert-AI-conundrum\": while domain experts, such as clinician\nscientists, should be able to build predictive models such as risk scores, they\nface substantial barriers in accessing state-of-the-art (SOTA) tools. While\nautomated machine learning (AutoML) has been proposed as a partner in clinical\npredictive modeling, many additional requirements need to be fulfilled to make\nmachine learning accessible for clinician scientists.\n  To address this gap, we introduce CliMB, a no-code AI-enabled partner\ndesigned to empower clinician scientists to create predictive models using\nnatural language. CliMB guides clinician scientists through the entire medical\ndata science pipeline, thus empowering them to create predictive models from\nreal-world data in just one conversation. CliMB also creates structured reports\nand interpretable visuals. In evaluations involving clinician scientists and\nsystematic comparisons against a baseline GPT-4, CliMB consistently\ndemonstrated superior performance in key areas such as planning, error\nprevention, code execution, and model performance. Moreover, in blinded\nassessments involving 45 clinicians from diverse specialties and career stages,\nmore than 80% preferred CliMB over GPT-4. Overall, by providing a no-code\ninterface with clear guidance and access to SOTA methods in the fields of\ndata-centric AI, AutoML, and interpretable ML, CliMB empowers clinician\nscientists to build robust predictive models.\n  The proof-of-concept version of CliMB is available as open-source software on\nGitHub: https://github.com/vanderschaarlab/climb.\n", "date": "2024-09-30", "categories": []}
{"id": "2411.03731", "title": "Reducing Hyperparameter Tuning Costs in ML, Vision and Language Model\n  Training Pipelines via Memoization-Awareness", "abstract": "  The training or fine-tuning of machine learning, vision, and language models\nis often implemented as a pipeline: a sequence of stages encompassing data\npreparation, model training and evaluation. In this paper, we exploit pipeline\nstructures to reduce the cost of hyperparameter tuning for model\ntraining/fine-tuning, which is particularly valuable for language models given\ntheir high costs in GPU-days. We propose a \"memoization-aware\" Bayesian\nOptimization (BO) algorithm, EEIPU, that works in tandem with a pipeline\ncaching system, allowing it to evaluate significantly more hyperparameter\ncandidates per GPU-day than other tuning algorithms. The result is\nbetter-quality hyperparameters in the same amount of search time, or\nequivalently, reduced search time to reach the same hyperparameter quality. In\nour benchmarks on machine learning (model ensembles), vision (convolutional\narchitecture) and language (T5 architecture) pipelines, we compare EEIPU\nagainst recent BO algorithms: EEIPU produces an average of $103\\%$ more\nhyperparameter candidates (within the same budget), and increases the\nvalidation metric by an average of $108\\%$ more than other algorithms (where\nthe increase is measured starting from the end of warm-up iterations).\n", "date": "2024-11-06", "categories": []}
{"id": "2410.17469", "title": "AdaptoML-UX: An Adaptive User-centered GUI-based AutoML Toolkit for\n  Non-AI Experts and HCI Researchers", "abstract": "  The increasing integration of machine learning across various domains has\nunderscored the necessity for accessible systems that non-experts can utilize\neffectively. To address this need, the field of automated machine learning\n(AutoML) has developed tools to simplify the construction and optimization of\nML pipelines. However, existing AutoML solutions often lack efficiency in\ncreating online pipelines and ease of use for Human-Computer Interaction (HCI)\napplications. Therefore, in this paper, we introduce AdaptoML-UX, an adaptive\nframework that incorporates automated feature engineering, machine learning,\nand incremental learning to assist non-AI experts in developing robust,\nuser-centered ML models. Our toolkit demonstrates the capability to adapt\nefficiently to diverse problem domains and datasets, particularly in HCI,\nthereby reducing the necessity for manual experimentation and conserving time\nand resources. Furthermore, it supports model personalization through\nincremental learning, customizing models to individual user behaviors. HCI\nresearchers can employ AdaptoML-UX\n(\\url{https://github.com/MichaelSargious/AdaptoML_UX}) without requiring\nspecialized expertise, as it automates the selection of algorithms, feature\nengineering, and hyperparameter tuning based on the unique characteristics of\nthe data.\n", "date": "2024-10-22", "categories": []}
{"id": "2410.17238", "title": "SELA: Tree-Search Enhanced LLM Agents for Automated Machine Learning", "abstract": "  Automated Machine Learning (AutoML) approaches encompass traditional methods\nthat optimize fixed pipelines for model selection and ensembling, as well as\nnewer LLM-based frameworks that autonomously build pipelines. While LLM-based\nagents have shown promise in automating machine learning tasks, they often\ngenerate low-diversity and suboptimal code, even after multiple iterations. To\novercome these limitations, we introduce Tree-Search Enhanced LLM Agents\n(SELA), an innovative agent-based system that leverages Monte Carlo Tree Search\n(MCTS) to optimize the AutoML process. By representing pipeline configurations\nas trees, our framework enables agents to conduct experiments intelligently and\niteratively refine their strategies, facilitating a more effective exploration\nof the machine learning solution space. This novel approach allows SELA to\ndiscover optimal pathways based on experimental feedback, improving the overall\nquality of the solutions. In an extensive evaluation across 20 machine learning\ndatasets, we compare the performance of traditional and agent-based AutoML\nmethods, demonstrating that SELA achieves a win rate of 65% to 80% against each\nbaseline across all datasets. These results underscore the significant\npotential of agent-based strategies in AutoML, offering a fresh perspective on\ntackling complex machine learning challenges.\n", "date": "2024-10-22", "categories": []}
{"id": "2410.14743", "title": "Efficient Deep Learning Board: Training Feedback Is Not All You Need", "abstract": "  Current automatic deep learning (i.e., AutoDL) frameworks rely on training\nfeedback from actual runs, which often hinder their ability to provide quick\nand clear performance predictions for selecting suitable DL systems. To address\nthis issue, we propose EfficientDL, an innovative deep learning board designed\nfor automatic performance prediction and component recommendation. EfficientDL\ncan quickly and precisely recommend twenty-seven system components and predict\nthe performance of DL models without requiring any training feedback. The magic\nof no training feedback comes from our proposed comprehensive,\nmulti-dimensional, fine-grained system component dataset, which enables us to\ndevelop a static performance prediction model and comprehensive optimized\ncomponent recommendation algorithm (i.e., {\\alpha}\\b{eta}-BO search), removing\nthe dependency on actually running parameterized models during the traditional\noptimization search process. The simplicity and power of EfficientDL stem from\nits compatibility with most DL models. For example, EfficientDL operates\nseamlessly with mainstream models such as ResNet50, MobileNetV3,\nEfficientNet-B0, MaxViT-T, Swin-B, and DaViT-T, bringing competitive\nperformance improvements. Besides, experimental results on the CIFAR-10 dataset\nreveal that EfficientDL outperforms existing AutoML tools in both accuracy and\nefficiency (approximately 20 times faster along with 1.31% Top-1 accuracy\nimprovement than the cutting-edge methods). Source code, pretrained models, and\ndatasets are available at https://github.com/OpenSELab/EfficientDL.\n", "date": "2024-10-17", "categories": []}
{"id": "2410.12841", "title": "UniAutoML: A Human-Centered Framework for Unified Discriminative and\n  Generative AutoML with Large Language Models", "abstract": "  Automated Machine Learning (AutoML) has simplified complex ML processes such\nas data pre-processing, model selection, and hyper-parameter searching.\nHowever, traditional AutoML frameworks focus solely on discriminative tasks,\noften falling short in tackling AutoML for generative models. Additionally,\nthese frameworks lack interpretability and user engagement during the training\nprocess, primarily due to the absence of human-centered design. It leads to a\nlack of transparency in final decision-making and limited user control,\npotentially reducing trust and adoption of AutoML methods. To address these\nlimitations, we introduce UniAutoML, a human-centered AutoML framework that\nleverages Large Language Models (LLMs) to unify AutoML for both discriminative\n(e.g., Transformers and CNNs for classification or regression tasks) and\ngenerative tasks (e.g., fine-tuning diffusion models or LLMs). The\nhuman-centered design of UniAutoML innovatively features a conversational user\ninterface (CUI) that facilitates natural language interactions, providing users\nwith real-time guidance, feedback, and progress updates for better\ninterpretability. This design enhances transparency and user control throughout\nthe AutoML training process, allowing users to seamlessly break down or modify\nthe model being trained. To mitigate potential risks associated with LLM\ngenerated content, UniAutoML incorporates a safety guardline that filters\ninputs and censors outputs. We evaluated UniAutoML's performance and usability\nthrough experiments on eight diverse datasets and user studies involving 25\nparticipants, demonstrating that UniAutoML not only enhances performance but\nalso improves user control and trust. Our human-centered design bridges the gap\nbetween AutoML capabilities and user understanding, making ML more accessible\nto a broader audience.\n", "date": "2024-10-09", "categories": []}
{"id": "2410.12435", "title": "Approaching Metaheuristic Deep Learning Combos for Automated Data Mining", "abstract": "  Lack of data on which to perform experimentation is a recurring issue in many\nareas of research, particularly in machine learning. The inability of most\nautomated data mining techniques to be generalized to all types of data is\ninherently related with their dependency on those types which deems them\nineffective against anything slightly different. Meta-heuristics are algorithms\nwhich attempt to optimize some solution independently of the type of data used,\nwhilst classifiers or neural networks focus on feature extrapolation and\ndimensionality reduction to fit some model onto data arranged in a particular\nway. These two algorithmic fields encompass a group of characteristics which\nwhen combined are seemingly capable of achieving data mining regardless of how\nit is arranged. To this end, this work proposes a means of combining\nmeta-heuristic methods with conventional classifiers and neural networks in\norder to perform automated data mining. Experiments on the MNIST dataset for\nhandwritten digit recognition were performed and it was empirically observed\nthat using a ground truth labeled dataset's validation accuracy is inadequate\nfor correcting labels of other previously unseen data instances.\n", "date": "2024-10-16", "categories": []}
{"id": "2410.09239", "title": "Scaling Gaussian Processes for Learning Curve Prediction via Latent\n  Kronecker Structure", "abstract": "  A key task in AutoML is to model learning curves of machine learning models\njointly as a function of model hyper-parameters and training progression. While\nGaussian processes (GPs) are suitable for this task, na\\\"ive GPs require\n$\\mathcal{O}(n^3m^3)$ time and $\\mathcal{O}(n^2 m^2)$ space for $n$\nhyper-parameter configurations and $\\mathcal{O}(m)$ learning curve observations\nper hyper-parameter. Efficient inference via Kronecker structure is typically\nincompatible with early-stopping due to missing learning curve values. We\nimpose $\\textit{latent Kronecker structure}$ to leverage efficient product\nkernels while handling missing values. In particular, we interpret the joint\ncovariance matrix of observed values as the projection of a latent Kronecker\nproduct. Combined with iterative linear solvers and structured matrix-vector\nmultiplication, our method only requires $\\mathcal{O}(n^3 + m^3)$ time and\n$\\mathcal{O}(n^2 + m^2)$ space. We show that our GP model can match the\nperformance of a Transformer on a learning curve prediction task.\n", "date": "2024-10-11", "categories": []}
{"id": "2410.07696", "title": "Meta-Learning from Learning Curves for Budget-Limited Algorithm\n  Selection", "abstract": "  Training a large set of machine learning algorithms to convergence in order\nto select the best-performing algorithm for a dataset is computationally\nwasteful. Moreover, in a budget-limited scenario, it is crucial to carefully\nselect an algorithm candidate and allocate a budget for training it, ensuring\nthat the limited budget is optimally distributed to favor the most promising\ncandidates. Casting this problem as a Markov Decision Process, we propose a\nnovel framework in which an agent must select in the process of learning the\nmost promising algorithm without waiting until it is fully trained. At each\ntime step, given an observation of partial learning curves of algorithms, the\nagent must decide whether to allocate resources to further train the most\npromising algorithm (exploitation), to wake up another algorithm previously put\nto sleep, or to start training a new algorithm (exploration). In addition, our\nframework allows the agent to meta-learn from learning curves on past datasets\nalong with dataset meta-features and algorithm hyperparameters. By\nincorporating meta-learning, we aim to avoid myopic decisions based solely on\npremature learning curves on the dataset at hand. We introduce two benchmarks\nof learning curves that served in international competitions at WCCI'22 and\nAutoML-conf'22, of which we analyze the results. Our findings show that both\nmeta-learning and the progression of learning curves enhance the algorithm\nselection process, as evidenced by methods of winning teams and our DDQN\nbaseline, compared to heuristic baselines or a random search. Interestingly,\nour cost-effective baseline, which selects the best-performing algorithm w.r.t.\na small budget, can perform decently when learning curves do not intersect\nfrequently.\n", "date": "2024-10-10", "categories": []}
{"id": "2403.04086", "title": "Automated Multi-Task Learning for Joint Disease Prediction on Electronic\n  Health Records", "abstract": "  In the realm of big data and digital healthcare, Electronic Health Records\n(EHR) have become a rich source of information with the potential to improve\npatient care and medical research. In recent years, machine learning models\nhave proliferated for analyzing EHR data to predict patients future health\nconditions. Among them, some studies advocate for multi-task learning (MTL) to\njointly predict multiple target diseases for improving the prediction\nperformance over single task learning. Nevertheless, current MTL frameworks for\nEHR data have significant limitations due to their heavy reliance on human\nexperts to identify task groups for joint training and design model\narchitectures. To reduce human intervention and improve the framework design,\nwe propose an automated approach named AutoDP, which can search for the optimal\nconfiguration of task grouping and architectures simultaneously. To tackle the\nvast joint search space encompassing task combinations and architectures, we\nemploy surrogate model-based optimization, enabling us to efficiently discover\nthe optimal solution. Experimental results on real-world EHR data demonstrate\nthe efficacy of the proposed AutoDP framework. It achieves significant\nperformance improvements over both hand-crafted and automated state-of-the-art\nmethods, also maintains a feasible search cost at the same time. Source code\ncan be found via the link: \\url{https://github.com/SH-Src/AutoDP}.\n", "date": "2024-03-06", "categories": []}
{"id": "2410.06408", "title": "Automating Data Science Pipelines with Tensor Completion", "abstract": "  Hyperparameter optimization is an essential component in many data science\npipelines and typically entails exhaustive time and resource-consuming\ncomputations in order to explore the combinatorial search space. Similar to\nthis problem, other key operations in data science pipelines exhibit the exact\nsame properties. Important examples are: neural architecture search, where the\ngoal is to identify the best design choices for a neural network, and query\ncardinality estimation, where given different predicate values for a SQL query\nthe goal is to estimate the size of the output. In this paper, we abstract away\nthose essential components of data science pipelines and we model them as\ninstances of tensor completion, where each variable of the search space\ncorresponds to one mode of the tensor, and the goal is to identify all missing\nentries of the tensor, corresponding to all combinations of variable values,\nstarting from a very small sample of observed entries. In order to do so, we\nfirst conduct a thorough experimental evaluation of existing state-of-the-art\ntensor completion techniques and introduce domain-inspired adaptations (such as\nsmoothness across the discretized variable space) and an ensemble technique\nwhich is able to achieve state-of-the-art performance. We extensively evaluate\nexisting and proposed methods in a number of datasets generated corresponding\nto (a) hyperparameter optimization for non-neural network models, (b) neural\narchitecture search, and (c) variants of query cardinality estimation,\ndemonstrating the effectiveness of tensor completion as a tool for automating\ndata science pipelines. Furthermore, we release our generated datasets and code\nin order to provide benchmarks for future work on this topic.\n", "date": "2024-10-08", "categories": []}
{"id": "2409.16218", "title": "Problem-oriented AutoML in Clustering", "abstract": "  The Problem-oriented AutoML in Clustering (PoAC) framework introduces a\nnovel, flexible approach to automating clustering tasks by addressing the\nshortcomings of traditional AutoML solutions. Conventional methods often rely\non predefined internal Clustering Validity Indexes (CVIs) and static\nmeta-features, limiting their adaptability and effectiveness across diverse\nclustering tasks. In contrast, PoAC establishes a dynamic connection between\nthe clustering problem, CVIs, and meta-features, allowing users to customize\nthese components based on the specific context and goals of their task. At its\ncore, PoAC employs a surrogate model trained on a large meta-knowledge base of\nprevious clustering datasets and solutions, enabling it to infer the quality of\nnew clustering pipelines and synthesize optimal solutions for unseen datasets.\nUnlike many AutoML frameworks that are constrained by fixed evaluation metrics\nand algorithm sets, PoAC is algorithm-agnostic, adapting seamlessly to\ndifferent clustering problems without requiring additional data or retraining.\nExperimental results demonstrate that PoAC not only outperforms\nstate-of-the-art frameworks on a variety of datasets but also excels in\nspecific tasks such as data visualization, and highlight its ability to\ndynamically adjust pipeline configurations based on dataset complexity.\n", "date": "2024-09-24", "categories": []}
{"id": "2409.04585", "title": "CubicML: Automated ML for Large ML Systems Co-design with ML Prediction\n  of Performance", "abstract": "  Scaling up deep learning models has been proven effective to improve\nintelligence of machine learning (ML) models, especially for industry\nrecommendation models and large language models. The co-design of large\ndistributed ML systems and algorithms (to maximize training performance) plays\na pivotal role for its success. As it scales, the number of co-design\nhyper-parameters grows rapidly which brings challenges to feasibly find the\noptimal setup for system performance maximization. In this paper, we propose\nCubicML which uses ML to automatically optimize training performance of large\ndistributed ML systems. In CubicML, we use an ML model as a proxy to predict\nthe training performance for search efficiency and performance modeling\nflexibility. We proved that CubicML can effectively optimize training speed of\nin-house ads recommendation models with 73 billion parameters and large\nlanguage models up to 405 billion parameters at Meta.\n", "date": "2024-09-06", "categories": []}
{"id": "2409.08823", "title": "AutoIRT: Calibrating Item Response Theory Models with Automated Machine\n  Learning", "abstract": "  Item response theory (IRT) is a class of interpretable factor models that are\nwidely used in computerized adaptive tests (CATs), such as language proficiency\ntests. Traditionally, these are fit using parametric mixed effects models on\nthe probability of a test taker getting the correct answer to a test item\n(i.e., question). Neural net extensions of these models, such as BertIRT,\nrequire specialized architectures and parameter tuning. We propose a multistage\nfitting procedure that is compatible with out-of-the-box Automated Machine\nLearning (AutoML) tools. It is based on a Monte Carlo EM (MCEM) outer loop with\na two stage inner loop, which trains a non-parametric AutoML grade model using\nitem features followed by an item specific parametric model. This greatly\naccelerates the modeling workflow for scoring tests. We demonstrate its\neffectiveness by applying it to the Duolingo English Test, a high stakes,\nonline English proficiency test. We show that the resulting model is typically\nmore well calibrated, gets better predictive performance, and more accurate\nscores than existing methods (non-explanatory IRT models and explanatory IRT\nmodels like BERT-IRT). Along the way, we provide a brief survey of machine\nlearning methods for calibration of item parameters for CATs.\n", "date": "2024-09-13", "categories": []}
{"id": "2405.03727", "title": "Large Language Models Synergize with Automated Machine Learning", "abstract": "  Recently, program synthesis driven by large language models (LLMs) has become\nincreasingly popular. However, program synthesis for machine learning (ML)\ntasks still poses significant challenges. This paper explores a novel form of\nprogram synthesis, targeting ML programs, by combining LLMs and automated\nmachine learning (autoML). Specifically, our goal is to fully automate the\ngeneration and optimization of the code of the entire ML workflow, from data\npreparation to modeling and post-processing, utilizing only textual\ndescriptions of the ML tasks. To manage the length and diversity of ML\nprograms, we propose to break each ML program into smaller, manageable parts.\nEach part is generated separately by the LLM, with careful consideration of\ntheir compatibilities. To ensure compatibilities, we design a testing technique\nfor ML programs. Unlike traditional program synthesis, which typically relies\non binary evaluations (i.e., correct or incorrect), evaluating ML programs\nnecessitates more than just binary judgments. Our approach automates the\nnumerical evaluation and optimization of these programs, selecting the best\ncandidates through autoML techniques. In experiments across various ML tasks,\nour method outperforms existing methods in 10 out of 12 tasks for generating ML\nprograms. In addition, autoML significantly improves the performance of the\ngenerated ML programs. In experiments, given the textual task description, our\nmethod, Text-to-ML, generates the complete and optimized ML program in a fully\nautonomous process. The implementation of our method is available at\nhttps://github.com/JLX0/llm-automl.\n", "date": "2024-05-06", "categories": []}
{"id": "2409.04789", "title": "forester: A Tree-Based AutoML Tool in R", "abstract": "  The majority of automated machine learning (AutoML) solutions are developed\nin Python, however a large percentage of data scientists are associated with\nthe R language. Unfortunately, there are limited R solutions available.\nMoreover high entry level means they are not accessible to everyone, due to\nrequired knowledge about machine learning (ML). To fill this gap, we present\nthe forester package, which offers ease of use regardless of the user's\nproficiency in the area of machine learning.\n  The forester is an open-source AutoML package implemented in R designed for\ntraining high-quality tree-based models on tabular data. It fully supports\nbinary and multiclass classification, regression, and partially survival\nanalysis tasks. With just a few functions, the user is capable of detecting\nissues regarding the data quality, preparing the preprocessing pipeline,\ntraining and tuning tree-based models, evaluating the results, and creating the\nreport for further analysis.\n", "date": "2024-09-07", "categories": []}
{"id": "2409.05258", "title": "Towards Automated Machine Learning Research", "abstract": "  This paper explores a top-down approach to automating incremental advances in\nmachine learning research through component-level innovation, facilitated by\nLarge Language Models (LLMs). Our framework systematically generates novel\ncomponents, validates their feasibility, and evaluates their performance\nagainst existing baselines. A key distinction of this approach lies in how\nthese novel components are generated. Unlike traditional AutoML and NAS\nmethods, which often rely on a bottom-up combinatorial search over predefined,\nhardcoded base components, our method leverages the cross-domain knowledge\nembedded in LLMs to propose new components that may not be confined to any\nhard-coded predefined set. By incorporating a reward model to prioritize\npromising hypotheses, we aim to improve the efficiency of the hypothesis\ngeneration and evaluation process. We hope this approach offers a new avenue\nfor exploration and contributes to the ongoing dialogue in the field.\n", "date": "2024-09-09", "categories": []}
{"id": "2409.01446", "title": "Landscape-Aware Automated Algorithm Configuration using Multi-output\n  Mixed Regression and Classification", "abstract": "  In landscape-aware algorithm selection problem, the effectiveness of\nfeature-based predictive models strongly depends on the representativeness of\ntraining data for practical applications. In this work, we investigate the\npotential of randomly generated functions (RGF) for the model training, which\ncover a much more diverse set of optimization problem classes compared to the\nwidely-used black-box optimization benchmarking (BBOB) suite. Correspondingly,\nwe focus on automated algorithm configuration (AAC), that is, selecting the\nbest suited algorithm and fine-tuning its hyperparameters based on the\nlandscape features of problem instances. Precisely, we analyze the performance\nof dense neural network (NN) models in handling the multi-output mixed\nregression and classification tasks using different training data sets, such as\nRGF and many-affine BBOB (MA-BBOB) functions. Based on our results on the BBOB\nfunctions in 5d and 20d, near optimal configurations can be identified using\nthe proposed approach, which can most of the time outperform the off-the-shelf\ndefault configuration considered by practitioners with limited knowledge about\nAAC. Furthermore, the predicted configurations are competitive against the\nsingle best solver in many cases. Overall, configurations with better\nperformance can be best identified by using NN models trained on a combination\nof RGF and MA-BBOB functions.\n", "date": "2024-09-02", "categories": []}
{"id": "2306.04750", "title": "AutoML Systems For Medical Imaging", "abstract": "  The integration of machine learning in medical image analysis can greatly\nenhance the quality of healthcare provided by physicians. The combination of\nhuman expertise and computerized systems can result in improved diagnostic\naccuracy. An automated machine learning approach simplifies the creation of\ncustom image recognition models by utilizing neural architecture search and\ntransfer learning techniques. Medical imaging techniques are used to\nnon-invasively create images of internal organs and body parts for diagnostic\nand procedural purposes. This article aims to highlight the potential\napplications, strategies, and techniques of AutoML in medical imaging through\ntheoretical and empirical evidence.\n", "date": "2023-06-07", "categories": []}
{"id": "2311.02971", "title": "TabRepo: A Large Scale Repository of Tabular Model Evaluations and its\n  AutoML Applications", "abstract": "  We introduce TabRepo, a new dataset of tabular model evaluations and\npredictions. TabRepo contains the predictions and metrics of 1310 models\nevaluated on 200 classification and regression datasets. We illustrate the\nbenefit of our dataset in multiple ways. First, we show that it allows to\nperform analysis such as comparing Hyperparameter Optimization against current\nAutoML systems while also considering ensembling at marginal cost by using\nprecomputed model predictions. Second, we show that our dataset can be readily\nleveraged to perform transfer-learning. In particular, we show that applying\nstandard transfer-learning techniques allows to outperform current\nstate-of-the-art tabular systems in accuracy, runtime and latency.\n", "date": "2023-11-06", "categories": []}
{"id": "2408.14331", "title": "Automated Machine Learning in Insurance", "abstract": "  Machine Learning (ML) has gained popularity in actuarial research and\ninsurance industrial applications. However, the performance of most ML tasks\nheavily depends on data preprocessing, model selection, and hyperparameter\noptimization, which are considered to be intensive in terms of domain\nknowledge, experience, and manual labor. Automated Machine Learning (AutoML)\naims to automatically complete the full life-cycle of ML tasks and provides\nstate-of-the-art ML models without human intervention or supervision. This\npaper introduces an AutoML workflow that allows users without domain knowledge\nor prior experience to achieve robust and effortless ML deployment by writing\nonly a few lines of code. This proposed AutoML is specifically tailored for the\ninsurance application, with features like the balancing step in data\npreprocessing, ensemble pipelines, and customized loss functions. These\nfeatures are designed to address the unique challenges of the insurance domain,\nincluding the imbalanced nature of common insurance datasets. The full code and\ndocumentation are available on the GitHub repository.\n(https://github.com/PanyiDong/InsurAutoML)\n", "date": "2024-08-26", "categories": []}
{"id": "2408.02280", "title": "Hardware Aware Ensemble Selection for Balancing Predictive Accuracy and\n  Cost", "abstract": "  Automated Machine Learning (AutoML) significantly simplifies the deployment\nof machine learning models by automating tasks from data preprocessing to model\nselection to ensembling. AutoML systems for tabular data often employ post hoc\nensembling, where multiple models are combined to improve predictive accuracy.\nThis typically results in longer inference times, a major limitation in\npractical deployments. Addressing this, we introduce a hardware-aware ensemble\nselection approach that integrates inference time into post hoc ensembling. By\nleveraging an existing framework for ensemble selection with quality diversity\noptimization, our method evaluates ensemble candidates for their predictive\naccuracy and hardware efficiency. This dual focus allows for a balanced\nconsideration of accuracy and operational efficiency. Thus, our approach\nenables practitioners to choose from a Pareto front of accurate and efficient\nensembles. Our evaluation using 83 classification datasets shows that our\napproach sustains competitive accuracy and can significantly improve ensembles'\noperational efficiency. The results of this study provide a foundation for\nextending these principles to additional hardware constraints, setting the\nstage for the development of more resource-efficient AutoML systems.\n", "date": "2024-08-05", "categories": []}
{"id": "2405.03389", "title": "Don't Waste Your Time: Early Stopping Cross-Validation", "abstract": "  State-of-the-art automated machine learning systems for tabular data often\nemploy cross-validation; ensuring that measured performances generalize to\nunseen data, or that subsequent ensembling does not overfit. However, using\nk-fold cross-validation instead of holdout validation drastically increases the\ncomputational cost of validating a single configuration. While ensuring better\ngeneralization and, by extension, better performance, the additional cost is\noften prohibitive for effective model selection within a time budget. We aim to\nmake model selection with cross-validation more effective. Therefore, we study\nearly stopping the process of cross-validation during model selection. We\ninvestigate the impact of early stopping on random search for two algorithms,\nMLP and random forest, across 36 classification datasets. We further analyze\nthe impact of the number of folds by considering 3-, 5-, and 10-folds. In\naddition, we investigate the impact of early stopping with Bayesian\noptimization instead of random search and also repeated cross-validation. Our\nexploratory study shows that even a simple-to-understand and easy-to-implement\nmethod consistently allows model selection to converge faster; in ~94% of all\ndatasets, on average by ~214%. Moreover, stopping cross-validation enables\nmodel selection to explore the search space more exhaustively by considering\n+167% configurations on average within one hour, while also obtaining better\noverall performance.\n", "date": "2024-05-06", "categories": []}
{"id": "2408.00421", "title": "Towards Evolutionary-based Automated Machine Learning for Small Molecule\n  Pharmacokinetic Prediction", "abstract": "  Machine learning (ML) is revolutionising drug discovery by expediting the\nprediction of small molecule properties essential for developing new drugs.\nThese properties -- including absorption, distribution, metabolism and\nexcretion (ADME)-- are crucial in the early stages of drug development since\nthey provide an understanding of the course of the drug in the organism, i.e.,\nthe drug's pharmacokinetics. However, existing methods lack personalisation and\nrely on manually crafted ML algorithms or pipelines, which can introduce\ninefficiencies and biases into the process. To address these challenges, we\npropose a novel evolutionary-based automated ML method (AutoML) specifically\ndesigned for predicting small molecule properties, with a particular focus on\npharmacokinetics. Leveraging the advantages of grammar-based genetic\nprogramming, our AutoML method streamlines the process by automatically\nselecting algorithms and designing predictive pipelines tailored to the\nparticular characteristics of input molecular data. Results demonstrate\nAutoML's effectiveness in selecting diverse ML algorithms, resulting in\ncomparable or even improved predictive performances compared to conventional\napproaches. By offering personalised ML-driven pipelines, our method promises\nto enhance small molecule research in drug discovery, providing researchers\nwith a valuable tool for accelerating the development of novel therapeutic\ndrugs.\n", "date": "2024-08-01", "categories": []}
{"id": "2408.00665", "title": "AutoM3L: An Automated Multimodal Machine Learning Framework with Large\n  Language Models", "abstract": "  Automated Machine Learning (AutoML) offers a promising approach to streamline\nthe training of machine learning models. However, existing AutoML frameworks\nare often limited to unimodal scenarios and require extensive manual\nconfiguration. Recent advancements in Large Language Models (LLMs) have\nshowcased their exceptional abilities in reasoning, interaction, and code\ngeneration, presenting an opportunity to develop a more automated and\nuser-friendly framework. To this end, we introduce AutoM3L, an innovative\nAutomated Multimodal Machine Learning framework that leverages LLMs as\ncontrollers to automatically construct multimodal training pipelines. AutoM3L\ncomprehends data modalities and selects appropriate models based on user\nrequirements, providing automation and interactivity. By eliminating the need\nfor manual feature engineering and hyperparameter optimization, our framework\nsimplifies user engagement and enables customization through directives,\naddressing the limitations of previous rule-based AutoML approaches. We\nevaluate the performance of AutoM3L on six diverse multimodal datasets spanning\nclassification, regression, and retrieval tasks, as well as a comprehensive set\nof unimodal datasets. The results demonstrate that AutoM3L achieves competitive\nor superior performance compared to traditional rule-based AutoML methods.\nFurthermore, a user study highlights the user-friendliness and usability of our\nframework, compared to the rule-based AutoML methods.\n", "date": "2024-08-01", "categories": []}
{"id": "2404.11276", "title": "Towards Data-Centric Automatic R&D", "abstract": "  The progress of humanity is driven by those successful discoveries\naccompanied by countless failed experiments. Researchers often seek the\npotential research directions by reading and then verifying them through\nexperiments. The process imposes a significant burden on researchers. In the\npast decade, the data-driven black-box deep learning method has demonstrated\nits effectiveness in a wide range of real-world scenarios, which exacerbates\nthe experimental burden of researchers and thus renders the potential\nsuccessful discoveries veiled. Therefore, automating such a research and\ndevelopment (R&D) process is an urgent need. In this paper, we serve as the\nfirst effort to formalize the goal by proposing a Real-world Data-centric\nautomatic R&D Benchmark, namely RD2Bench. RD2Bench benchmarks all the\noperations in data-centric automatic R&D (D-CARD) as a whole to navigate future\nwork toward our goal directly. We focus on evaluating the interaction and\nsynergistic effects of various model capabilities and aiding in selecting\nwell-performing trustworthy models. Although RD2Bench is very challenging to\nthe state-of-the-art (SOTA) large language model (LLM) named GPT-4, indicating\nample research opportunities and more research efforts, LLMs possess promising\npotential to bring more significant development to D-CARD: They are able to\nimplement some simple methods without adopting any additional techniques. We\nappeal to future work to take developing techniques for tackling automatic R&D\ninto consideration, thus bringing the opportunities of the potential\nrevolutionary upgrade to human productivity.\n", "date": "2024-04-17", "categories": []}
{"id": "2407.18690", "title": "Collaborative Evolving Strategy for Automatic Data-Centric Development", "abstract": "  Artificial Intelligence (AI) significantly influences many fields, largely\nthanks to the vast amounts of high-quality data for machine learning models.\nThe emphasis is now on a data-centric AI strategy, prioritizing data\ndevelopment over model design progress. Automating this process is crucial. In\nthis paper, we serve as the first work to introduce the automatic data-centric\ndevelopment (AD^2) task and outline its core challenges, which require\ndomain-experts-like task scheduling and implementation capability, largely\nunexplored by previous work.\n  By leveraging the strong complex problem-solving capabilities of large\nlanguage models (LLMs), we propose an LLM-based autonomous agent, equipped with\na strategy named Collaborative Knowledge-STudying-Enhanced Evolution by\nRetrieval (Co-STEER), to simultaneously address all the challenges.\nSpecifically, our proposed Co-STEER agent enriches its domain knowledge through\nour proposed evolving strategy and develops both its scheduling and\nimplementation skills by accumulating and retrieving domain-specific practical\nexperience. With an improved schedule, the capability for implementation\naccelerates. Simultaneously, as implementation feedback becomes more thorough,\nthe scheduling accuracy increases. These two capabilities evolve together\nthrough practical feedback, enabling a collaborative evolution process.\n  Extensive experimental results demonstrate that our Co-STEER agent breaks new\nground in AD^2 research, possesses strong evolvable schedule and implementation\nability, and demonstrates the significant effectiveness of its components. Our\nCo-STEER paves the way for AD^2 advancements.\n", "date": "2024-07-26", "categories": []}
{"id": "2407.17358", "title": "Quantile Learn-Then-Test: Quantile-Based Risk Control for Hyperparameter\n  Optimization", "abstract": "  The increasing adoption of Artificial Intelligence (AI) in engineering\nproblems calls for the development of calibration methods capable of offering\nrobust statistical reliability guarantees. The calibration of black box AI\nmodels is carried out via the optimization of hyperparameters dictating\narchitecture, optimization, and/or inference configuration. Prior work has\nintroduced learn-then-test (LTT), a calibration procedure for hyperparameter\noptimization (HPO) that provides statistical guarantees on average performance\nmeasures. Recognizing the importance of controlling risk-aware objectives in\nengineering contexts, this work introduces a variant of LTT that is designed to\nprovide statistical guarantees on quantiles of a risk measure. We illustrate\nthe practical advantages of this approach by applying the proposed algorithm to\na radio access scheduling problem.\n", "date": "2024-07-24", "categories": []}
{"id": "2403.11446", "title": "LLM Guided Evolution -- The Automation of Models Advancing Models", "abstract": "  In the realm of machine learning, traditional model development and automated\napproaches like AutoML typically rely on layers of abstraction, such as\ntree-based or Cartesian genetic programming. Our study introduces \"Guided\nEvolution\" (GE), a novel framework that diverges from these methods by\nutilizing Large Language Models (LLMs) to directly modify code. GE leverages\nLLMs for a more intelligent, supervised evolutionary process, guiding mutations\nand crossovers. Our unique \"Evolution of Thought\" (EoT) technique further\nenhances GE by enabling LLMs to reflect on and learn from the outcomes of\nprevious mutations. This results in a self-sustaining feedback loop that\naugments decision-making in model evolution. GE maintains genetic diversity,\ncrucial for evolutionary algorithms, by leveraging LLMs' capability to generate\ndiverse responses from expertly crafted prompts and modulate model temperature.\nThis not only accelerates the evolution process but also injects expert like\ncreativity and insight into the process. Our application of GE in evolving the\nExquisiteNetV2 model demonstrates its efficacy: the LLM-driven GE autonomously\nproduced variants with improved accuracy, increasing from 92.52% to 93.34%,\nwithout compromising model compactness. This underscores the potential of LLMs\nto accelerate the traditional model design pipeline, enabling models to\nautonomously evolve and enhance their own designs.\n", "date": "2024-03-18", "categories": []}
{"id": "2407.11286", "title": "CLAMS: A System for Zero-Shot Model Selection for Clustering", "abstract": "  We propose an AutoML system that enables model selection on clustering\nproblems by leveraging optimal transport-based dataset similarity. Our\nobjective is to establish a comprehensive AutoML pipeline for clustering\nproblems and provide recommendations for selecting the most suitable\nalgorithms, thus opening up a new area of AutoML beyond the traditional\nsupervised learning settings. We compare our results against multiple\nclustering baselines and find that it outperforms all of them, hence\ndemonstrating the utility of similarity-based automated model selection for\nsolving clustering applications.\n", "date": "2024-07-15", "categories": []}
{"id": "2407.05788", "title": "Automated Computational Energy Minimization of ML Algorithms using\n  Constrained Bayesian Optimization", "abstract": "  Bayesian optimization (BO) is an efficient framework for optimization of\nblack-box objectives when function evaluations are costly and gradient\ninformation is not easily accessible. BO has been successfully applied to\nautomate the task of hyperparameter optimization (HPO) in machine learning (ML)\nmodels with the primary objective of optimizing predictive performance on\nheld-out data. In recent years, however, with ever-growing model sizes, the\nenergy cost associated with model training has become an important factor for\nML applications. Here we evaluate Constrained Bayesian Optimization (CBO) with\nthe primary objective of minimizing energy consumption and subject to the\nconstraint that the generalization performance is above some threshold. We\nevaluate our approach on regression and classification tasks and demonstrate\nthat CBO achieves lower energy consumption without compromising the predictive\nperformance of ML models.\n", "date": "2024-07-08", "categories": []}
{"id": "2407.00613", "title": "A Linear Programming Enhanced Genetic Algorithm for Hyperparameter\n  Tuning in Machine Learning", "abstract": "  In this paper, we formulate the hyperparameter tuning problem in machine\nlearning as a bilevel program. The bilevel program is solved using a micro\ngenetic algorithm that is enhanced with a linear program. While the genetic\nalgorithm searches over discrete hyperparameters, the linear program\nenhancement allows hyper local search over continuous hyperparameters. The\nmajor contribution in this paper is the formulation of a linear program that\nsupports fast search over continuous hyperparameters, and can be integrated\nwith any hyperparameter search technique. It can also be applied directly on\nany trained machine learning or deep learning model for the purpose of\nfine-tuning. We test the performance of the proposed approach on two datasets,\nMNIST and CIFAR-10. Our results clearly demonstrate that using the linear\nprogram enhancement offers significant promise when incorporated with any\npopulation-based approach for hyperparameter tuning.\n", "date": "2024-06-30", "categories": []}
{"id": "2406.08099", "title": "Confidence Interval Estimation of Predictive Performance in the Context\n  of AutoML", "abstract": "  Any supervised machine learning analysis is required to provide an estimate\nof the out-of-sample predictive performance. However, it is imperative to also\nprovide a quantification of the uncertainty of this performance in the form of\na confidence or credible interval (CI) and not just a point estimate. In an\nAutoML setting, estimating the CI is challenging due to the ``winner's curse\",\ni.e., the bias of estimation due to cross-validating several machine learning\npipelines and selecting the winning one. In this work, we perform a comparative\nevaluation of 9 state-of-the-art methods and variants in CI estimation in an\nAutoML setting on a corpus of real and simulated datasets. The methods are\ncompared in terms of inclusion percentage (does a 95\\% CI include the true\nperformance at least 95\\% of the time), CI tightness (tighter CIs are\npreferable as being more informative), and execution time. The evaluation is\nthe first one that covers most, if not all, such methods and extends previous\nwork to imbalanced and small-sample tasks. In addition, we present a variant,\ncalled BBC-F, of an existing method (the Bootstrap Bias Correction, or BBC)\nthat maintains the statistical properties of the BBC but is more\ncomputationally efficient. The results support that BBC-F and BBC dominate the\nother methods in all metrics measured.\n", "date": "2024-06-12", "categories": []}
{"id": "2406.07404", "title": "Enhancing Tabular Data Optimization with a Flexible Graph-based\n  Reinforced Exploration Strategy", "abstract": "  Tabular data optimization methods aim to automatically find an optimal\nfeature transformation process that generates high-value features and improves\nthe performance of downstream machine learning tasks. Current frameworks for\nautomated feature transformation rely on iterative sequence generation tasks,\noptimizing decision strategies through performance feedback from downstream\ntasks. However, these approaches fail to effectively utilize historical\ndecision-making experiences and overlook potential relationships among\ngenerated features, thus limiting the depth of knowledge extraction. Moreover,\nthe granularity of the decision-making process lacks dynamic backtracking\ncapabilities for individual features, leading to insufficient adaptability when\nencountering inefficient pathways, adversely affecting overall robustness and\nexploration efficiency. To address the limitations observed in current\nautomatic feature engineering frameworks, we introduce a novel method that\nutilizes a feature-state transformation graph to effectively preserve the\nentire feature transformation journey, where each node represents a specific\ntransformation state. During exploration, three cascading agents iteratively\nselect nodes and idea mathematical operations to generate new transformation\nstates. This strategy leverages the inherent properties of the graph structure,\nallowing for the preservation and reuse of valuable transformations. It also\nenables backtracking capabilities through graph pruning techniques, which can\nrectify inefficient transformation paths. To validate the efficacy and\nflexibility of our approach, we conducted comprehensive experiments and\ndetailed case studies, demonstrating superior performance in diverse scenarios.\n", "date": "2024-06-11", "categories": []}
{"id": "2206.07438", "title": "Multi-Objective Hyperparameter Optimization in Machine Learning -- An\n  Overview", "abstract": "  Hyperparameter optimization constitutes a large part of typical modern\nmachine learning workflows. This arises from the fact that machine learning\nmethods and corresponding preprocessing steps often only yield optimal\nperformance when hyperparameters are properly tuned. But in many applications,\nwe are not only interested in optimizing ML pipelines solely for predictive\naccuracy; additional metrics or constraints must be considered when determining\nan optimal configuration, resulting in a multi-objective optimization problem.\nThis is often neglected in practice, due to a lack of knowledge and readily\navailable software implementations for multi-objective hyperparameter\noptimization. In this work, we introduce the reader to the basics of\nmulti-objective hyperparameter optimization and motivate its usefulness in\napplied ML. Furthermore, we provide an extensive survey of existing\noptimization strategies, both from the domain of evolutionary algorithms and\nBayesian optimization. We illustrate the utility of MOO in several specific ML\napplications, considering objectives such as operating conditions, prediction\ntime, sparseness, fairness, interpretability and robustness.\n", "date": "2022-06-15", "categories": []}
{"id": "2406.04153", "title": "Learned Feature Importance Scores for Automated Feature Engineering", "abstract": "  Feature engineering has demonstrated substantial utility for many machine\nlearning workflows, such as in the small data regime or when distribution\nshifts are severe. Thus automating this capability can relieve much manual\neffort and improve model performance. Towards this, we propose AutoMAN, or\nAutomated Mask-based Feature Engineering, an automated feature engineering\nframework that achieves high accuracy, low latency, and can be extended to\nheterogeneous and time-varying data. AutoMAN is based on effectively exploring\nthe candidate transforms space, without explicitly manifesting transformed\nfeatures. This is achieved by learning feature importance masks, which can be\nextended to support other modalities such as time series. AutoMAN learns\nfeature transform importance end-to-end, incorporating a dataset's task target\ndirectly into feature engineering, resulting in state-of-the-art performance\nwith significantly lower latency compared to alternatives.\n", "date": "2024-06-06", "categories": []}
{"id": "2406.03348", "title": "Position: A Call to Action for a Human-Centered AutoML Paradigm", "abstract": "  Automated machine learning (AutoML) was formed around the fundamental\nobjectives of automatically and efficiently configuring machine learning (ML)\nworkflows, aiding the research of new ML algorithms, and contributing to the\ndemocratization of ML by making it accessible to a broader audience. Over the\npast decade, commendable achievements in AutoML have primarily focused on\noptimizing predictive performance. This focused progress, while substantial,\nraises questions about how well AutoML has met its broader, original goals. In\nthis position paper, we argue that a key to unlocking AutoML's full potential\nlies in addressing the currently underexplored aspect of user interaction with\nAutoML systems, including their diverse roles, expectations, and expertise. We\nenvision a more human-centered approach in future AutoML research, promoting\nthe collaborative design of ML systems that tightly integrates the\ncomplementary strengths of human expertise and AutoML methodologies.\n", "date": "2024-06-05", "categories": []}
{"id": "2406.01789", "title": "AI-based Classification of Customer Support Tickets: State of the Art\n  and Implementation with AutoML", "abstract": "  Automation of support ticket classification is crucial to improve customer\nsupport performance and shortening resolution time for customer inquiries. This\nresearch aims to test the applicability of automated machine learning (AutoML)\nas a technology to train a machine learning model (ML model) that can classify\nsupport tickets. The model evaluation conducted in this research shows that\nAutoML can be used to train ML models with good classification performance.\nMoreover, this paper fills a research gap by providing new insights into\ndeveloping AI solutions without a dedicated professional by utilizing AutoML,\nwhich makes this technology more accessible for companies without specialized\nAI departments and staff.\n", "date": "2024-06-03", "categories": []}
{"id": "2402.17453", "title": "DS-Agent: Automated Data Science by Empowering Large Language Models\n  with Case-Based Reasoning", "abstract": "  In this work, we investigate the potential of large language models (LLMs)\nbased agents to automate data science tasks, with the goal of comprehending\ntask requirements, then building and training the best-fit machine learning\nmodels. Despite their widespread success, existing LLM agents are hindered by\ngenerating unreasonable experiment plans within this scenario. To this end, we\npresent DS-Agent, a novel automatic framework that harnesses LLM agent and\ncase-based reasoning (CBR). In the development stage, DS-Agent follows the CBR\nframework to structure an automatic iteration pipeline, which can flexibly\ncapitalize on the expert knowledge from Kaggle, and facilitate consistent\nperformance improvement through the feedback mechanism. Moreover, DS-Agent\nimplements a low-resource deployment stage with a simplified CBR paradigm to\nadapt past successful solutions from the development stage for direct code\ngeneration, significantly reducing the demand on foundational capabilities of\nLLMs. Empirically, DS-Agent with GPT-4 achieves 100\\% success rate in the\ndevelopment stage, while attaining 36\\% improvement on average one pass rate\nacross alternative LLMs in the deployment stage. In both stages, DS-Agent\nachieves the best rank in performance, costing \\$1.60 and \\$0.13 per run with\nGPT-4, respectively. Our data and code are open-sourced at\nhttps://github.com/guosyjlu/DS-Agent.\n", "date": "2024-02-27", "categories": []}
{"id": "2311.14014", "title": "On the Hyperparameter Loss Landscapes of Machine Learning Models: An\n  Exploratory Study", "abstract": "  Previous efforts on hyperparameter optimization (HPO) of machine learning\n(ML) models predominately focus on algorithmic advances, yet little is known\nabout the topography of the underlying hyperparameter (HP) loss landscape,\nwhich plays a fundamental role in governing the search process of HPO. While\nseveral works have conducted fitness landscape analysis (FLA) on various ML\nsystems, they are limited to properties of isolated landscape without\ninterrogating the potential structural similarities among them. The exploration\nof such similarities can provide a novel perspective for understanding the\nmechanism behind modern HPO methods, but has been missing, possibly due to the\nexpensive cost of large-scale landscape construction, and the lack of effective\nanalysis methods. In this paper, we mapped 1,500 HP loss landscapes of 6\nrepresentative ML models on 63 datasets across different fidelity levels, with\n11M+ configurations. By conducting exploratory analysis on these landscapes\nwith fine-grained visualizations and dedicated FLA metrics, we observed a\nsimilar landscape topography across a wide range of models, datasets, and\nfidelities, and shed light on several central topics in HPO.\n", "date": "2023-11-23", "categories": []}
{"id": "2311.13184", "title": "Large Language Model-Enhanced Algorithm Selection: Towards Comprehensive\n  Algorithm Representation", "abstract": "  Algorithm selection, a critical process of automated machine learning, aims\nto identify the most suitable algorithm for solving a specific problem prior to\nexecution. Mainstream algorithm selection techniques heavily rely on problem\nfeatures, while the role of algorithm features remains largely unexplored. Due\nto the intrinsic complexity of algorithms, effective methods for universally\nextracting algorithm information are lacking. This paper takes a significant\nstep towards bridging this gap by introducing Large Language Models (LLMs) into\nalgorithm selection for the first time. By comprehending the code text, LLM not\nonly captures the structural and semantic aspects of the algorithm, but also\ndemonstrates contextual awareness and library function understanding. The\nhigh-dimensional algorithm representation extracted by LLM, after undergoing a\nfeature selection module, is combined with the problem representation and\npassed to the similarity calculation module. The selected algorithm is\ndetermined by the matching degree between a given problem and different\nalgorithms. Extensive experiments validate the performance superiority of the\nproposed model and the efficacy of each key module. Furthermore, we present a\ntheoretical upper bound on model complexity, showcasing the influence of\nalgorithm representation and feature selection modules. This provides valuable\ntheoretical guidance for the practical implementation of our method.\n", "date": "2023-11-22", "categories": []}
{"id": "2304.02396", "title": "AutoRL Hyperparameter Landscapes", "abstract": "  Although Reinforcement Learning (RL) has shown to be capable of producing\nimpressive results, its use is limited by the impact of its hyperparameters on\nperformance. This often makes it difficult to achieve good results in practice.\nAutomated RL (AutoRL) addresses this difficulty, yet little is known about the\ndynamics of the hyperparameter landscapes that hyperparameter optimization\n(HPO) methods traverse in search of optimal configurations. In view of existing\nAutoRL approaches dynamically adjusting hyperparameter configurations, we\npropose an approach to build and analyze these hyperparameter landscapes not\njust for one point in time but at multiple points in time throughout training.\nAddressing an important open question on the legitimacy of such dynamic AutoRL\napproaches, we provide thorough empirical evidence that the hyperparameter\nlandscapes strongly vary over time across representative algorithms from RL\nliterature (DQN, PPO, and SAC) in different kinds of environments (Cartpole,\nBipedal Walker, and Hopper) This supports the theory that hyperparameters\nshould be dynamically adjusted during training and shows the potential for more\ninsights on AutoRL problems that can be gained through landscape analyses. Our\ncode can be found at https://github.com/automl/AutoRL-Landscape\n", "date": "2023-04-05", "categories": []}
{"id": "2405.08842", "title": "Automated Deep Learning for Load Forecasting", "abstract": "  Accurate forecasting of electricity consumption is essential to ensure the\nperformance and stability of the grid, especially as the use of renewable\nenergy increases. Forecasting electricity is challenging because it depends on\nmany external factors, such as weather and calendar variables. While\nregression-based models are currently effective, the emergence of new\nexplanatory variables and the need to refine the temporality of the signals to\nbe forecasted is encouraging the exploration of novel methodologies, in\nparticular deep learning models. However, Deep Neural Networks (DNNs) struggle\nwith this task due to the lack of data points and the different types of\nexplanatory variables (e.g. integer, float, or categorical). In this paper, we\nexplain why and how we used Automated Deep Learning (AutoDL) to find performing\nDNNs for load forecasting. We ended up creating an AutoDL framework called\nEnergyDragon by extending the DRAGON package and applying it to load\nforecasting. EnergyDragon automatically selects the features embedded in the\nDNN training in an innovative way and optimizes the architecture and the\nhyperparameters of the networks. We demonstrate on the French load signal that\nEnergyDragon can find original DNNs that outperform state-of-the-art load\nforecasting methods as well as other AutoDL approaches.\n", "date": "2024-05-14", "categories": []}
{"id": "2311.17279", "title": "LiveTune: Dynamic Parameter Tuning for Feedback-Driven Optimization", "abstract": "  Feedback-driven optimization, such as traditional machine learning training,\nis a static process that lacks real-time adaptability of hyperparameters.\nTuning solutions for optimization require trial and error paired with\ncheckpointing and schedulers, in many cases feedback from the algorithm is\noverlooked. Adjusting hyperparameters during optimization usually requires the\nprogram to be restarted, wasting utilization and time, while placing\nunnecessary strain on memory and processors. We present LiveTune, a novel\nframework allowing real-time parameter adjustment of optimization loops through\nLiveVariables. Live Variables allow for continuous feedback-driven optimization\nby storing parameters on designated ports on the system, allowing them to be\ndynamically adjusted. Extensive evaluations of our framework on standard\nmachine learning training pipelines show saving up to 60 seconds and 5.4\nKilojoules of energy per hyperparameter change. We also show the feasibility\nand value of LiveTune in a reinforcement learning application where the users\nchange the dynamics of the reward structure while the agent is learning showing\n5x improvement over the baseline. Finally, we outline a fully automated\nworkflow to provide end-to-end, unsupervised feedback-driven optimization.\n", "date": "2023-11-28", "categories": []}
{"id": "2405.07662", "title": "Squeezing Lemons with Hammers: An Evaluation of AutoML and Tabular Deep\n  Learning for Data-Scarce Classification Applications", "abstract": "  Many industry verticals are confronted with small-sized tabular data. In this\nlow-data regime, it is currently unclear whether the best performance can be\nexpected from simple baselines, or more complex machine learning approaches\nthat leverage meta-learning and ensembling. On 44 tabular classification\ndatasets with sample sizes $\\leq$ 500, we find that L2-regularized logistic\nregression performs similar to state-of-the-art automated machine learning\n(AutoML) frameworks (AutoPrognosis, AutoGluon) and off-the-shelf deep neural\nnetworks (TabPFN, HyperFast) on the majority of the benchmark datasets. We\ntherefore recommend to consider logistic regression as the first choice for\ndata-scarce applications with tabular data and provide practitioners with best\npractices for further method selection.\n", "date": "2024-05-13", "categories": []}
{"id": "2404.16233", "title": "AutoGluon-Multimodal (AutoMM): Supercharging Multimodal AutoML with\n  Foundation Models", "abstract": "  AutoGluon-Multimodal (AutoMM) is introduced as an open-source AutoML library\ndesigned specifically for multimodal learning. Distinguished by its exceptional\nease of use, AutoMM enables fine-tuning of foundation models with just three\nlines of code. Supporting various modalities including image, text, and tabular\ndata, both independently and in combination, the library offers a comprehensive\nsuite of functionalities spanning classification, regression, object detection,\nsemantic matching, and image segmentation. Experiments across diverse datasets\nand tasks showcases AutoMM's superior performance in basic classification and\nregression tasks compared to existing AutoML tools, while also demonstrating\ncompetitive results in advanced tasks, aligning with specialized toolboxes\ndesigned for such purposes.\n", "date": "2024-04-24", "categories": []}
{"id": "2405.00182", "title": "M-DEW: Extending Dynamic Ensemble Weighting to Handle Missing Values", "abstract": "  Missing value imputation is a crucial preprocessing step for many machine\nlearning problems. However, it is often considered as a separate subtask from\ndownstream applications such as classification, regression, or clustering, and\nthus is not optimized together with them. We hypothesize that treating the\nimputation model and downstream task model together and optimizing over full\npipelines will yield better results than treating them separately. Our work\ndescribes a novel AutoML technique for making downstream predictions with\nmissing data that automatically handles preprocessing, model weighting, and\nselection during inference time, with minimal compute overhead. Specifically we\ndevelop M-DEW, a Dynamic missingness-aware Ensemble Weighting (DEW) approach,\nthat constructs a set of two-stage imputation-prediction pipelines, trains each\ncomponent separately, and dynamically calculates a set of pipeline weights for\neach sample during inference time. We thus extend previous work on dynamic\nensemble weighting to handle missing data at the level of full\nimputation-prediction pipelines, improving performance and calibration on\ndownstream machine learning tasks over standard model averaging techniques.\nM-DEW is shown to outperform the state-of-the-art in that it produces\nstatistically significant reductions in model perplexity in 17 out of 18\nexperiments, while improving average precision in 13 out of 18 experiments.\n", "date": "2024-04-30", "categories": []}
{"id": "2312.08528", "title": "auto-sktime: Automated Time Series Forecasting", "abstract": "  In today's data-driven landscape, time series forecasting is pivotal in\ndecision-making across various sectors. Yet, the proliferation of more diverse\ntime series data, coupled with the expanding landscape of available forecasting\nmethods, poses significant challenges for forecasters. To meet the growing\ndemand for efficient forecasting, we introduce auto-sktime, a novel framework\nfor automated time series forecasting. The proposed framework uses the power of\nautomated machine learning (AutoML) techniques to automate the creation of the\nentire forecasting pipeline. The framework employs Bayesian optimization, to\nautomatically construct pipelines from statistical, machine learning (ML) and\ndeep neural network (DNN) models. Furthermore, we propose three essential\nimprovements to adapt AutoML to time series data. First, pipeline templates to\naccount for the different supported forecasting models. Second, a novel\nwarm-starting technique to start the optimization from prior optimization runs.\nThird, we adapt multi-fidelity optimizations to make them applicable to a\nsearch space containing statistical, ML and DNN models. Experimental results on\n64 diverse real-world time series datasets demonstrate the effectiveness and\nefficiency of the framework, outperforming traditional methods while requiring\nminimal human involvement.\n", "date": "2023-12-13", "categories": []}
{"id": "2404.03419", "title": "Integrating Hyperparameter Search into Model-Free AutoML with\n  Context-Free Grammars", "abstract": "  Automated Machine Learning (AutoML) has become increasingly popular in recent\nyears due to its ability to reduce the amount of time and expertise required to\ndesign and develop machine learning systems. This is very important for the\npractice of machine learning, as it allows building strong baselines quickly,\nimproving the efficiency of the data scientists, and reducing the time to\nproduction. However, despite the advantages of AutoML, it faces several\nchallenges, such as defining the solutions space and exploring it efficiently.\nRecently, some approaches have been shown to be able to do it using tree-based\nsearch algorithms and context-free grammars. In particular, GramML presents a\nmodel-free reinforcement learning approach that leverages pipeline\nconfiguration grammars and operates using Monte Carlo tree search. However, one\nof the limitations of GramML is that it uses default hyperparameters, limiting\nthe search problem to finding optimal pipeline structures for the available\ndata preprocessors and models. In this work, we propose an extension to GramML\nthat supports larger search spaces including hyperparameter search. We\nevaluated the approach using an OpenML benchmark and found significant\nimprovements compared to other state-of-the-art techniques.\n", "date": "2024-04-04", "categories": []}
{"id": "2302.10827", "title": "AutoML in The Wild: Obstacles, Workarounds, and Expectations", "abstract": "  Automated machine learning (AutoML) is envisioned to make ML techniques\naccessible to ordinary users. Recent work has investigated the role of humans\nin enhancing AutoML functionality throughout a standard ML workflow. However,\nit is also critical to understand how users adopt existing AutoML solutions in\ncomplex, real-world settings from a holistic perspective. To fill this gap,\nthis study conducted semi-structured interviews of AutoML users (N=19) focusing\non understanding (1) the limitations of AutoML encountered by users in their\nreal-world practices, (2) the strategies users adopt to cope with such\nlimitations, and (3) how the limitations and workarounds impact their use of\nAutoML. Our findings reveal that users actively exercise user agency to\novercome three major challenges arising from customizability, transparency, and\nprivacy. Furthermore, users make cautious decisions about whether and how to\napply AutoML on a case-by-case basis. Finally, we derive design implications\nfor developing future AutoML solutions.\n", "date": "2023-02-21", "categories": []}
{"id": "2403.11395", "title": "Automated data processing and feature engineering for deep learning and\n  big data applications: a survey", "abstract": "  Modern approach to artificial intelligence (AI) aims to design algorithms\nthat learn directly from data. This approach has achieved impressive results\nand has contributed significantly to the progress of AI, particularly in the\nsphere of supervised deep learning. It has also simplified the design of\nmachine learning systems as the learning process is highly automated. However,\nnot all data processing tasks in conventional deep learning pipelines have been\nautomated. In most cases data has to be manually collected, preprocessed and\nfurther extended through data augmentation before they can be effective for\ntraining. Recently, special techniques for automating these tasks have emerged.\nThe automation of data processing tasks is driven by the need to utilize large\nvolumes of complex, heterogeneous data for machine learning and big data\napplications. Today, end-to-end automated data processing systems based on\nautomated machine learning (AutoML) techniques are capable of taking raw data\nand transforming them into useful features for Big Data tasks by automating all\nintermediate processing stages. In this work, we present a thorough review of\napproaches for automating data processing tasks in deep learning pipelines,\nincluding automated data preprocessing--e.g., data cleaning, labeling, missing\ndata imputation, and categorical data encoding--as well as data augmentation\n(including synthetic data generation using generative AI methods) and feature\nengineering--specifically, automated feature extraction, feature construction\nand feature selection. In addition to automating specific data processing\ntasks, we discuss the use of AutoML methods and tools to simultaneously\noptimize all stages of the machine learning pipeline.\n", "date": "2024-03-18", "categories": []}
{"id": "2403.12664", "title": "Deciphering AutoML Ensembles: cattleia's Assistance in Decision-Making", "abstract": "  In many applications, model ensembling proves to be better than a single\npredictive model. Hence, it is the most common post-processing technique in\nAutomated Machine Learning (AutoML). The most popular frameworks use ensembles\nat the expense of reducing the interpretability of the final models. In our\nwork, we propose cattleia - an application that deciphers the ensembles for\nregression, multiclass, and binary classification tasks. This tool works with\nmodels built by three AutoML packages: auto-sklearn, AutoGluon, and FLAML. The\ngiven ensemble is analyzed from different perspectives. We conduct a predictive\nperformance investigation through evaluation metrics of the ensemble and its\ncomponent models. We extend the validation perspective by introducing new\nmeasures to assess the diversity and complementarity of the model predictions.\nMoreover, we apply explainable artificial intelligence (XAI) techniques to\nexamine the importance of variables. Summarizing obtained insights, we can\ninvestigate and adjust the weights with a modification tool to tune the\nensemble in the desired way. The application provides the aforementioned\naspects through dedicated interactive visualizations, making it accessible to a\ndiverse audience. We believe the cattleia can support users in decision-making\nand deepen the comprehension of AutoML frameworks.\n", "date": "2024-03-19", "categories": []}
{"id": "2210.12090", "title": "AutoPrognosis 2.0: Democratizing Diagnostic and Prognostic Modeling in\n  Healthcare with Automated Machine Learning", "abstract": "  Diagnostic and prognostic models are increasingly important in medicine and\ninform many clinical decisions. Recently, machine learning approaches have\nshown improvement over conventional modeling techniques by better capturing\ncomplex interactions between patient covariates in a data-driven manner.\nHowever, the use of machine learning introduces a number of technical and\npractical challenges that have thus far restricted widespread adoption of such\ntechniques in clinical settings. To address these challenges and empower\nhealthcare professionals, we present a machine learning framework,\nAutoPrognosis 2.0, to develop diagnostic and prognostic models. AutoPrognosis\nleverages state-of-the-art advances in automated machine learning to develop\noptimized machine learning pipelines, incorporates model explainability tools,\nand enables deployment of clinical demonstrators, without requiring significant\ntechnical expertise. Our framework eliminates the major technical obstacles to\npredictive modeling with machine learning that currently impede clinical\nadoption. To demonstrate AutoPrognosis 2.0, we provide an illustrative\napplication where we construct a prognostic risk score for diabetes using the\nUK Biobank, a prospective study of 502,467 individuals. The models produced by\nour automated framework achieve greater discrimination for diabetes than expert\nclinical risk scores. Our risk score has been implemented as a web-based\ndecision support tool and can be publicly accessed by patients and clinicians\nworldwide. In addition, AutoPrognosis 2.0 is provided as an open-source python\npackage. By open-sourcing our framework as a tool for the community, clinicians\nand other medical practitioners will be able to readily develop new risk\nscores, personalized diagnostics, and prognostics using modern machine learning\ntechniques.\n", "date": "2022-10-21", "categories": []}
{"id": "2402.18198", "title": "Automated Machine Learning for Multi-Label Classification", "abstract": "  Automated machine learning (AutoML) aims to select and configure machine\nlearning algorithms and combine them into machine learning pipelines tailored\nto a dataset at hand. For supervised learning tasks, most notably binary and\nmultinomial classification, aka single-label classification (SLC), such AutoML\napproaches have shown promising results. However, the task of multi-label\nclassification (MLC), where data points are associated with a set of class\nlabels instead of a single class label, has received much less attention so\nfar. In the context of multi-label classification, the data-specific selection\nand configuration of multi-label classifiers are challenging even for experts\nin the field, as it is a high-dimensional optimization problem with multi-level\nhierarchical dependencies. While for SLC, the space of machine learning\npipelines is already huge, the size of the MLC search space outnumbers the one\nof SLC by several orders.\n  In the first part of this thesis, we devise a novel AutoML approach for\nsingle-label classification tasks optimizing pipelines of machine learning\nalgorithms, consisting of two algorithms at most. This approach is then\nextended first to optimize pipelines of unlimited length and eventually\nconfigure the complex hierarchical structures of multi-label classification\nmethods. Furthermore, we investigate how well AutoML approaches that form the\nstate of the art for single-label classification tasks scale with the increased\nproblem complexity of AutoML for multi-label classification.\n  In the second part, we explore how methods for SLC and MLC could be\nconfigured more flexibly to achieve better generalization performance and how\nto increase the efficiency of execution-based AutoML systems.\n", "date": "2024-02-28", "categories": []}
{"id": "2402.18505", "title": "Evolving machine learning workflows through interactive AutoML", "abstract": "  Automatic workflow composition (AWC) is a relevant problem in automated\nmachine learning (AutoML) that allows finding suitable sequences of\npreprocessing and prediction models together with their optimal\nhyperparameters. This problem can be solved using evolutionary algorithms and,\nin particular, grammar-guided genetic programming (G3P). Current G3P approaches\nto AWC define a fixed grammar that formally specifies how workflow elements can\nbe combined and which algorithms can be included. In this paper we present\n\\ourmethod, an interactive G3P algorithm that allows users to dynamically\nmodify the grammar to prune the search space and focus on their regions of\ninterest. Our proposal is the first to combine the advantages of a G3P method\nwith ideas from interactive optimisation and human-guided machine learning, an\narea little explored in the context of AutoML. To evaluate our approach, we\npresent an experimental study in which 20 participants interact with \\ourmethod\nto evolve workflows according to their preferences. Our results confirm that\nthe collaboration between \\ourmethod and humans allows us to find\nhigh-performance workflows in terms of accuracy that require less tuning time\nthan those found without human intervention.\n", "date": "2024-02-28", "categories": []}
{"id": "2306.08107", "title": "AutoML in the Age of Large Language Models: Current Challenges, Future\n  Opportunities and Risks", "abstract": "  The fields of both Natural Language Processing (NLP) and Automated Machine\nLearning (AutoML) have achieved remarkable results over the past years. In NLP,\nespecially Large Language Models (LLMs) have experienced a rapid series of\nbreakthroughs very recently. We envision that the two fields can radically push\nthe boundaries of each other through tight integration. To showcase this\nvision, we explore the potential of a symbiotic relationship between AutoML and\nLLMs, shedding light on how they can benefit each other. In particular, we\ninvestigate both the opportunities to enhance AutoML approaches with LLMs from\ndifferent perspectives and the challenges of leveraging AutoML to further\nimprove LLMs. To this end, we survey existing work, and we critically assess\nrisks. We strongly believe that the integration of the two fields has the\npotential to disrupt both fields, NLP and AutoML. By highlighting conceivable\nsynergies, but also risks, we aim to foster further exploration at the\nintersection of AutoML and LLMs.\n", "date": "2023-06-13", "categories": []}
{"id": "2303.08485", "title": "Can Fairness be Automated? Guidelines and Opportunities for\n  Fairness-aware AutoML", "abstract": "  The field of automated machine learning (AutoML) introduces techniques that\nautomate parts of the development of machine learning (ML) systems,\naccelerating the process and reducing barriers for novices. However, decisions\nderived from ML models can reproduce, amplify, or even introduce unfairness in\nour societies, causing harm to (groups of) individuals. In response,\nresearchers have started to propose AutoML systems that jointly optimize\nfairness and predictive performance to mitigate fairness-related harm. However,\nfairness is a complex and inherently interdisciplinary subject, and solely\nposing it as an optimization problem can have adverse side effects. With this\nwork, we aim to raise awareness among developers of AutoML systems about such\nlimitations of fairness-aware AutoML, while also calling attention to the\npotential of AutoML as a tool for fairness research. We present a comprehensive\noverview of different ways in which fairness-related harm can arise and the\nensuing implications for the design of fairness-aware AutoML. We conclude that\nwhile fairness cannot be automated, fairness-aware AutoML can play an important\nrole in the toolbox of ML practitioners. We highlight several open technical\nchallenges for future work in this direction. Additionally, we advocate for the\ncreation of more user-centered assistive systems designed to tackle challenges\nencountered in fairness work\n", "date": "2023-03-15", "categories": []}
{"id": "2402.12408", "title": "ModelGPT: Unleashing LLM's Capabilities for Tailored Model Generation", "abstract": "  The rapid advancement of Large Language Models (LLMs) has revolutionized\nvarious sectors by automating routine tasks, marking a step toward the\nrealization of Artificial General Intelligence (AGI). However, they still\nstruggle to accommodate the diverse and specific needs of users and simplify\nthe utilization of AI models for the average user. In response, we propose\nModelGPT, a novel framework designed to determine and generate AI models\nspecifically tailored to the data or task descriptions provided by the user,\nleveraging the capabilities of LLMs. Given user requirements, ModelGPT is able\nto provide tailored models at most 270x faster than the previous paradigms\n(e.g. all-parameter or LoRA finetuning). Comprehensive experiments on NLP, CV,\nand Tabular datasets attest to the effectiveness of our framework in making AI\nmodels more accessible and user-friendly. Our code is available at\nhttps://github.com/IshiKura-a/ModelGPT.\n", "date": "2024-02-18", "categories": []}
{"id": "2402.12916", "title": "Data Pipeline Training: Integrating AutoML to Optimize the Data Flow of\n  Machine Learning Models", "abstract": "  Data Pipeline plays an indispensable role in tasks such as modeling machine\nlearning and developing data products. With the increasing diversification and\ncomplexity of Data sources, as well as the rapid growth of data volumes,\nbuilding an efficient Data Pipeline has become crucial for improving work\nefficiency and solving complex problems. This paper focuses on exploring how to\noptimize data flow through automated machine learning methods by integrating\nAutoML with Data Pipeline. We will discuss how to leverage AutoML technology to\nenhance the intelligence of Data Pipeline, thereby achieving better results in\nmachine learning tasks. By delving into the automation and optimization of Data\nflows, we uncover key strategies for constructing efficient data pipelines that\ncan adapt to the ever-changing data landscape. This not only accelerates the\nmodeling process but also provides innovative solutions to complex problems,\nenabling more significant outcomes in increasingly intricate data domains.\nKeywords- Data Pipeline Training;AutoML; Data environment; Machine learning\n", "date": "2024-02-20", "categories": []}
{"id": "2304.14979", "title": "MLCopilot: Unleashing the Power of Large Language Models in Solving\n  Machine Learning Tasks", "abstract": "  The field of machine learning (ML) has gained widespread adoption, leading to\nsignificant demand for adapting ML to specific scenarios, which is yet\nexpensive and non-trivial. The predominant approaches towards the automation of\nsolving ML tasks (e.g., AutoML) are often time-consuming and hard to understand\nfor human developers. In contrast, though human engineers have the incredible\nability to understand tasks and reason about solutions, their experience and\nknowledge are often sparse and difficult to utilize by quantitative approaches.\nIn this paper, we aim to bridge the gap between machine intelligence and human\nknowledge by introducing a novel framework, which leverages the\nstate-of-the-art large language models to develop ML solutions for novel tasks.\nWe showcase the possibility of extending the capability of LLMs to comprehend\nstructured inputs and perform thorough reasoning for solving novel ML tasks.\nAnd we find that, after some dedicated design, the LLM can (i) observe from the\nexisting experiences of ML tasks and (ii) reason effectively to deliver\npromising results for new tasks. The solution generated can be used directly to\nachieve high levels of competitiveness. Examples and code available at\nhttps://github.com/microsoft/CoML.\n", "date": "2023-04-28", "categories": []}
{"id": "2402.11594", "title": "Simplifying Hyperparameter Tuning in Online Machine Learning -- The\n  spotRiverGUI", "abstract": "  Batch Machine Learning (BML) reaches its limits when dealing with very large\namounts of streaming data. This is especially true for available memory,\nhandling drift in data streams, and processing new, unknown data. Online\nMachine Learning (OML) is an alternative to BML that overcomes the limitations\nof BML. OML is able to process data in a sequential manner, which is especially\nuseful for data streams. The `river` package is a Python OML-library, which\nprovides a variety of online learning algorithms for classification,\nregression, clustering, anomaly detection, and more. The `spotRiver` package\nprovides a framework for hyperparameter tuning of OML models. The\n`spotRiverGUI` is a graphical user interface for the `spotRiver` package. The\n`spotRiverGUI` releases the user from the burden of manually searching for the\noptimal hyperparameter setting. After the data is provided, users can compare\ndifferent OML algorithms from the powerful `river` package in a convenient way\nand tune the selected algorithms very efficiently.\n", "date": "2024-02-18", "categories": []}
{"id": "2402.05821", "title": "Guided Evolution with Binary Discriminators for ML Program Search", "abstract": "  How to automatically design better machine learning programs is an open\nproblem within AutoML. While evolution has been a popular tool to search for\nbetter ML programs, using learning itself to guide the search has been less\nsuccessful and less understood on harder problems but has the promise to\ndramatically increase the speed and final performance of the optimization\nprocess. We propose guiding evolution with a binary discriminator, trained\nonline to distinguish which program is better given a pair of programs. The\ndiscriminator selects better programs without having to perform a costly\nevaluation and thus speed up the convergence of evolution. Our method can\nencode a wide variety of ML components including symbolic optimizers, neural\narchitectures, RL loss functions, and symbolic regression equations with the\nsame directed acyclic graph representation. By combining this representation\nwith modern GNNs and an adaptive mutation strategy, we demonstrate our method\ncan speed up evolution across a set of diverse problems including a 3.7x\nspeedup on the symbolic search for ML optimizers and a 4x speedup for RL loss\nfunctions.\n", "date": "2024-02-08", "categories": []}
{"id": "2402.04453", "title": "The Potential of AutoML for Recommender Systems", "abstract": "  Automated Machine Learning (AutoML) has greatly advanced applications of\nMachine Learning (ML) including model compression, machine translation, and\ncomputer vision. Recommender Systems (RecSys) can be seen as an application of\nML. Yet, AutoML has found little attention in the RecSys community; nor has\nRecSys found notable attention in the AutoML community. Only few and relatively\nsimple Automated Recommender Systems (AutoRecSys) libraries exist that adopt\nAutoML techniques. However, these libraries are based on student projects and\ndo not offer the features and thorough development of AutoML libraries. We set\nout to determine how AutoML libraries perform in the scenario of an\ninexperienced user who wants to implement a recommender system. We compared the\npredictive performance of 60 AutoML, AutoRecSys, ML, and RecSys algorithms from\n15 libraries, including a mean predictor baseline, on 14 explicit feedback\nRecSys datasets. To simulate the perspective of an inexperienced user, the\nalgorithms were evaluated with default hyperparameters. We found that AutoML\nand AutoRecSys libraries performed best. AutoML libraries performed best for\nsix of the 14 datasets (43%), but it was not always the same AutoML library\nperforming best. The single-best library was the AutoRecSys library\nAuto-Surprise, which performed best on five datasets (36%). On three datasets\n(21%), AutoML libraries performed poorly, and RecSys libraries with default\nparameters performed best. Although, while obtaining 50% of all placements in\nthe top five per dataset, RecSys algorithms fall behind AutoML on average. ML\nalgorithms generally performed the worst.\n", "date": "2024-02-06", "categories": []}
{"id": "2402.02124", "title": "Grammar-based evolutionary approach for automated workflow composition\n  with domain-specific operators and ensemble diversity", "abstract": "  The process of extracting valuable and novel insights from raw data involves\na series of complex steps. In the realm of Automated Machine Learning (AutoML),\na significant research focus is on automating aspects of this process,\nspecifically tasks like selecting algorithms and optimising their\nhyper-parameters. A particularly challenging task in AutoML is automatic\nworkflow composition (AWC). AWC aims to identify the most effective sequence of\ndata preprocessing and ML algorithms, coupled with their best hyper-parameters,\nfor a specific dataset. However, existing AWC methods are limited in how many\nand in what ways they can combine algorithms within a workflow.\n  Addressing this gap, this paper introduces EvoFlow, a grammar-based\nevolutionary approach for AWC. EvoFlow enhances the flexibility in designing\nworkflow structures, empowering practitioners to select algorithms that best\nfit their specific requirements. EvoFlow stands out by integrating two\ninnovative features. First, it employs a suite of genetic operators, designed\nspecifically for AWC, to optimise both the structure of workflows and their\nhyper-parameters. Second, it implements a novel updating mechanism that\nenriches the variety of predictions made by different workflows. Promoting this\ndiversity helps prevent the algorithm from overfitting. With this aim, EvoFlow\nbuilds an ensemble whose workflows differ in their misclassified instances.\n  To evaluate EvoFlow's effectiveness, we carried out empirical validation\nusing a set of classification benchmarks. We begin with an ablation study to\ndemonstrate the enhanced performance attributable to EvoFlow's unique\ncomponents. Then, we compare EvoFlow with other AWC approaches, encompassing\nboth evolutionary and non-evolutionary techniques. Our findings show that\nEvoFlow's specialised genetic operators and updating mechanism substantially\noutperform current leading methods[..]\n", "date": "2024-02-03", "categories": []}
{"id": "2402.01379", "title": "Regularized boosting with an increasing coefficient magnitude stop\n  criterion as meta-learner in hyperparameter optimization stacking ensemble", "abstract": "  In Hyperparameter Optimization (HPO), only the hyperparameter configuration\nwith the best performance is chosen after performing several trials, then,\ndiscarding the effort of training all the models with every hyperparameter\nconfiguration trial and performing an ensemble of all them. This ensemble\nconsists of simply averaging the model predictions or weighting the models by a\ncertain probability. Recently, other more sophisticated ensemble strategies,\nsuch as the Caruana method or the stacking strategy has been proposed. On the\none hand, the Caruana method performs well in HPO ensemble, since it is not\naffected by the effects of multicollinearity, which is prevalent in HPO. It\njust computes the average over a subset of predictions with replacement. But it\ndoes not benefit from the generalization power of a learning process. On the\nother hand, stacking methods include a learning procedure since a meta-learner\nis required to perform the ensemble. Yet, one hardly finds advice about which\nmeta-learner is adequate. Besides, some meta-learners may suffer from the\neffects of multicollinearity or need to be tuned to reduce them. This paper\nexplores meta-learners for stacking ensemble in HPO, free of hyperparameter\ntuning, able to reduce the effects of multicollinearity and considering the\nensemble learning process generalization power. At this respect, the boosting\nstrategy seems promising as a stacking meta-learner. In fact, it completely\nremoves the effects of multicollinearity. This paper also proposes an implicit\nregularization in the classical boosting method and a novel non-parametric stop\ncriterion suitable only for boosting and specifically designed for HPO. The\nsynergy between these two improvements over boosting exhibits competitive and\npromising predictive power performance compared to other existing meta-learners\nand ensemble approaches for HPO other than the stacking ensemble.\n", "date": "2024-02-02", "categories": []}
{"id": "2401.15810", "title": "Green Runner: A tool for efficient deep learning component selection", "abstract": "  For software that relies on machine-learned functionality, model selection is\nkey to finding the right model for the task with desired performance\ncharacteristics. Evaluating a model requires developers to i) select from many\nmodels (e.g. the Hugging face model repository), ii) select evaluation metrics\nand training strategy, and iii) tailor trade-offs based on the problem domain.\nHowever, current evaluation approaches are either ad-hoc resulting in\nsub-optimal model selection or brute force leading to wasted compute. In this\nwork, we present \\toolname, a novel tool to automatically select and evaluate\nmodels based on the application scenario provided in natural language. We\nleverage the reasoning capabilities of large language models to propose a\ntraining strategy and extract desired trade-offs from a problem description.\n\\toolname~features a resource-efficient experimentation engine that integrates\nconstraints and trade-offs based on the problem into the model selection\nprocess. Our preliminary evaluation demonstrates that \\toolname{} is both\nefficient and accurate compared to ad-hoc evaluations and brute force. This\nwork presents an important step toward energy-efficient tools to help reduce\nthe environmental impact caused by the growing demand for software with\nmachine-learned functionality.\n", "date": "2024-01-29", "categories": []}
{"id": "2312.06305", "title": "A Meta-Level Learning Algorithm for Sequential Hyper-Parameter Space\n  Reduction in AutoML", "abstract": "  AutoML platforms have numerous options for the algorithms to try for each\nstep of the analysis, i.e., different possible algorithms for imputation,\ntransformations, feature selection, and modelling. Finding the optimal\ncombination of algorithms and hyper-parameter values is computationally\nexpensive, as the number of combinations to explore leads to an exponential\nexplosion of the space. In this paper, we present the Sequential\nHyper-parameter Space Reduction (SHSR) algorithm that reduces the space for an\nAutoML tool with negligible drop in its predictive performance. SHSR is a\nmeta-level learning algorithm that analyzes past runs of an AutoML tool on\nseveral datasets and learns which hyper-parameter values to filter out from\nconsideration on a new dataset to analyze. SHSR is evaluated on 284\nclassification and 375 regression problems, showing an approximate 30%\nreduction in execution time with a performance drop of less than 0.1%.\n", "date": "2023-12-11", "categories": []}
{"id": "2309.17011", "title": "Feature Interaction Aware Automated Data Representation Transformation", "abstract": "  Creating an effective representation space is crucial for mitigating the\ncurse of dimensionality, enhancing model generalization, addressing data\nsparsity, and leveraging classical models more effectively. Recent advancements\nin automated feature engineering (AutoFE) have made significant progress in\naddressing various challenges associated with representation learning, issues\nsuch as heavy reliance on intensive labor and empirical experiences, lack of\nexplainable explicitness, and inflexible feature space reconstruction embedded\ninto downstream tasks. However, these approaches are constrained by: 1)\ngeneration of potentially unintelligible and illogical reconstructed feature\nspaces, stemming from the neglect of expert-level cognitive processes; 2) lack\nof systematic exploration, which subsequently results in slower model\nconvergence for identification of optimal feature space. To address these, we\nintroduce an interaction-aware reinforced generation perspective. We redefine\nfeature space reconstruction as a nested process of creating meaningful\nfeatures and controlling feature set size through selection. We develop a\nhierarchical reinforcement learning structure with cascading Markov Decision\nProcesses to automate feature and operation selection, as well as feature\ncrossing. By incorporating statistical measures, we reward agents based on the\ninteraction strength between selected features, resulting in intelligent and\nefficient exploration of the feature space that emulates human decision-making.\nExtensive experiments are conducted to validate our proposed approach.\n", "date": "2023-09-29", "categories": []}
{"id": "2401.06452", "title": "Automated Machine Learning for Positive-Unlabelled Learning", "abstract": "  Positive-Unlabelled (PU) learning is a growing field of machine learning that\naims to learn classifiers from data consisting of labelled positive and\nunlabelled instances, which can be in reality positive or negative, but whose\nlabel is unknown. An extensive number of methods have been proposed to address\nPU learning over the last two decades, so many so that selecting an optimal\nmethod for a given PU learning task presents a challenge. Our previous work has\naddressed this by proposing GA-Auto-PU, the first Automated Machine Learning\n(Auto-ML) system for PU learning. In this work, we propose two new Auto-ML\nsystems for PU learning: BO-Auto-PU, based on a Bayesian Optimisation approach,\nand EBO-Auto-PU, based on a novel evolutionary/Bayesian optimisation approach.\nWe also present an extensive evaluation of the three Auto-ML systems, comparing\nthem to each other and to well-established PU learning methods across 60\ndatasets (20 real-world datasets, each with 3 versions in terms of PU learning\ncharacteristics).\n", "date": "2024-01-12", "categories": []}
{"id": "2309.03581", "title": "Interactive Hyperparameter Optimization in Multi-Objective Problems via\n  Preference Learning", "abstract": "  Hyperparameter optimization (HPO) is important to leverage the full potential\nof machine learning (ML). In practice, users are often interested in\nmulti-objective (MO) problems, i.e., optimizing potentially conflicting\nobjectives, like accuracy and energy consumption. To tackle this, the vast\nmajority of MO-ML algorithms return a Pareto front of non-dominated machine\nlearning models to the user. Optimizing the hyperparameters of such algorithms\nis non-trivial as evaluating a hyperparameter configuration entails evaluating\nthe quality of the resulting Pareto front. In literature, there are known\nindicators that assess the quality of a Pareto front (e.g., hypervolume, R2) by\nquantifying different properties (e.g., volume, proximity to a reference\npoint). However, choosing the indicator that leads to the desired Pareto front\nmight be a hard task for a user. In this paper, we propose a human-centered\ninteractive HPO approach tailored towards multi-objective ML leveraging\npreference learning to extract desiderata from users that guide the\noptimization. Instead of relying on the user guessing the most suitable\nindicator for their needs, our approach automatically learns an appropriate\nindicator. Concretely, we leverage pairwise comparisons of distinct Pareto\nfronts to learn such an appropriate quality indicator. Then, we optimize the\nhyperparameters of the underlying MO-ML algorithm towards this learned\nindicator using a state-of-the-art HPO approach. In an experimental study\ntargeting the environmental impact of ML, we demonstrate that our approach\nleads to substantially better Pareto fronts compared to optimizing based on a\nwrong indicator pre-selected by the user, and performs comparable in the case\nof an advanced user knowing which indicator to pick.\n", "date": "2023-09-07", "categories": []}
{"id": "2312.17404", "title": "Parameter Optimization with Conscious Allocation (POCA)", "abstract": "  The performance of modern machine learning algorithms depends upon the\nselection of a set of hyperparameters. Common examples of hyperparameters are\nlearning rate and the number of layers in a dense neural network. Auto-ML is a\nbranch of optimization that has produced important contributions in this area.\nWithin Auto-ML, hyperband-based approaches, which eliminate poorly-performing\nconfigurations after evaluating them at low budgets, are among the most\neffective. However, the performance of these algorithms strongly depends on how\neffectively they allocate the computational budget to various hyperparameter\nconfigurations. We present the new Parameter Optimization with Conscious\nAllocation (POCA), a hyperband-based algorithm that adaptively allocates the\ninputted budget to the hyperparameter configurations it generates following a\nBayesian sampling scheme. We compare POCA to its nearest competitor at\noptimizing the hyperparameters of an artificial toy function and a deep neural\nnetwork and find that POCA finds strong configurations faster in both settings.\n", "date": "2023-12-29", "categories": []}
{"id": "2311.04945", "title": "Auto deep learning for bioacoustic signals", "abstract": "  This study investigates the potential of automated deep learning to enhance\nthe accuracy and efficiency of multi-class classification of bird\nvocalizations, compared against traditional manually-designed deep learning\nmodels. Using the Western Mediterranean Wetland Birds dataset, we investigated\nthe use of AutoKeras, an automated machine learning framework, to automate\nneural architecture search and hyperparameter tuning. Comparative analysis\nvalidates our hypothesis that the AutoKeras-derived model consistently\noutperforms traditional models like MobileNet, ResNet50 and VGG16. Our approach\nand findings underscore the transformative potential of automated deep learning\nfor advancing bioacoustics research and models. In fact, the automated\ntechniques eliminate the need for manual feature engineering and model design\nwhile improving performance. This study illuminates best practices in sampling,\nevaluation and reporting to enhance reproducibility in this nascent field. All\nthe code used is available at https:\n//github.com/giuliotosato/AutoKeras-bioacustic\n  Keywords: AutoKeras; automated deep learning; audio classification; Wetlands\nBird dataset; comparative analysis; bioacoustics; validation dataset;\nmulti-class classification; spectrograms.\n", "date": "2023-11-08", "categories": []}
{"id": "2312.14770", "title": "Integration Of Evolutionary Automated Machine Learning With Structural\n  Sensitivity Analysis For Composite Pipelines", "abstract": "  Automated machine learning (AutoML) systems propose an end-to-end solution to\na given machine learning problem, creating either fixed or flexible pipelines.\nFixed pipelines are task independent constructs: their general composition\nremains the same, regardless of the data. In contrast, the structure of\nflexible pipelines varies depending on the input, making them finely tailored\nto individual tasks. However, flexible pipelines can be structurally\novercomplicated and have poor explainability. We propose the EVOSA approach\nthat compensates for the negative points of flexible pipelines by incorporating\na sensitivity analysis which increases the robustness and interpretability of\nthe flexible solutions. EVOSA quantitatively estimates positive and negative\nimpact of an edge or a node on a pipeline graph, and feeds this information to\nthe evolutionary AutoML optimizer. The correctness and efficiency of EVOSA was\nvalidated in tabular, multimodal and computer vision tasks, suggesting\ngeneralizability of the proposed approach across domains.\n", "date": "2023-12-22", "categories": []}
{"id": "2312.12460", "title": "Democratize with Care: The need for fairness specific features in\n  user-interface based open source AutoML tools", "abstract": "  AI is increasingly playing a pivotal role in businesses and organizations,\nimpacting the outcomes and interests of human users. Automated Machine Learning\n(AutoML) streamlines the machine learning model development process by\nautomating repetitive tasks and making data-driven decisions, enabling even\nnon-experts to construct high-quality models efficiently. This democratization\nallows more users (including non-experts) to access and utilize\nstate-of-the-art machine-learning expertise. However, AutoML tools may also\npropagate bias in the way these tools handle the data, model choices, and\noptimization approaches adopted. We conducted an experimental study of\nUser-interface-based open source AutoML tools (DataRobot, H2O Studio, Dataiku,\nand Rapidminer Studio) to examine if they had features to assist users in\ndeveloping fairness-aware machine learning models. The experiments covered the\nfollowing considerations for the evaluation of features: understanding use case\ncontext, data representation, feature relevance and sensitivity, data bias and\npreprocessing techniques, data handling capabilities, training-testing split,\nhyperparameter handling, and constraints, fairness-oriented model development,\nexplainability and ability to download and edit models by the user. The results\nrevealed inadequacies in features that could support in fairness-aware model\ndevelopment. Further, the results also highlight the need to establish certain\nessential features for promoting fairness in AutoML tools.\n", "date": "2023-12-16", "categories": []}
{"id": "2312.13038", "title": "AutoXPCR: Automated Multi-Objective Model Selection for Time Series\n  Forecasting", "abstract": "  Automated machine learning (AutoML) streamlines the creation of ML models.\nWhile most methods select the \"best\" model based on predictive quality, it's\ncrucial to acknowledge other aspects, such as interpretability and resource\nconsumption. This holds particular importance in the context of deep neural\nnetworks (DNNs), as these models are often perceived as computationally\nintensive black boxes. In the challenging domain of time series forecasting,\nDNNs achieve stunning results, but specialized approaches for automatically\nselecting models are scarce. In this paper, we propose AutoXPCR - a novel\nmethod for automated and explainable multi-objective model selection. Our\napproach leverages meta-learning to estimate any model's performance along PCR\ncriteria, which encompass (P)redictive error, (C)omplexity, and (R)esource\ndemand. Explainability is addressed on multiple levels, as our interactive\nframework can prioritize less complex models and provide by-product\nexplanations of recommendations. We demonstrate practical feasibility by\ndeploying AutoXPCR on over 1000 configurations across 114 data sets from\nvarious domains. Our method clearly outperforms other model selection\napproaches - on average, it only requires 20% of computation costs for\nrecommending models with 90% of the best-possible quality.\n", "date": "2023-12-20", "categories": []}
{"id": "2312.05461", "title": "STREAMLINE: An Automated Machine Learning Pipeline for Biomedicine\n  Applied to Examine the Utility of Photography-Based Phenotypes for OSA\n  Prediction Across International Sleep Centers", "abstract": "  While machine learning (ML) includes a valuable array of tools for analyzing\nbiomedical data, significant time and expertise is required to assemble\neffective, rigorous, and unbiased pipelines. Automated ML (AutoML) tools seek\nto facilitate ML application by automating a subset of analysis pipeline\nelements. In this study we develop and validate a Simple, Transparent,\nEnd-to-end Automated Machine Learning Pipeline (STREAMLINE) and apply it to\ninvestigate the added utility of photography-based phenotypes for predicting\nobstructive sleep apnea (OSA); a common and underdiagnosed condition associated\nwith a variety of health, economic, and safety consequences. STREAMLINE is\ndesigned to tackle biomedical binary classification tasks while adhering to\nbest practices and accommodating complexity, scalability, reproducibility,\ncustomization, and model interpretation. Benchmarking analyses validated the\nefficacy of STREAMLINE across data simulations with increasingly complex\npatterns of association. Then we applied STREAMLINE to evaluate the utility of\ndemographics (DEM), self-reported comorbidities (DX), symptoms (SYM), and\nphotography-based craniofacial (CF) and intraoral (IO) anatomy measures in\npredicting any OSA or moderate/severe OSA using 3,111 participants from Sleep\nApnea Global Interdisciplinary Consortium (SAGIC). OSA analyses identified a\nsignificant increase in ROC-AUC when adding CF to DEM+DX+SYM to predict\nmoderate/severe OSA. A consistent but non-significant increase in PRC-AUC was\nobserved with the addition of each subsequent feature set to predict any OSA,\nwith CF and IO yielding minimal improvements. Application of STREAMLINE to OSA\ndata suggests that CF features provide additional value in predicting\nmoderate/severe OSA, but neither CF nor IO features meaningfully improved the\nprediction of any OSA beyond established demographics, comorbidity and symptom\ncharacteristics.\n", "date": "2023-12-09", "categories": []}
{"id": "2312.06121", "title": "Can LLMs Configure Software Tools", "abstract": "  In software engineering, the meticulous configuration of software tools is\ncrucial in ensuring optimal performance within intricate systems. However, the\ncomplexity inherent in selecting optimal configurations is exacerbated by the\nhigh-dimensional search spaces presented in modern applications. Conventional\ntrial-and-error or intuition-driven methods are both inefficient and\nerror-prone, impeding scalability and reproducibility. In this study, we embark\non an exploration of leveraging Large-Language Models (LLMs) to streamline the\nsoftware configuration process. We identify that the task of hyperparameter\nconfiguration for machine learning components within intelligent applications\nis particularly challenging due to the extensive search space and\nperformance-critical nature. Existing methods, including Bayesian optimization,\nhave limitations regarding initial setup, computational cost, and convergence\nefficiency. Our work presents a novel approach that employs LLMs, such as\nChat-GPT, to identify starting conditions and narrow down the search space,\nimproving configuration efficiency. We conducted a series of experiments to\ninvestigate the variability of LLM-generated responses, uncovering intriguing\nfindings such as potential response caching and consistent behavior based on\ndomain-specific keywords. Furthermore, our results from hyperparameter\noptimization experiments reveal the potential of LLMs in expediting\ninitialization processes and optimizing configurations. While our initial\ninsights are promising, they also indicate the need for further in-depth\ninvestigations and experiments in this domain.\n", "date": "2023-12-11", "categories": []}
{"id": "2312.01567", "title": "Toward Automated Quantum Variational Machine Learning", "abstract": "  In this work, we address the problem of automating quantum variational\nmachine learning. We develop a multi-locality parallelizable search algorithm,\ncalled MUSE, to find the initial points and the sets of parameters that achieve\nthe best performance for quantum variational circuit learning. Simulations with\nfive real-world classification datasets indicate that on average, MUSE improves\nthe detection accuracy of quantum variational classifiers 2.3 times with\nrespect to the observed lowest scores. Moreover, when applied to two real-world\nregression datasets, MUSE improves the quality of the predictions from negative\ncoefficients of determination to positive ones. Furthermore, the classification\nand regression scores of the quantum variational models trained with MUSE are\non par with the classical counterparts.\n", "date": "2023-12-04", "categories": []}
{"id": "2311.17124", "title": "A knowledge-driven AutoML architecture", "abstract": "  This paper proposes a knowledge-driven AutoML architecture for pipeline and\ndeep feature synthesis. The main goal is to render the AutoML process\nexplainable and to leverage domain knowledge in the synthesis of pipelines and\nfeatures. The architecture explores several novel ideas: first, the\nconstruction of pipelines and deep features is approached in an unified way.\nNext, synthesis is driven by a shared knowledge system, interactively queried\nas to what pipeline operations to use or features to compute. Lastly, the\nsynthesis processes takes decisions at runtime using partial solutions and\nresults of their application on data. Two experiments are conducted to\ndemonstrate the functionality of a na\\\"{\\i}ve implementation of the proposed\narchitecture and to discuss its advantages, trade-offs as well as future\npotential for AutoML.\n", "date": "2023-11-28", "categories": []}
{"id": "2306.05734", "title": "DP-HyPO: An Adaptive Private Hyperparameter Optimization Framework", "abstract": "  Hyperparameter optimization, also known as hyperparameter tuning, is a widely\nrecognized technique for improving model performance. Regrettably, when\ntraining private ML models, many practitioners often overlook the privacy risks\nassociated with hyperparameter optimization, which could potentially expose\nsensitive information about the underlying dataset. Currently, the sole\nexisting approach to allow privacy-preserving hyperparameter optimization is to\nuniformly and randomly select hyperparameters for a number of runs,\nsubsequently reporting the best-performing hyperparameter. In contrast, in\nnon-private settings, practitioners commonly utilize ``adaptive''\nhyperparameter optimization methods such as Gaussian process-based\noptimization, which select the next candidate based on information gathered\nfrom previous outputs. This substantial contrast between private and\nnon-private hyperparameter optimization underscores a critical concern. In our\npaper, we introduce DP-HyPO, a pioneering framework for ``adaptive'' private\nhyperparameter optimization, aiming to bridge the gap between private and\nnon-private hyperparameter optimization. To accomplish this, we provide a\ncomprehensive differential privacy analysis of our framework. Furthermore, we\nempirically demonstrate the effectiveness of DP-HyPO on a diverse set of\nreal-world datasets.\n", "date": "2023-06-09", "categories": []}
{"id": "2311.15854", "title": "A systematic study comparing hyperparameter optimization engines on\n  tabular data", "abstract": "  We run an independent comparison of all hyperparameter optimization\n(hyperopt) engines available in the Ray Tune library. We introduce two ways to\nnormalize and aggregate statistics across data sets and models, one rank-based,\nand another one sandwiching the score between the random search score and the\nfull grid search score. This affords us i) to rank the hyperopt engines, ii) to\nmake generalized and statistically significant statements on how much they\nimprove over random search, and iii) to make recommendations on which engine\nshould be used to hyperopt a given learning algorithm. We find that most\nengines beat random search, but that only three of them (HEBO, AX, and\nBlendSearch) clearly stand out. We also found that some engines seem to\nspecialize in hyperopting certain learning algorithms, which makes it tricky to\nuse hyperopt in comparison studies, since the choice of the hyperopt technique\nmay favor some of the models in the comparison.\n", "date": "2023-11-27", "categories": []}
{"id": "2311.06622", "title": "TrainerAgent: Customizable and Efficient Model Training through\n  LLM-Powered Multi-Agent System", "abstract": "  Training AI models has always been challenging, especially when there is a\nneed for custom models to provide personalized services. Algorithm engineers\noften face a lengthy process to iteratively develop models tailored to specific\nbusiness requirements, making it even more difficult for non-experts. The quest\nfor high-quality and efficient model development, along with the emergence of\nLarge Language Model (LLM) Agents, has become a key focus in the industry.\nLeveraging the powerful analytical, planning, and decision-making capabilities\nof LLM, we propose a TrainerAgent system comprising a multi-agent framework\nincluding Task, Data, Model and Server agents. These agents analyze\nuser-defined tasks, input data, and requirements (e.g., accuracy, speed),\noptimizing them comprehensively from both data and model perspectives to obtain\nsatisfactory models, and finally deploy these models as online service.\nExperimental evaluations on classical discriminative and generative tasks in\ncomputer vision and natural language processing domains demonstrate that our\nsystem consistently produces models that meet the desired criteria.\nFurthermore, the system exhibits the ability to critically identify and reject\nunattainable tasks, such as fantastical scenarios or unethical requests,\nensuring robustness and safety. This research presents a significant\nadvancement in achieving desired models with increased efficiency and quality\nas compared to traditional model development, facilitated by the integration of\nLLM-powered analysis, decision-making, and execution capabilities, as well as\nthe collaboration among four agents. We anticipate that our work will\ncontribute to the advancement of research on TrainerAgent in both academic and\nindustry communities, potentially establishing it as a new paradigm for model\ndevelopment in the field of AI.\n", "date": "2023-11-11", "categories": []}
{"id": "2311.11963", "title": "What Can AutoML Do For Continual Learning?", "abstract": "  This position paper outlines the potential of AutoML for incremental\n(continual) learning to encourage more research in this direction. Incremental\nlearning involves incorporating new data from a stream of tasks and\ndistributions to learn enhanced deep representations and adapt better to new\ntasks. However, a significant limitation of incremental learners is that most\ncurrent techniques freeze the backbone architecture, hyperparameters, and the\norder & structure of the learning tasks throughout the learning and adaptation\nprocess. We strongly believe that AutoML offers promising solutions to address\nthese limitations, enabling incremental learning to adapt to more diverse\nreal-world tasks. Therefore, instead of directly proposing a new method, this\npaper takes a step back by posing the question: \"What can AutoML do for\nincremental learning?\" We outline three key areas of research that can\ncontribute to making incremental learners more dynamic, highlighting concrete\nopportunities to apply AutoML methods in novel ways as well as entirely new\nchallenges for AutoML research.\n", "date": "2023-11-20", "categories": []}
{"id": "2311.07870", "title": "AutoML for Large Capacity Modeling of Meta's Ranking Systems", "abstract": "  Web-scale ranking systems at Meta serving billions of users is complex.\nImproving ranking models is essential but engineering heavy. Automated Machine\nLearning (AutoML) can release engineers from labor intensive work of tuning\nranking models; however, it is unknown if AutoML is efficient enough to meet\ntight production timeline in real-world and, at the same time, bring additional\nimprovements to the strong baselines. Moreover, to achieve higher ranking\nperformance, there is an ever-increasing demand to scale up ranking models to\neven larger capacity, which imposes more challenges on the efficiency. The\nlarge scale of models and tight production schedule requires AutoML to\noutperform human baselines by only using a small number of model evaluation\ntrials (around 100). We presents a sampling-based AutoML method, focusing on\nneural architecture search and hyperparameter optimization, addressing these\nchallenges in Meta-scale production when building large capacity models. Our\napproach efficiently handles large-scale data demands. It leverages a\nlightweight predictor-based searcher and reinforcement learning to explore vast\nsearch spaces, significantly reducing the number of model evaluations. Through\nexperiments in large capacity modeling for CTR and CVR applications, we show\nthat our method achieves outstanding Return on Investment (ROI) versus human\ntuned baselines, with up to 0.09% Normalized Entropy (NE) loss reduction or\n$25\\%$ Query per Second (QPS) increase by only sampling one hundred models on\naverage from a curated search space. The proposed AutoML method has already\nmade real-world impact where a discovered Instagram CTR model with up to -0.36%\nNE gain (over existing production baseline) was selected for large-scale online\nA/B test and show statistically significant gain. These production results\nproved AutoML efficacy and accelerated its adoption in ranking systems at Meta.\n", "date": "2023-11-14", "categories": []}
{"id": "2306.12370", "title": "PriorBand: Practical Hyperparameter Optimization in the Age of Deep\n  Learning", "abstract": "  Hyperparameters of Deep Learning (DL) pipelines are crucial for their\ndownstream performance. While a large number of methods for Hyperparameter\nOptimization (HPO) have been developed, their incurred costs are often\nuntenable for modern DL. Consequently, manual experimentation is still the most\nprevalent approach to optimize hyperparameters, relying on the researcher's\nintuition, domain knowledge, and cheap preliminary explorations. To resolve\nthis misalignment between HPO algorithms and DL researchers, we propose\nPriorBand, an HPO algorithm tailored to DL, able to utilize both expert beliefs\nand cheap proxy tasks. Empirically, we demonstrate PriorBand's efficiency\nacross a range of DL benchmarks and show its gains under informative expert\ninput and robustness against poor expert beliefs\n", "date": "2023-06-21", "categories": []}
{"id": "2311.04034", "title": "Impact of HPO on AutoML Forecasting Ensembles", "abstract": "  A forecasting ensemble consisting of a diverse range of estimators for both\nlocal and global univariate forecasting, in particular MQ-CNN,DeepAR, Prophet,\nNPTS, ARIMA and ETS, can be used to make forecasts for a variety of problems.\nThis paper delves into the aspect of adding different hyperparameter\noptimization strategies to the deep learning models in such a setup (DeepAR and\nMQ-CNN), exploring the trade-off between added training cost and the increase\nin accuracy for different configurations. It shows that in such a setup, adding\nhyperparameter optimization can lead to performance improvements, with the\nfinal setup having a 9.9 % percent accuracy improvement with respect to the\navg-wQL over the baseline ensemble without HPO, accompanied by a 65.8 %\nincrease in end-to-end ensemble latency. This improvement is based on an\nempirical analysis of combining the ensemble pipeline with different tuning\nstrategies, namely Bayesian Optimisation and Hyperband and different\nconfigurations of those strategies. In the final configuration, the proposed\ncombination of ensemble learning and HPO outperforms the state of the art\ncommercial AutoML forecasting solution, Amazon Forecast, with a 3.5 % lower\nerror and 16.0 % lower end-to-end ensemble latency.\n", "date": "2023-11-07", "categories": []}
{"id": "2310.12914", "title": "Network-Aware AutoML Framework for Software-Defined Sensor Networks", "abstract": "  As the current detection solutions of distributed denial of service attacks\n(DDoS) need additional infrastructures to handle high aggregate data rates,\nthey are not suitable for sensor networks or the Internet of Things. Besides,\nthe security architecture of software-defined sensor networks needs to pay\nattention to the vulnerabilities of both software-defined networks and sensor\nnetworks. In this paper, we propose a network-aware automated machine learning\n(AutoML) framework which detects DDoS attacks in software-defined sensor\nnetworks. Our framework selects an ideal machine learning algorithm to detect\nDDoS attacks in network-constrained environments, using metrics such as\nvariable traffic load, heterogeneous traffic rate, and detection time while\npreventing over-fitting. Our contributions are two-fold: (i) we first\ninvestigate the trade-off between the efficiency of ML algorithms and\nnetwork/traffic state in the scope of DDoS detection. (ii) we design and\nimplement a software architecture containing open-source network tools, with\nthe deployment of multiple ML algorithms. Lastly, we show that under the denial\nof service attacks, our framework ensures the traffic packets are still\ndelivered within the network with additional delays.\n", "date": "2023-10-19", "categories": []}
{"id": "2310.11478", "title": "ASP: Automatic Selection of Proxy dataset for efficient AutoML", "abstract": "  Deep neural networks have gained great success due to the increasing amounts\nof data, and diverse effective neural network designs. However, it also brings\na heavy computing burden as the amount of training data is proportional to the\ntraining time. In addition, a well-behaved model requires repeated trials of\ndifferent structure designs and hyper-parameters, which may take a large amount\nof time even with state-of-the-art (SOTA) hyper-parameter optimization (HPO)\nalgorithms and neural architecture search (NAS) algorithms. In this paper, we\npropose an Automatic Selection of Proxy dataset framework (ASP) aimed to\ndynamically find the informative proxy subsets of training data at each epoch,\nreducing the training data size as well as saving the AutoML processing time.\nWe verify the effectiveness and generalization of ASP on CIFAR10, CIFAR100,\nImageNet16-120, and ImageNet-1k, across various public model benchmarks. The\nexperiment results show that ASP can obtain better results than other data\nselection methods at all selection ratios. ASP can also enable much more\nefficient AutoML processing with a speedup of 2x-20x while obtaining better\narchitectures and better hyper-parameters compared to utilizing the entire\ndataset.\n", "date": "2023-10-17", "categories": []}
{"id": "2306.16913", "title": "AutoML in Heavily Constrained Applications", "abstract": "  Optimizing a machine learning pipeline for a task at hand requires careful\nconfiguration of various hyperparameters, typically supported by an AutoML\nsystem that optimizes the hyperparameters for the given training dataset. Yet,\ndepending on the AutoML system's own second-order meta-configuration, the\nperformance of the AutoML process can vary significantly. Current AutoML\nsystems cannot automatically adapt their own configuration to a specific use\ncase. Further, they cannot compile user-defined application constraints on the\neffectiveness and efficiency of the pipeline and its generation. In this paper,\nwe propose CAML, which uses meta-learning to automatically adapt its own AutoML\nparameters, such as the search strategy, the validation strategy, and the\nsearch space, for a task at hand. The dynamic AutoML strategy of CAML takes\nuser-defined constraints into account and obtains constraint-satisfying\npipelines with high predictive performance.\n", "date": "2023-06-29", "categories": []}
{"id": "2310.04238", "title": "Bringing Quantum Algorithms to Automated Machine Learning: A Systematic\n  Review of AutoML Frameworks Regarding Extensibility for QML Algorithms", "abstract": "  This work describes the selection approach and analysis of existing AutoML\nframeworks regarding their capability of a) incorporating Quantum Machine\nLearning (QML) algorithms into this automated solving approach of the AutoML\nframing and b) solving a set of industrial use-cases with different ML problem\ntypes by benchmarking their most important characteristics. For that, available\nopen-source tools are condensed into a market overview and suitable frameworks\nare systematically selected on a multi-phase, multi-criteria approach. This is\ndone by considering software selection approaches, as well as in terms of the\ntechnical perspective of AutoML. The requirements for the framework selection\nare divided into hard and soft criteria regarding their software and ML\nattributes. Additionally, a classification of AutoML frameworks is made into\nhigh- and low-level types, inspired by the findings of. Finally, we select Ray\nand AutoGluon as the suitable low- and high-level frameworks respectively, as\nthey fulfil all requirements sufficiently and received the best evaluation\nfeedback during the use-case study. Based on those findings, we build an\nextended Automated Quantum Machine Learning (AutoQML) framework with\nQC-specific pipeline steps and decision characteristics for hardware and\nsoftware constraints.\n", "date": "2023-10-06", "categories": []}
{"id": "2305.03403", "title": "Large Language Models for Automated Data Science: Introducing CAAFE for\n  Context-Aware Automated Feature Engineering", "abstract": "  As the field of automated machine learning (AutoML) advances, it becomes\nincreasingly important to incorporate domain knowledge into these systems. We\npresent an approach for doing so by harnessing the power of large language\nmodels (LLMs). Specifically, we introduce Context-Aware Automated Feature\nEngineering (CAAFE), a feature engineering method for tabular datasets that\nutilizes an LLM to iteratively generate additional semantically meaningful\nfeatures for tabular datasets based on the description of the dataset. The\nmethod produces both Python code for creating new features and explanations for\nthe utility of the generated features. Despite being methodologically simple,\nCAAFE improves performance on 11 out of 14 datasets -- boosting mean ROC AUC\nperformance from 0.798 to 0.822 across all dataset - similar to the improvement\nachieved by using a random forest instead of logistic regression on our\ndatasets. Furthermore, CAAFE is interpretable by providing a textual\nexplanation for each generated feature. CAAFE paves the way for more extensive\nsemi-automation in data science tasks and emphasizes the significance of\ncontext-aware solutions that can extend the scope of AutoML systems to semantic\nAutoML. We release our $\\href{https://github.com/automl/CAAFE}{code}$, a simple\n$\\href{https://colab.research.google.com/drive/1mCA8xOAJZ4MaB_alZvyARTMjhl6RZf0a}{demo}$\nand a $\\href{https://pypi.org/project/caafe/}{python\\ package}$.\n", "date": "2023-05-05", "categories": []}
{"id": "2304.14735", "title": "Benchmarking Automated Machine Learning Methods for Price Forecasting\n  Applications", "abstract": "  Price forecasting for used construction equipment is a challenging task due\nto spatial and temporal price fluctuations. It is thus of high interest to\nautomate the forecasting process based on current market data. Even though\napplying machine learning (ML) to these data represents a promising approach to\npredict the residual value of certain tools, it is hard to implement for small\nand medium-sized enterprises due to their insufficient ML expertise. To this\nend, we demonstrate the possibility of substituting manually created ML\npipelines with automated machine learning (AutoML) solutions, which\nautomatically generate the underlying pipelines. We combine AutoML methods with\nthe domain knowledge of the companies. Based on the CRISP-DM process, we split\nthe manual ML pipeline into a machine learning and non-machine learning part.\nTo take all complex industrial requirements into account and to demonstrate the\napplicability of our new approach, we designed a novel metric named method\nevaluation score, which incorporates the most important technical and\nnon-technical metrics for quality and usability. Based on this metric, we show\nin a case study for the industrial use case of price forecasting, that domain\nknowledge combined with AutoML can weaken the dependence on ML experts for\ninnovative small and medium-sized enterprises which are interested in\nconducting such solutions.\n", "date": "2023-04-28", "categories": []}
{"id": "2307.15422", "title": "Is One Epoch All You Need For Multi-Fidelity Hyperparameter\n  Optimization?", "abstract": "  Hyperparameter optimization (HPO) is crucial for fine-tuning machine learning\nmodels but can be computationally expensive. To reduce costs, Multi-fidelity\nHPO (MF-HPO) leverages intermediate accuracy levels in the learning process and\ndiscards low-performing models early on. We compared various representative\nMF-HPO methods against a simple baseline on classical benchmark data. The\nbaseline involved discarding all models except the Top-K after training for\nonly one epoch, followed by further training to select the best model.\nSurprisingly, this baseline achieved similar results to its counterparts, while\nrequiring an order of magnitude less computation. Upon analyzing the learning\ncurves of the benchmark data, we observed a few dominant learning curves, which\nexplained the success of our baseline. This suggests that researchers should\n(1) always use the suggested baseline in benchmarks and (2) broaden the\ndiversity of MF-HPO benchmarks to include more complex cases.\n", "date": "2023-07-28", "categories": []}
{"id": "2309.06604", "title": "Hybrid Algorithm Selection and Hyperparameter Tuning on Distributed\n  Machine Learning Resources: A Hierarchical Agent-based Approach", "abstract": "  Algorithm selection and hyperparameter tuning are critical steps in both\nacademic and applied machine learning. On the other hand, these steps are\nbecoming ever increasingly delicate due to the extensive rise in the number,\ndiversity, and distributedness of machine learning resources. Multi-agent\nsystems, when applied to the design of machine learning platforms, bring about\nseveral distinctive characteristics such as scalability, flexibility, and\nrobustness, just to name a few. This paper proposes a fully automatic and\ncollaborative agent-based mechanism for selecting distributedly organized\nmachine learning algorithms and simultaneously tuning their hyperparameters.\nOur method builds upon an existing agent-based hierarchical machine-learning\nplatform and augments its query structure to support the aforementioned\nfunctionalities without being limited to specific learning, selection, and\ntuning mechanisms. We have conducted theoretical assessments, formal\nverification, and analytical study to demonstrate the correctness, resource\nutilization, and computational efficiency of our technique. According to the\nresults, our solution is totally correct and exhibits linear time and space\ncomplexity in relation to the size of available resources. To provide concrete\nexamples of how the proposed methodologies can effectively adapt and perform\nacross a range of algorithmic options and datasets, we have also conducted a\nseries of experiments using a system comprised of 24 algorithms and 9 datasets.\n", "date": "2023-09-12", "categories": []}
{"id": "2309.01125", "title": "AutoML-GPT: Large Language Model for AutoML", "abstract": "  With the emerging trend of GPT models, we have established a framework called\nAutoML-GPT that integrates a comprehensive set of tools and libraries. This\nframework grants users access to a wide range of data preprocessing techniques,\nfeature engineering methods, and model selection algorithms. Through a\nconversational interface, users can specify their requirements, constraints,\nand evaluation metrics. Throughout the process, AutoML-GPT employs advanced\ntechniques for hyperparameter optimization and model selection, ensuring that\nthe resulting model achieves optimal performance. The system effectively\nmanages the complexity of the machine learning pipeline, guiding users towards\nthe best choices without requiring deep domain knowledge. Through our\nexperimental results on diverse datasets, we have demonstrated that AutoML-GPT\nsignificantly reduces the time and effort required for machine learning tasks.\nIts ability to leverage the vast knowledge encoded in large language models\nenables it to provide valuable insights, identify potential pitfalls, and\nsuggest effective solutions to common challenges faced during model training.\n", "date": "2023-09-03", "categories": []}
{"id": "2308.15647", "title": "A General Recipe for Automated Machine Learning in Practice", "abstract": "  Automated Machine Learning (AutoML) is an area of research that focuses on\ndeveloping methods to generate machine learning models automatically. The idea\nof being able to build machine learning models with very little human\nintervention represents a great opportunity for the practice of applied machine\nlearning. However, there is very little information on how to design an AutoML\nsystem in practice. Most of the research focuses on the problems facing\noptimization algorithms and leaves out the details of how that would be done in\npractice. In this paper, we propose a frame of reference for building general\nAutoML systems. Through a narrative review of the main approaches in the area,\nour main idea is to distill the fundamental concepts in order to support them\nin a single design. Finally, we discuss some open problems related to the\napplication of AutoML for future research.\n", "date": "2023-08-29", "categories": []}
{"id": "2308.10915", "title": "DiffPrep: Differentiable Data Preprocessing Pipeline Search for Learning\n  over Tabular Data", "abstract": "  Data preprocessing is a crucial step in the machine learning process that\ntransforms raw data into a more usable format for downstream ML models.\nHowever, it can be costly and time-consuming, often requiring the expertise of\ndomain experts. Existing automated machine learning (AutoML) frameworks claim\nto automate data preprocessing. However, they often use a restricted search\nspace of data preprocessing pipelines which limits the potential performance\ngains, and they are often too slow as they require training the ML model\nmultiple times. In this paper, we propose DiffPrep, a method that can\nautomatically and efficiently search for a data preprocessing pipeline for a\ngiven tabular dataset and a differentiable ML model such that the performance\nof the ML model is maximized. We formalize the problem of data preprocessing\npipeline search as a bi-level optimization problem. To solve this problem\nefficiently, we transform and relax the discrete, non-differential search space\ninto a continuous and differentiable one, which allows us to perform the\npipeline search using gradient descent with training the ML model only once.\nOur experiments show that DiffPrep achieves the best test accuracy on 15 out of\nthe 18 real-world datasets evaluated and improves the model's test accuracy by\nup to 6.6 percentage points.\n", "date": "2023-08-20", "categories": []}
{"id": "2308.09947", "title": "Study on the effectiveness of AutoML in detecting cardiovascular disease", "abstract": "  Cardiovascular diseases are widespread among patients with chronic\nnoncommunicable diseases and are one of the leading causes of death, including\nin the working age. The article presents the relevance of the development and\napplication of patient-oriented systems, in which machine learning (ML) is a\npromising technology that allows predicting cardiovascular diseases. Automated\nmachine learning (AutoML) makes it possible to simplify and speed up the\nprocess of developing AI/ML applications, which is key in the development of\npatient-oriented systems by application users, in particular medical\nspecialists. The authors propose a framework for the application of automatic\nmachine learning and three scenarios that allowed for data combining five data\nsets of cardiovascular disease indicators from the UCI Machine Learning\nRepository to investigate the effectiveness in detecting this class of\ndiseases. The study investigated one AutoML model that used and optimized the\nhyperparameters of thirteen basic ML models (KNeighborsUnif, KNeighborsDist,\nLightGBMXT, LightGBM, RandomForestGini, RandomForestEntr, CatBoost,\nExtraTreesGini, ExtraTreesEntr, NeuralNetFastA, XGBoost, NeuralNetTorch,\nLightGBMLarge) and included the most accurate models in the weighted ensemble.\nThe results of the study showed that the structure of the AutoML model for\ndetecting cardiovascular diseases depends not only on the efficiency and\naccuracy of the basic models used, but also on the scenarios for preprocessing\nthe initial data, in particular, on the technique of data normalization. The\ncomparative analysis showed that the accuracy of the AutoML model in detecting\ncardiovascular disease varied in the range from 87.41% to 92.3%, and the\nmaximum accuracy was obtained when normalizing the source data into binary\nvalues, and the minimum was obtained when using the built-in AutoML technique.\n", "date": "2023-08-19", "categories": []}
{"id": "2308.05566", "title": "AutoGluon-TimeSeries: AutoML for Probabilistic Time Series Forecasting", "abstract": "  We introduce AutoGluon-TimeSeries - an open-source AutoML library for\nprobabilistic time series forecasting. Focused on ease of use and robustness,\nAutoGluon-TimeSeries enables users to generate accurate point and quantile\nforecasts with just 3 lines of Python code. Built on the design philosophy of\nAutoGluon, AutoGluon-TimeSeries leverages ensembles of diverse forecasting\nmodels to deliver high accuracy within a short training time.\nAutoGluon-TimeSeries combines both conventional statistical models,\nmachine-learning based forecasting approaches, and ensembling techniques. In\nour evaluation on 29 benchmark datasets, AutoGluon-TimeSeries demonstrates\nstrong empirical performance, outperforming a range of forecasting methods in\nterms of both point and quantile forecast accuracy, and often even improving\nupon the best-in-hindsight combination of prior methods.\n", "date": "2023-08-10", "categories": []}
{"id": "2308.03317", "title": "HomOpt: A Homotopy-Based Hyperparameter Optimization Method", "abstract": "  Machine learning has achieved remarkable success over the past couple of\ndecades, often attributed to a combination of algorithmic innovations and the\navailability of high-quality data available at scale. However, a third critical\ncomponent is the fine-tuning of hyperparameters, which plays a pivotal role in\nachieving optimal model performance. Despite its significance, hyperparameter\noptimization (HPO) remains a challenging task for several reasons. Many HPO\ntechniques rely on naive search methods or assume that the loss function is\nsmooth and continuous, which may not always be the case. Traditional methods,\nlike grid search and Bayesian optimization, often struggle to quickly adapt and\nefficiently search the loss landscape. Grid search is computationally\nexpensive, while Bayesian optimization can be slow to prime. Since the search\nspace for HPO is frequently high-dimensional and non-convex, it is often\nchallenging to efficiently find a global minimum. Moreover, optimal\nhyperparameters can be sensitive to the specific dataset or task, further\ncomplicating the search process. To address these issues, we propose a new\nhyperparameter optimization method, HomOpt, using a data-driven approach based\non a generalized additive model (GAM) surrogate combined with homotopy\noptimization. This strategy augments established optimization methodologies to\nboost the performance and effectiveness of any given method with faster\nconvergence to the optimum on continuous, discrete, and categorical domain\nspaces. We compare the effectiveness of HomOpt applied to multiple optimization\ntechniques (e.g., Random Search, TPE, Bayes, and SMAC) showing improved\nobjective performance on many standardized machine learning benchmarks and\nchallenging open-set recognition tasks.\n", "date": "2023-08-07", "categories": []}
{"id": "2307.08364", "title": "Q(D)O-ES: Population-based Quality (Diversity) Optimisation for Post Hoc\n  Ensemble Selection in AutoML", "abstract": "  Automated machine learning (AutoML) systems commonly ensemble models post hoc\nto improve predictive performance, typically via greedy ensemble selection\n(GES). However, we believe that GES may not always be optimal, as it performs a\nsimple deterministic greedy search. In this work, we introduce two novel\npopulation-based ensemble selection methods, QO-ES and QDO-ES, and compare them\nto GES. While QO-ES optimises solely for predictive performance, QDO-ES also\nconsiders the diversity of ensembles within the population, maintaining a\ndiverse set of well-performing ensembles during optimisation based on ideas of\nquality diversity optimisation. The methods are evaluated using 71\nclassification datasets from the AutoML benchmark, demonstrating that QO-ES and\nQDO-ES often outrank GES, albeit only statistically significant on validation\ndata. Our results further suggest that diversity can be beneficial for post hoc\nensembling but also increases the risk of overfitting.\n", "date": "2023-07-17", "categories": []}
{"id": "2307.10262", "title": "Hyperparameter Tuning Cookbook: A guide for scikit-learn, PyTorch,\n  river, and spotPython", "abstract": "  This document provides a comprehensive guide to hyperparameter tuning using\nspotPython for scikit-learn, PyTorch, and river. The first part introduces\nspotPython's surrogate model-based optimization process, while the second part\nfocuses on hyperparameter tuning. Several case studies are presented, including\nhyperparameter tuning for sklearn models such as Support Vector Classification,\nRandom Forests, Gradient Boosting (XGB), and K-nearest neighbors (KNN), as well\nas a Hoeffding Adaptive Tree Regressor from river. The integration of\nspotPython into the PyTorch and PyTorch Lightning training workflow is also\ndiscussed. With a hands-on approach and step-by-step explanations, this\ncookbook serves as a practical starting point for anyone interested in\nhyperparameter tuning with Python. Highlights include the interplay between\nTensorboard, PyTorch Lightning, spotPython, and river. This publication is\nunder development, with updates available on the corresponding webpage.\n", "date": "2023-07-17", "categories": []}
{"id": "2307.11004", "title": "Efficient and Joint Hyperparameter and Architecture Search for\n  Collaborative Filtering", "abstract": "  Automated Machine Learning (AutoML) techniques have recently been introduced\nto design Collaborative Filtering (CF) models in a data-specific manner.\nHowever, existing works either search architectures or hyperparameters while\nignoring the fact they are intrinsically related and should be considered\ntogether. This motivates us to consider a joint hyperparameter and architecture\nsearch method to design CF models. However, this is not easy because of the\nlarge search space and high evaluation cost. To solve these challenges, we\nreduce the space by screening out usefulness yperparameter choices through a\ncomprehensive understanding of individual hyperparameters. Next, we propose a\ntwo-stage search algorithm to find proper configurations from the reduced\nspace. In the first stage, we leverage knowledge from subsampled datasets to\nreduce evaluation costs; in the second stage, we efficiently fine-tune top\ncandidate models on the whole dataset. Extensive experiments on real-world\ndatasets show better performance can be achieved compared with both\nhand-designed and previous searched models. Besides, ablation and case studies\ndemonstrate the effectiveness of our search framework.\n", "date": "2023-07-12", "categories": []}
{"id": "2307.04849", "title": "SigOpt Mulch: An Intelligent System for AutoML of Gradient Boosted Trees", "abstract": "  Gradient boosted trees (GBTs) are ubiquitous models used by researchers,\nmachine learning (ML) practitioners, and data scientists because of their\nrobust performance, interpretable behavior, and ease-of-use. One critical\nchallenge in training GBTs is the tuning of their hyperparameters. In practice,\nselecting these hyperparameters is often done manually. Recently, the ML\ncommunity has advocated for tuning hyperparameters through black-box\noptimization and developed state-of-the-art systems to do so. However, applying\nsuch systems to tune GBTs suffers from two drawbacks. First, these systems are\nnot \\textit{model-aware}, rather they are designed to apply to a\n\\textit{generic} model; this leaves significant optimization performance on the\ntable. Second, using these systems requires \\textit{domain knowledge} such as\nthe choice of hyperparameter search space, which is an antithesis to the\nautomatic experimentation that black-box optimization aims to provide. In this\npaper, we present SigOpt Mulch, a model-aware hyperparameter tuning system\nspecifically designed for automated tuning of GBTs that provides two\nimprovements over existing systems. First, Mulch leverages powerful techniques\nin metalearning and multifidelity optimization to perform model-aware\nhyperparameter optimization. Second, it automates the process of learning\nperformant hyperparameters by making intelligent decisions about the\noptimization search space, thus reducing the need for user domain knowledge.\nThese innovations allow Mulch to identify good GBT hyperparameters far more\nefficiently -- and in a more seamless and user-friendly way -- than existing\nblack-box hyperparameter tuning systems.\n", "date": "2023-07-10", "categories": []}
{"id": "2303.00501", "title": "OmniForce: On Human-Centered, Large Model Empowered and Cloud-Edge\n  Collaborative AutoML System", "abstract": "  Automated machine learning (AutoML) seeks to build ML models with minimal\nhuman effort. While considerable research has been conducted in the area of\nAutoML in general, aiming to take humans out of the loop when building\nartificial intelligence (AI) applications, scant literature has focused on how\nAutoML works well in open-environment scenarios such as the process of training\nand updating large models, industrial supply chains or the industrial\nmetaverse, where people often face open-loop problems during the search\nprocess: they must continuously collect data, update data and models, satisfy\nthe requirements of the development and deployment environment, support massive\ndevices, modify evaluation metrics, etc. Addressing the open-environment issue\nwith pure data-driven approaches requires considerable data, computing\nresources, and effort from dedicated data engineers, making current AutoML\nsystems and platforms inefficient and computationally intractable.\nHuman-computer interaction is a practical and feasible way to tackle the\nproblem of open-environment AI. In this paper, we introduce OmniForce, a\nhuman-centered AutoML (HAML) system that yields both human-assisted ML and\nML-assisted human techniques, to put an AutoML system into practice and build\nadaptive AI in open-environment scenarios. Specifically, we present OmniForce\nin terms of ML version management; pipeline-driven development and deployment\ncollaborations; a flexible search strategy framework; and widely provisioned\nand crowdsourced application algorithms, including large models. Furthermore,\nthe (large) models constructed by OmniForce can be automatically turned into\nremote services in a few minutes; this process is dubbed model as a service\n(MaaS). Experimental results obtained in multiple search spaces and real-world\nuse cases demonstrate the efficacy and efficiency of OmniForce.\n", "date": "2023-03-01", "categories": []}
{"id": "2207.09849", "title": "Automated machine learning for borehole resistivity measurements", "abstract": "  Deep neural networks (DNNs) offer a real-time solution for the inversion of\nborehole resistivity measurements to approximate forward and inverse operators.\nIt is possible to use extremely large DNNs to approximate the operators, but it\ndemands a considerable training time. Moreover, evaluating the network after\ntraining also requires a significant amount of memory and processing power. In\naddition, we may overfit the model. In this work, we propose a scoring function\nthat accounts for the accuracy and size of the DNNs compared to a reference DNN\nthat provides a good approximation for the operators. Using this scoring\nfunction, we use DNN architecture search algorithms to obtain a quasi-optimal\nDNN smaller than the reference network; hence, it requires less computational\neffort during training and evaluation. The quasi-optimal DNN delivers\ncomparable accuracy to the original large DNN.\n", "date": "2022-07-20", "categories": []}
{"id": "2307.00285", "title": "Assembled-OpenML: Creating Efficient Benchmarks for Ensembles in AutoML\n  with OpenML", "abstract": "  Automated Machine Learning (AutoML) frameworks regularly use ensembles.\nDevelopers need to compare different ensemble techniques to select appropriate\ntechniques for an AutoML framework from the many potential techniques. So far,\nthe comparison of ensemble techniques is often computationally expensive,\nbecause many base models must be trained and evaluated one or multiple times.\nTherefore, we present Assembled-OpenML. Assembled-OpenML is a Python tool,\nwhich builds meta-datasets for ensembles using OpenML. A meta-dataset, called\nMetatask, consists of the data of an OpenML task, the task's dataset, and\nprediction data from model evaluations for the task. We can make the comparison\nof ensemble techniques computationally cheaper by using the predictions stored\nin a metatask instead of training and evaluating base models. To introduce\nAssembled-OpenML, we describe the first version of our tool. Moreover, we\npresent an example of using Assembled-OpenML to compare a set of ensemble\ntechniques. For this example comparison, we built a benchmark using\nAssembled-OpenML and implemented ensemble techniques expecting predictions\ninstead of base models as input. In our example comparison, we gathered the\nprediction data of $1523$ base models for $31$ datasets. Obtaining the\nprediction data for all base models using Assembled-OpenML took ${\\sim} 1$ hour\nin total. In comparison, obtaining the prediction data by training and\nevaluating just one base model on the most computationally expensive dataset\ntook ${\\sim} 37$ minutes.\n", "date": "2023-07-01", "categories": []}
{"id": "2307.00286", "title": "CMA-ES for Post Hoc Ensembling in AutoML: A Great Success and\n  Salvageable Failure", "abstract": "  Many state-of-the-art automated machine learning (AutoML) systems use greedy\nensemble selection (GES) by Caruana et al. (2004) to ensemble models found\nduring model selection post hoc. Thereby, boosting predictive performance and\nlikely following Auto-Sklearn 1's insight that alternatives, like stacking or\ngradient-free numerical optimization, overfit. Overfitting in Auto-Sklearn 1 is\nmuch more likely than in other AutoML systems because it uses only low-quality\nvalidation data for post hoc ensembling. Therefore, we were motivated to\nanalyze whether Auto-Sklearn 1's insight holds true for systems with\nhigher-quality validation data. Consequently, we compared the performance of\ncovariance matrix adaptation evolution strategy (CMA-ES), state-of-the-art\ngradient-free numerical optimization, to GES on the 71 classification datasets\nfrom the AutoML benchmark for AutoGluon. We found that Auto-Sklearn's insight\ndepends on the chosen metric. For the metric ROC AUC, CMA-ES overfits\ndrastically and is outperformed by GES -- statistically significantly for\nmulti-class classification. For the metric balanced accuracy, CMA-ES does not\noverfit and outperforms GES significantly. Motivated by the successful\napplication of CMA-ES for balanced accuracy, we explored methods to stop CMA-ES\nfrom overfitting for ROC AUC. We propose a method to normalize the weights\nproduced by CMA-ES, inspired by GES, that avoids overfitting for CMA-ES and\nmakes CMA-ES perform better than or similar to GES for ROC AUC.\n", "date": "2023-07-01", "categories": []}
{"id": "2305.11390", "title": "ALT: An Automatic System for Long Tail Scenario Modeling", "abstract": "  In this paper, we consider the problem of long tail scenario modeling with\nbudget limitation, i.e., insufficient human resources for model training stage\nand limited time and computing resources for model inference stage. This\nproblem is widely encountered in various applications, yet has received\ndeficient attention so far. We present an automatic system named ALT to deal\nwith this problem. Several efforts are taken to improve the algorithms used in\nour system, such as employing various automatic machine learning related\ntechniques, adopting the meta learning philosophy, and proposing an essential\nbudget-limited neural architecture search method, etc. Moreover, to build the\nsystem, many optimizations are performed from a systematic perspective, and\nessential modules are armed, making the system more feasible and efficient. We\nperform abundant experiments to validate the effectiveness of our system and\ndemonstrate the usefulness of the critical modules in our system. Moreover,\nonline results are provided, which fully verified the efficacy of our system.\n", "date": "2023-05-19", "categories": []}
{"id": "2211.12507", "title": "OpenFE: Automated Feature Generation with Expert-level Performance", "abstract": "  The goal of automated feature generation is to liberate machine learning\nexperts from the laborious task of manual feature generation, which is crucial\nfor improving the learning performance of tabular data. The major challenge in\nautomated feature generation is to efficiently and accurately identify\neffective features from a vast pool of candidate features. In this paper, we\npresent OpenFE, an automated feature generation tool that provides competitive\nresults against machine learning experts. OpenFE achieves high efficiency and\naccuracy with two components: 1) a novel feature boosting method for accurately\nevaluating the incremental performance of candidate features and 2) a two-stage\npruning algorithm that performs feature pruning in a coarse-to-fine manner.\nExtensive experiments on ten benchmark datasets show that OpenFE outperforms\nexisting baseline methods by a large margin. We further evaluate OpenFE in two\nKaggle competitions with thousands of data science teams participating. In the\ntwo competitions, features generated by OpenFE with a simple baseline model can\nbeat 99.3% and 99.6% data science teams respectively. In addition to the\nempirical results, we provide a theoretical perspective to show that feature\ngeneration can be beneficial in a simple yet representative setting. The code\nis available at https://github.com/ZhangTP1996/OpenFE.\n", "date": "2022-11-22", "categories": []}
{"id": "2305.14009", "title": "Deep Pipeline Embeddings for AutoML", "abstract": "  Automated Machine Learning (AutoML) is a promising direction for\ndemocratizing AI by automatically deploying Machine Learning systems with\nminimal human expertise. The core technical challenge behind AutoML is\noptimizing the pipelines of Machine Learning systems (e.g. the choice of\npreprocessing, augmentations, models, optimizers, etc.). Existing Pipeline\nOptimization techniques fail to explore deep interactions between pipeline\nstages/components. As a remedy, this paper proposes a novel neural architecture\nthat captures the deep interaction between the components of a Machine Learning\npipeline. We propose embedding pipelines into a latent representation through a\nnovel per-component encoder mechanism. To search for optimal pipelines, such\npipeline embeddings are used within deep-kernel Gaussian Process surrogates\ninside a Bayesian Optimization setup. Furthermore, we meta-learn the parameters\nof the pipeline embedding network using existing evaluations of pipelines on\ndiverse collections of related datasets (a.k.a. meta-datasets). Through\nextensive experiments on three large-scale meta-datasets, we demonstrate that\npipeline embeddings yield state-of-the-art results in Pipeline Optimization.\n", "date": "2023-05-23", "categories": []}
{"id": "2210.12257", "title": "Efficient Automatic Machine Learning via Design Graphs", "abstract": "  Despite the success of automated machine learning (AutoML), which aims to\nfind the best design, including the architecture of deep networks and\nhyper-parameters, conventional AutoML methods are computationally expensive and\nhardly provide insights into the relations of different model design choices.\nTo tackle the challenges, we propose FALCON, an efficient sample-based method\nto search for the optimal model design. Our key insight is to model the design\nspace of possible model designs as a design graph, where the nodes represent\ndesign choices, and the edges denote design similarities. FALCON features 1) a\ntask-agnostic module, which performs message passing on the design graph via a\nGraph Neural Network (GNN), and 2) a task-specific module, which conducts label\npropagation of the known model performance information on the design graph.\nBoth modules are combined to predict the design performances in the design\nspace, navigating the search direction. We conduct extensive experiments on 27\nnode and graph classification tasks from various application domains, and an\nimage classification task on the CIFAR-10 dataset. We empirically show that\nFALCON can efficiently obtain the well-performing designs for each task using\nonly 30 explored nodes. Specifically, FALCON has a comparable time cost with\nthe one-shot approaches while achieving an average improvement of 3.3% compared\nwith the best baselines.\n", "date": "2022-10-21", "categories": []}
{"id": "2305.13926", "title": "Clustering Indices based Automatic Classification Model Selection", "abstract": "  Classification model selection is a process of identifying a suitable model\nclass for a given classification task on a dataset. Traditionally, model\nselection is based on cross-validation, meta-learning, and user preferences,\nwhich are often time-consuming and resource-intensive. The performance of any\nmachine learning classification task depends on the choice of the model class,\nthe learning algorithm, and the dataset's characteristics. Our work proposes a\nnovel method for automatic classification model selection from a set of\ncandidate model classes by determining the empirical model-fitness for a\ndataset based only on its clustering indices. Clustering Indices measure the\nability of a clustering algorithm to induce good quality neighborhoods with\nsimilar data characteristics. We propose a regression task for a given model\nclass, where the clustering indices of a given dataset form the features and\nthe dependent variable represents the expected classification performance. We\ncompute the dataset clustering indices and directly predict the expected\nclassification performance using the learned regressor for each candidate model\nclass to recommend a suitable model class for dataset classification. We\nevaluate our model selection method through cross-validation with 60 publicly\navailable binary class datasets and show that our top3 model recommendation is\naccurate for over 45 of 60 datasets. We also propose an end-to-end Automated ML\nsystem for data classification based on our model selection method. We evaluate\nour end-to-end system against popular commercial and noncommercial Automated ML\nsystems using a different collection of 25 public domain binary class datasets.\nWe show that the proposed system outperforms other methods with an excellent\naverage rank of 1.68.\n", "date": "2023-05-23", "categories": []}
{"id": "2303.15212", "title": "Deep Ranking Ensembles for Hyperparameter Optimization", "abstract": "  Automatically optimizing the hyperparameters of Machine Learning algorithms\nis one of the primary open questions in AI. Existing work in Hyperparameter\nOptimization (HPO) trains surrogate models for approximating the response\nsurface of hyperparameters as a regression task. In contrast, we hypothesize\nthat the optimal strategy for training surrogates is to preserve the ranks of\nthe performances of hyperparameter configurations as a Learning to Rank\nproblem. As a result, we present a novel method that meta-learns neural network\nsurrogates optimized for ranking the configurations' performances while\nmodeling their uncertainty via ensembling. In a large-scale experimental\nprotocol comprising 12 baselines, 16 HPO search spaces and 86 datasets/tasks,\nwe demonstrate that our method achieves new state-of-the-art results in HPO.\n", "date": "2023-03-27", "categories": []}
{"id": "2302.06675", "title": "Symbolic Discovery of Optimization Algorithms", "abstract": "  We present a method to formulate algorithm discovery as program search, and\napply it to discover optimization algorithms for deep neural network training.\nWe leverage efficient search techniques to explore an infinite and sparse\nprogram space. To bridge the large generalization gap between proxy and target\ntasks, we also introduce program selection and simplification strategies. Our\nmethod discovers a simple and effective optimization algorithm, $\\textbf{Lion}$\n($\\textit{Evo$\\textbf{L}$ved S$\\textbf{i}$gn M$\\textbf{o}$me$\\textbf{n}$tum}$).\nIt is more memory-efficient than Adam as it only keeps track of the momentum.\nDifferent from adaptive optimizers, its update has the same magnitude for each\nparameter calculated through the sign operation. We compare Lion with widely\nused optimizers, such as Adam and Adafactor, for training a variety of models\non different tasks. On image classification, Lion boosts the accuracy of ViT by\nup to 2% on ImageNet and saves up to 5x the pre-training compute on JFT. On\nvision-language contrastive learning, we achieve 88.3% $\\textit{zero-shot}$ and\n91.1% $\\textit{fine-tuning}$ accuracy on ImageNet, surpassing the previous best\nresults by 2% and 0.1%, respectively. On diffusion models, Lion outperforms\nAdam by achieving a better FID score and reducing the training compute by up to\n2.3x. For autoregressive, masked language modeling, and fine-tuning, Lion\nexhibits a similar or better performance compared to Adam. Our analysis of Lion\nreveals that its performance gain grows with the training batch size. It also\nrequires a smaller learning rate than Adam due to the larger norm of the update\nproduced by the sign function. Additionally, we examine the limitations of Lion\nand identify scenarios where its improvements are small or not statistically\nsignificant. Lion is also successfully deployed in production systems such as\nGoogle search ads CTR model.\n", "date": "2023-02-13", "categories": []}
{"id": "2305.02499", "title": "AutoML-GPT: Automatic Machine Learning with GPT", "abstract": "  AI tasks encompass a wide range of domains and fields. While numerous AI\nmodels have been designed for specific tasks and applications, they often\nrequire considerable human efforts in finding the right model architecture,\noptimization algorithm, and hyperparameters. Recent advances in large language\nmodels (LLMs) like ChatGPT show remarkable capabilities in various aspects of\nreasoning, comprehension, and interaction. Consequently, we propose developing\ntask-oriented prompts and automatically utilizing LLMs to automate the training\npipeline. To implement this concept, we present the AutoML-GPT, which employs\nGPT as the bridge to diverse AI models and dynamically trains models with\noptimized hyperparameters. AutoML-GPT dynamically takes user requests from the\nmodel and data cards and composes the corresponding prompt paragraph.\nUltimately, with this prompt paragraph, AutoML-GPT will automatically conduct\nthe experiments from data processing to model architecture, hyperparameter\ntuning, and predicted training log. By leveraging {\\ours}'s robust language\ncapabilities and the available AI models, AutoML-GPT can tackle numerous\nintricate AI tasks across various tasks and datasets. This approach achieves\nremarkable results in computer vision, natural language processing, and other\nchallenging areas. Extensive experiments and ablation studies demonstrate that\nour method can be general, effective, and beneficial for many AI tasks.\n", "date": "2023-05-04", "categories": []}
{"id": "2305.00449", "title": "Predictability of Machine Learning Algorithms and Related Feature\n  Extraction Techniques", "abstract": "  This thesis designs a prediction system based on matrix factorization to\npredict the classification accuracy of a specific model on a particular\ndataset. In this thesis, we conduct comprehensive empirical research on more\nthan fifty datasets that we collected from the openml website. We study the\nperformance prediction of three fundamental machine learning algorithms,\nnamely, random forest, XGBoost, and MultiLayer Perceptron(MLP). In particular,\nwe obtain the following results: 1. Predictability of fine-tuned models using\ncoarse-tuned variants. 2. Predictability of MLP using feature extraction\ntechniques. 3. Predict model performance using implicit feedback.\n", "date": "2023-04-30", "categories": []}
{"id": "2304.13636", "title": "AutoCure: Automated Tabular Data Curation Technique for ML Pipelines", "abstract": "  Machine learning algorithms have become increasingly prevalent in multiple\ndomains, such as autonomous driving, healthcare, and finance. In such domains,\ndata preparation remains a significant challenge in developing accurate models,\nrequiring significant expertise and time investment to search the huge search\nspace of well-suited data curation and transformation tools. To address this\nchallenge, we present AutoCure, a novel and configuration-free data curation\npipeline that improves the quality of tabular data. Unlike traditional data\ncuration methods, AutoCure synthetically enhances the density of the clean data\nfraction through an adaptive ensemble-based error detection method and a data\naugmentation module. In practice, AutoCure can be integrated with open source\ntools, e.g., Auto-sklearn, H2O, and TPOT, to promote the democratization of\nmachine learning. As a proof of concept, we provide a comparative evaluation of\nAutoCure against 28 combinations of traditional data curation tools,\ndemonstrating superior performance and predictive accuracy without user\nintervention. Our evaluation shows that AutoCure is an effective approach to\nautomating data preparation and improving the accuracy of machine learning\nmodels.\n", "date": "2023-04-26", "categories": []}
{"id": "2304.12239", "title": "Uni-QSAR: an Auto-ML Tool for Molecular Property Prediction", "abstract": "  Recently deep learning based quantitative structure-activity relationship\n(QSAR) models has shown surpassing performance than traditional methods for\nproperty prediction tasks in drug discovery. However, most DL based QSAR models\nare restricted to limited labeled data to achieve better performance, and also\nare sensitive to model scale and hyper-parameters. In this paper, we propose\nUni-QSAR, a powerful Auto-ML tool for molecule property prediction tasks.\nUni-QSAR combines molecular representation learning (MRL) of 1D sequential\ntokens, 2D topology graphs, and 3D conformers with pretraining models to\nleverage rich representation from large-scale unlabeled data. Without any\nmanual fine-tuning or model selection, Uni-QSAR outperforms SOTA in 21/22 tasks\nof the Therapeutic Data Commons (TDC) benchmark under designed parallel\nworkflow, with an average performance improvement of 6.09\\%. Furthermore, we\ndemonstrate the practical usefulness of Uni-QSAR in drug discovery domains.\n", "date": "2023-04-24", "categories": []}
{"id": "2304.10051", "title": "HyperTuner: A Cross-Layer Multi-Objective Hyperparameter Auto-Tuning\n  Framework for Data Analytic Services", "abstract": "  Hyper-parameters optimization (HPO) is vital for machine learning models.\nBesides model accuracy, other tuning intentions such as model training time and\nenergy consumption are also worthy of attention from data analytic service\nproviders. Hence, it is essential to take both model hyperparameters and system\nparameters into consideration to execute cross-layer multi-objective\nhyperparameter auto-tuning. Towards this challenging target, we propose\nHyperTuner in this paper. To address the formulated high-dimensional black-box\nmulti-objective optimization problem, HyperTuner first conducts multi-objective\nparameter importance ranking with its MOPIR algorithm and then leverages the\nproposed ADUMBO algorithm to find the Pareto-optimal configuration set. During\neach iteration, ADUMBO selects the most promising configuration from the\ngenerated Pareto candidate set via maximizing a new well-designed metric, which\ncan adaptively leverage the uncertainty as well as the predicted mean across\nall the surrogate models along with the iteration times. We evaluate HyperTuner\non our local distributed TensorFlow cluster and experimental results show that\nit is always able to find a better Pareto configuration front superior in both\nconvergence and diversity compared with the other four baseline algorithms.\nBesides, experiments with different training datasets, different optimization\nobjectives and different machine learning platforms verify that HyperTuner can\nwell adapt to various data analytic service scenarios.\n", "date": "2023-04-20", "categories": []}
{"id": "2208.04321", "title": "Neural Architecture Search as Multiobjective Optimization Benchmarks:\n  Problem Formulation and Performance Assessment", "abstract": "  The ongoing advancements in network architecture design have led to\nremarkable achievements in deep learning across various challenging computer\nvision tasks. Meanwhile, the development of neural architecture search (NAS)\nhas provided promising approaches to automating the design of network\narchitectures for lower prediction error. Recently, the emerging application\nscenarios of deep learning have raised higher demands for network architectures\nconsidering multiple design criteria: number of parameters/floating-point\noperations, and inference latency, among others. From an optimization point of\nview, the NAS tasks involving multiple design criteria are intrinsically\nmultiobjective optimization problems; hence, it is reasonable to adopt\nevolutionary multiobjective optimization (EMO) algorithms for tackling them.\nNonetheless, there is still a clear gap confining the related research along\nthis pathway: on the one hand, there is a lack of a general problem formulation\nof NAS tasks from an optimization point of view; on the other hand, there are\nchallenges in conducting benchmark assessments of EMO algorithms on NAS tasks.\nTo bridge the gap: (i) we formulate NAS tasks into general multi-objective\noptimization problems and analyze the complex characteristics from an\noptimization point of view; (ii) we present an end-to-end pipeline, dubbed\n$\\texttt{EvoXBench}$, to generate benchmark test problems for EMO algorithms to\nrun efficiently -- without the requirement of GPUs or Pytorch/Tensorflow; (iii)\nwe instantiate two test suites comprehensively covering two datasets, seven\nsearch spaces, and three hardware devices, involving up to eight objectives.\nBased on the above, we validate the proposed test suites using six\nrepresentative EMO algorithms and provide some empirical analyses. The code of\n$\\texttt{EvoXBench}$ is available from\n$\\href{https://github.com/EMI-Group/EvoXBench}{\\rm{here}}$.\n", "date": "2022-08-08", "categories": []}
{"id": "2304.08597", "title": "eTOP: Early Termination of Pipelines for Faster Training of AutoML\n  Systems", "abstract": "  Recent advancements in software and hardware technologies have enabled the\nuse of AI/ML models in everyday applications has significantly improved the\nquality of service rendered. However, for a given application, finding the\nright AI/ML model is a complex and costly process, that involves the\ngeneration, training, and evaluation of multiple interlinked steps (called\npipelines), such as data pre-processing, feature engineering, selection, and\nmodel tuning. These pipelines are complex (in structure) and costly (both in\ncompute resource and time) to execute end-to-end, with a hyper-parameter\nassociated with each step. AutoML systems automate the search of these\nhyper-parameters but are slow, as they rely on optimizing the pipeline's end\noutput. We propose the eTOP Framework which works on top of any AutoML system\nand decides whether or not to execute the pipeline to the end or terminate at\nan intermediate step. Experimental evaluation on 26 benchmark datasets and\nintegration of eTOPwith MLBox4 reduces the training time of the AutoML system\nupto 40x than baseline MLBox.\n", "date": "2023-04-17", "categories": []}
{"id": "2304.03374", "title": "Optimizing Neural Networks through Activation Function Discovery and\n  Automatic Weight Initialization", "abstract": "  Automated machine learning (AutoML) methods improve upon existing models by\noptimizing various aspects of their design. While present methods focus on\nhyperparameters and neural network topologies, other aspects of neural network\ndesign can be optimized as well. To further the state of the art in AutoML,\nthis dissertation introduces techniques for discovering more powerful\nactivation functions and establishing more robust weight initialization for\nneural networks. These contributions improve performance, but also provide new\nperspectives on neural network optimization. First, the dissertation\ndemonstrates that discovering solutions specialized to specific architectures\nand tasks gives better performance than reusing general approaches. Second, it\nshows that jointly optimizing different components of neural networks is\nsynergistic, and results in better performance than optimizing individual\ncomponents alone. Third, it demonstrates that learned representations are\neasier to optimize than hard-coded ones, creating further opportunities for\nAutoML. The dissertation thus makes concrete progress towards fully automatic\nmachine learning in the future.\n", "date": "2023-04-06", "categories": []}
{"id": "2303.15053", "title": "Hyperparameter optimization, quantum-assisted model performance\n  prediction, and benchmarking of AI-based High Energy Physics workloads using\n  HPC", "abstract": "  Training and Hyperparameter Optimization (HPO) of deep learning-based AI\nmodels are often compute resource intensive and calls for the use of\nlarge-scale distributed resources as well as scalable and resource efficient\nhyperparameter search algorithms. This work studies the potential of using\nmodel performance prediction to aid the HPO process carried out on High\nPerformance Computing systems. In addition, a quantum annealer is used to train\nthe performance predictor and a method is proposed to overcome some of the\nproblems derived from the current limitations in quantum systems as well as to\nincrease the stability of solutions. This allows for achieving results on a\nquantum machine comparable to those obtained on a classical machine, showing\nhow quantum computers could be integrated within classical machine learning\ntuning pipelines.\n  Furthermore, results are presented from the development of a containerized\nbenchmark based on an AI-model for collision event reconstruction that allows\nus to compare and assess the suitability of different hardware accelerators for\ntraining deep neural networks.\n", "date": "2023-03-27", "categories": []}
{"id": "2303.10732", "title": "AutoEn: An AutoML method based on ensembles of predefined Machine\n  Learning pipelines for supervised Traffic Forecasting", "abstract": "  Intelligent Transportation Systems are producing tons of hardly manageable\ntraffic data, which motivates the use of Machine Learning (ML) for data-driven\napplications, such as Traffic Forecasting (TF). TF is gaining relevance due to\nits ability to mitigate traffic congestion by forecasting future traffic\nstates. However, TF poses one big challenge to the ML paradigm, known as the\nModel Selection Problem (MSP): deciding the most suitable combination of data\npreprocessing techniques and ML method for traffic data collected under\ndifferent transportation circumstances. In this context, Automated Machine\nLearning (AutoML), the automation of the ML workflow from data preprocessing to\nmodel validation, arises as a promising strategy to deal with the MSP in\nproblem domains wherein expert ML knowledge is not always an available or\naffordable asset, such as TF. Various AutoML frameworks have been used to\napproach the MSP in TF. Most are based on online optimisation processes to\nsearch for the best-performing pipeline on a given dataset. This online\noptimisation could be complemented with meta-learning to warm-start the search\nphase and/or the construction of ensembles using pipelines derived from the\noptimisation process. However, given the complexity of the search space and the\nhigh computational cost of tuning-evaluating pipelines generated, online\noptimisation is only beneficial when there is a long time to obtain the final\nmodel. Thus, we introduce AutoEn, which is a simple and efficient method for\nautomatically generating multi-classifier ensembles from a predefined set of ML\npipelines. We compare AutoEn against Auto-WEKA and Auto-sklearn, two AutoML\nmethods commonly used in TF. Experimental results demonstrate that AutoEn can\nlead to better or more competitive results in the general-purpose domain and in\nTF.\n", "date": "2023-03-19", "categories": []}
{"id": "2212.02704", "title": "Benchmarking AutoML algorithms on a collection of synthetic\n  classification problems", "abstract": "  Automated machine learning (AutoML) algorithms have grown in popularity due\nto their high performance and flexibility to adapt to different problems and\ndata sets. With the increasing number of AutoML algorithms, deciding which\nwould best suit a given problem becomes increasingly more work. Therefore, it\nis essential to use complex and challenging benchmarks which would be able to\ndifferentiate the AutoML algorithms from each other. This paper compares the\nperformance of four different AutoML algorithms: Tree-based Pipeline\nOptimization Tool (TPOT), Auto-Sklearn, Auto-Sklearn 2, and H2O AutoML. We use\nthe Diverse and Generative ML benchmark (DIGEN), a diverse set of synthetic\ndatasets derived from generative functions designed to highlight the strengths\nand weaknesses of the performance of common machine learning algorithms. We\nconfirm that AutoML can identify pipelines that perform well on all included\ndatasets. Most AutoML algorithms performed similarly; however, there were some\ndifferences depending on the specific dataset and metric used.\n", "date": "2022-12-06", "categories": []}
{"id": "2302.14329", "title": "Towards Personalized Preprocessing Pipeline Search", "abstract": "  Feature preprocessing, which transforms raw input features into numerical\nrepresentations, is a crucial step in automated machine learning (AutoML)\nsystems. However, the existing systems often have a very small search space for\nfeature preprocessing with the same preprocessing pipeline applied to all the\nnumerical features. This may result in sub-optimal performance since different\ndatasets often have various feature characteristics, and features within a\ndataset may also have their own preprocessing preferences. To bridge this gap,\nwe explore personalized preprocessing pipeline search, where the search\nalgorithm is allowed to adopt a different preprocessing pipeline for each\nfeature. This is a challenging task because the search space grows\nexponentially with more features. To tackle this challenge, we propose\nClusterP3S, a novel framework for Personalized Preprocessing Pipeline Search\nvia Clustering. The key idea is to learn feature clusters such that the search\nspace can be significantly reduced by using the same preprocessing pipeline for\nthe features within a cluster. To this end, we propose a hierarchical search\nstrategy to jointly learn the clusters and search for the optimal pipelines,\nwhere the upper-level search optimizes the feature clustering to enable better\npipelines built upon the clusters, and the lower-level search optimizes the\npipeline given a specific cluster assignment. We instantiate this idea with a\ndeep clustering network that is trained with reinforcement learning at the\nupper level, and random search at the lower level. Experiments on benchmark\nclassification datasets demonstrate the effectiveness of enabling feature-wise\npreprocessing pipeline search.\n", "date": "2023-02-28", "categories": []}
{"id": "2302.10298", "title": "Quantum Machine Learning hyperparameter search", "abstract": "  This paper presents a quantum-based Fourier-regression approach for machine\nlearning hyperparameter optimization applied to a benchmark of models trained\non a dataset related to a forecast problem in the airline industry. Our\napproach utilizes the Fourier series method to represent the hyperparameter\nsearch space, which is then optimized using quantum algorithms to find the\noptimal set of hyperparameters for a given machine learning model. Our study\nevaluates the proposed method on a benchmark of models trained to predict a\nforecast problem in the airline industry using a standard HyperParameter\nOptimizer (HPO). The results show that our approach outperforms traditional\nhyperparameter optimization methods in terms of accuracy and convergence speed\nfor the given search space. Our study provides a new direction for future\nresearch in quantum-based machine learning hyperparameter optimization.\n", "date": "2023-02-20", "categories": []}
{"id": "2302.05433", "title": "Unified Functional Hashing in Automatic Machine Learning", "abstract": "  The field of Automatic Machine Learning (AutoML) has recently attained\nimpressive results, including the discovery of state-of-the-art machine\nlearning solutions, such as neural image classifiers. This is often done by\napplying an evolutionary search method, which samples multiple candidate\nsolutions from a large space and evaluates the quality of each candidate\nthrough a long training process. As a result, the search tends to be slow. In\nthis paper, we show that large efficiency gains can be obtained by employing a\nfast unified functional hash, especially through the functional equivalence\ncaching technique, which we also present. The central idea is to detect by\nhashing when the search method produces equivalent candidates, which occurs\nvery frequently, and this way avoid their costly re-evaluation. Our hash is\n\"functional\" in that it identifies equivalent candidates even if they were\nrepresented or coded differently, and it is \"unified\" in that the same\nalgorithm can hash arbitrary representations; e.g. compute graphs, imperative\ncode, or lambda functions. As evidence, we show dramatic improvements on\nmultiple AutoML domains, including neural architecture search and algorithm\ndiscovery. Finally, we consider the effect of hash collisions, evaluation\nnoise, and search distribution through empirical analysis. Altogether, we hope\nthis paper may serve as a guide to hashing techniques in AutoML.\n", "date": "2023-02-10", "categories": []}
{"id": "2302.03255", "title": "DivBO: Diversity-aware CASH for Ensemble Learning", "abstract": "  The Combined Algorithm Selection and Hyperparameters optimization (CASH)\nproblem is one of the fundamental problems in Automated Machine Learning\n(AutoML). Motivated by the success of ensemble learning, recent AutoML systems\nbuild post-hoc ensembles to output the final predictions instead of using the\nbest single learner. However, while most CASH methods focus on searching for a\nsingle learner with the best performance, they neglect the diversity among base\nlearners (i.e., they may suggest similar configurations to previously evaluated\nones), which is also a crucial consideration when building an ensemble. To\ntackle this issue and further enhance the ensemble performance, we propose\nDivBO, a diversity-aware framework to inject explicit search of diversity into\nthe CASH problems. In the framework, we propose to use a diversity surrogate to\npredict the pair-wise diversity of two unseen configurations. Furthermore, we\nintroduce a temporary pool and a weighted acquisition function to guide the\nsearch of both performance and diversity based on Bayesian optimization.\nEmpirical results on 15 public datasets show that DivBO achieves the best\naverage ranks (1.82 and 1.73) on both validation and test errors among 10\ncompared methods, including post-hoc designs in recent AutoML systems and\nstate-of-the-art baselines for ensemble learning on CASH problems.\n", "date": "2023-02-07", "categories": []}
{"id": "2302.00731", "title": "Faster Convergence with Lexicase Selection in Tree-based Automated\n  Machine Learning", "abstract": "  In many evolutionary computation systems, parent selection methods can\naffect, among other things, convergence to a solution. In this paper, we\npresent a study comparing the role of two commonly used parent selection\nmethods in evolving machine learning pipelines in an automated machine learning\nsystem called Tree-based Pipeline Optimization Tool (TPOT). Specifically, we\ndemonstrate, using experiments on multiple datasets, that lexicase selection\nleads to significantly faster convergence as compared to NSGA-II in TPOT. We\nalso compare the exploration of parts of the search space by these selection\nmethods using a trie data structure that contains information about the\npipelines explored in a particular run.\n", "date": "2023-02-01", "categories": []}
{"id": "2301.11316", "title": "Open Problems in Applied Deep Learning", "abstract": "  This work formulates the machine learning mechanism as a bi-level\noptimization problem. The inner level optimization loop entails minimizing a\nproperly chosen loss function evaluated on the training data. This is nothing\nbut the well-studied training process in pursuit of optimal model parameters.\nThe outer level optimization loop is less well-studied and involves maximizing\na properly chosen performance metric evaluated on the validation data. This is\nwhat we call the \"iteration process\", pursuing optimal model hyper-parameters.\nAmong many other degrees of freedom, this process entails model engineering\n(e.g., neural network architecture design) and management, experiment tracking,\ndataset versioning and augmentation. The iteration process could be automated\nvia Automatic Machine Learning (AutoML) or left to the intuitions of machine\nlearning students, engineers, and researchers. Regardless of the route we take,\nthere is a need to reduce the computational cost of the iteration step and as a\ndirect consequence reduce the carbon footprint of developing artificial\nintelligence algorithms. Despite the clean and unified mathematical formulation\nof the iteration step as a bi-level optimization problem, its solutions are\ncase specific and complex. This work will consider such cases while increasing\nthe level of complexity from supervised learning to semi-supervised,\nself-supervised, unsupervised, few-shot, federated, reinforcement, and\nphysics-informed learning. As a consequence of this exercise, this proposal\nsurfaces a plethora of open problems in the field, many of which can be\naddressed in parallel.\n", "date": "2023-01-26", "categories": []}
{"id": "2301.08727", "title": "Neural Architecture Search: Insights from 1000 Papers", "abstract": "  In the past decade, advances in deep learning have resulted in breakthroughs\nin a variety of areas, including computer vision, natural language\nunderstanding, speech recognition, and reinforcement learning. Specialized,\nhigh-performing neural architectures are crucial to the success of deep\nlearning in these areas. Neural architecture search (NAS), the process of\nautomating the design of neural architectures for a given task, is an\ninevitable next step in automating machine learning and has already outpaced\nthe best human-designed architectures on many tasks. In the past few years,\nresearch in NAS has been progressing rapidly, with over 1000 papers released\nsince 2020 (Deng and Lindauer, 2021). In this survey, we provide an organized\nand comprehensive guide to neural architecture search. We give a taxonomy of\nsearch spaces, algorithms, and speedup techniques, and we discuss resources\nsuch as benchmarks, best practices, other surveys, and open-source libraries.\n", "date": "2023-01-20", "categories": []}
{"id": "2301.10443", "title": "Learning to Rank Normalized Entropy Curves with Differentiable Window\n  Transformation", "abstract": "  Recent automated machine learning systems often use learning curves ranking\nmodels to inform decisions about when to stop unpromising trials and identify\nbetter model configurations. In this paper, we present a novel learning curve\nranking model specifically tailored for ranking normalized entropy (NE)\nlearning curves, which are commonly used in online advertising and\nrecommendation systems. Our proposed model, self-Adaptive Curve Transformation\naugmented Relative curve Ranking (ACTR2), features an adaptive curve\ntransformation layer that transforms raw lifetime NE curves into composite\nwindow NE curves with the window sizes adaptively optimized based on both the\nposition on the learning curve and the curve's dynamics. We also introduce a\nnovel differentiable indexing method for the proposed adaptive curve\ntransformation, which allows gradients with respect to the discrete indices to\nflow freely through the curve transformation layer, enabling the learned window\nsizes to be updated flexibly during training. Additionally, we propose a\npairwise curve ranking architecture that directly models the difference between\nthe two learning curves and is better at capturing subtle changes in relative\nperformance that may not be evident when modeling each curve individually as\nthe existing approaches did. Our extensive experiments on a real-world NE curve\ndataset demonstrate the effectiveness of each key component of ACTR2 and its\nimproved performance over the state-of-the-art.\n", "date": "2023-01-25", "categories": []}
{"id": "2301.05102", "title": "Improvement of Computational Performance of Evolutionary AutoML in a\n  Heterogeneous Environment", "abstract": "  Resource-intensive computations are a major factor that limits the\neffectiveness of automated machine learning solutions. In the paper, we propose\na modular approach that can be used to increase the quality of evolutionary\noptimization for modelling pipelines with a graph-based structure. It consists\nof several stages - parallelization, caching and evaluation. Heterogeneous and\nremote resources can be involved in the evaluation stage. The conducted\nexperiments confirm the correctness and effectiveness of the proposed approach.\nThe implemented algorithms are available as a part of the open-source framework\nFEDOT.\n", "date": "2023-01-12", "categories": []}
{"id": "2212.13402", "title": "Traceable Automatic Feature Transformation via Cascading Actor-Critic\n  Agents", "abstract": "  Feature transformation for AI is an essential task to boost the effectiveness\nand interpretability of machine learning (ML). Feature transformation aims to\ntransform original data to identify an optimal feature space that enhances the\nperformances of a downstream ML model. Existing studies either combines\npreprocessing, feature selection, and generation skills to empirically\ntransform data, or automate feature transformation by machine intelligence,\nsuch as reinforcement learning. However, existing studies suffer from: 1)\nhigh-dimensional non-discriminative feature space; 2) inability to represent\ncomplex situational states; 3) inefficiency in integrating local and global\nfeature information. To fill the research gap, we formulate the feature\ntransformation task as an iterative, nested process of feature generation and\nselection, where feature generation is to generate and add new features based\non original features, and feature selection is to remove redundant features to\ncontrol the size of feature space. Finally, we present extensive experiments\nand case studies to illustrate 24.7\\% improvements in F1 scores compared with\nSOTAs and robustness in high-dimensional data.\n", "date": "2022-12-27", "categories": []}
{"id": "2212.13152", "title": "Toward Efficient Automated Feature Engineering", "abstract": "  Automated Feature Engineering (AFE) refers to automatically generate and\nselect optimal feature sets for downstream tasks, which has achieved great\nsuccess in real-world applications. Current AFE methods mainly focus on\nimproving the effectiveness of the produced features, but ignoring the\nlow-efficiency issue for large-scale deployment. Therefore, in this work, we\npropose a generic framework to improve the efficiency of AFE. Specifically, we\nconstruct the AFE pipeline based on reinforcement learning setting, where each\nfeature is assigned an agent to perform feature transformation \\com{and}\nselection, and the evaluation score of the produced features in downstream\ntasks serve as the reward to update the policy. We improve the efficiency of\nAFE in two perspectives. On the one hand, we develop a Feature Pre-Evaluation\n(FPE) Model to reduce the sample size and feature size that are two main\nfactors on undermining the efficiency of feature evaluation. On the other hand,\nwe devise a two-stage policy training strategy by running FPE on the\npre-evaluation task as the initialization of the policy to avoid training\npolicy from scratch. We conduct comprehensive experiments on 36 datasets in\nterms of both classification and regression tasks. The results show $2.9\\%$\nhigher performance in average and 2x higher computational efficiency comparing\nto state-of-the-art AFE methods.\n", "date": "2022-12-26", "categories": []}
{"id": "2212.10671", "title": "evoML Yellow Paper: Evolutionary AI and Optimisation Studio", "abstract": "  Machine learning model development and optimisation can be a rather\ncumbersome and resource-intensive process. Custom models are often more\ndifficult to build and deploy, and they require infrastructure and expertise\nwhich are often costly to acquire and maintain. Machine learning product\ndevelopment lifecycle must take into account the need to navigate the\ndifficulties of developing and deploying machine learning models. evoML is an\nAI-powered tool that provides automated functionalities in machine learning\nmodel development, optimisation, and model code optimisation. Core\nfunctionalities of evoML include data cleaning, exploratory analysis, feature\nanalysis and generation, model optimisation, model evaluation, model code\noptimisation, and model deployment. Additionally, a key feature of evoML is\nthat it embeds code and model optimisation into the model development process,\nand includes multi-objective optimisation capabilities.\n", "date": "2022-12-20", "categories": []}
{"id": "2212.06735", "title": "POPNASv3: a Pareto-Optimal Neural Architecture Search Solution for Image\n  and Time Series Classification", "abstract": "  The automated machine learning (AutoML) field has become increasingly\nrelevant in recent years. These algorithms can develop models without the need\nfor expert knowledge, facilitating the application of machine learning\ntechniques in the industry. Neural Architecture Search (NAS) exploits deep\nlearning techniques to autonomously produce neural network architectures whose\nresults rival the state-of-the-art models hand-crafted by AI experts. However,\nthis approach requires significant computational resources and hardware\ninvestments, making it less appealing for real-usage applications. This article\npresents the third version of Pareto-Optimal Progressive Neural Architecture\nSearch (POPNASv3), a new sequential model-based optimization NAS algorithm\ntargeting different hardware environments and multiple classification tasks.\nOur method is able to find competitive architectures within large search\nspaces, while keeping a flexible structure and data processing pipeline to\nadapt to different tasks. The algorithm employs Pareto optimality to reduce the\nnumber of architectures sampled during the search, drastically improving the\ntime efficiency without loss in accuracy. The experiments performed on images\nand time series classification datasets provide evidence that POPNASv3 can\nexplore a large set of assorted operators and converge to optimal architectures\nsuited for the type of data provided under different scenarios.\n", "date": "2022-12-13", "categories": []}
{"id": "2212.04058", "title": "AutoPINN: When AutoML Meets Physics-Informed Neural Networks", "abstract": "  Physics-Informed Neural Networks (PINNs) have recently been proposed to solve\nscientific and engineering problems, where physical laws are introduced into\nneural networks as prior knowledge. With the embedded physical laws, PINNs\nenable the estimation of critical parameters, which are unobservable via\nphysical tools, through observable variables. For example, Power Electronic\nConverters (PECs) are essential building blocks for the green energy\ntransition. PINNs have been applied to estimate the capacitance, which is\nunobservable during PEC operations, using current and voltage, which can be\nobserved easily during operations. The estimated capacitance facilitates\nself-diagnostics of PECs. Existing PINNs are often manually designed, which is\ntime-consuming and may lead to suboptimal performance due to a large number of\ndesign choices for neural network architectures and hyperparameters. In\naddition, PINNs are often deployed on different physical devices, e.g., PECs,\nwith limited and varying resources. Therefore, it requires designing different\nPINN models under different resource constraints, making it an even more\nchallenging task for manual design. To contend with the challenges, we propose\nAutomated Physics-Informed Neural Networks (AutoPINN), a framework that enables\nthe automated design of PINNs by combining AutoML and PINNs. Specifically, we\nfirst tailor a search space that allows finding high-accuracy PINNs for PEC\ninternal parameter estimation. We then propose a resource-aware search strategy\nto explore the search space to find the best PINN model under different\nresource constraints. We experimentally demonstrate that AutoPINN is able to\nfind more accurate PINN models than human-designed, state-of-the-art PINN\nmodels using fewer resources.\n", "date": "2022-12-08", "categories": []}
{"id": "2211.10921", "title": "MEESO: A Multi-objective End-to-End Self-Optimized Approach for\n  Automatically Building Deep Learning Models", "abstract": "  Deep learning has been widely used in various applications from different\nfields such as computer vision, natural language processing, etc. However, the\ntraining models are often manually developed via many costly experiments. This\nmanual work usually requires substantial computing resources, time, and\nexperience. To simplify the use of deep learning and alleviate human effort,\nautomated deep learning has emerged as a potential tool that releases the\nburden for both users and researchers. Generally, an automatic approach should\nsupport the diversity of model selection and the evaluation should allow users\nto decide upon their demands. To that end, we propose a multi-objective\nend-to-end self-optimized approach for constructing deep learning models\nautomatically. Experimental results on well-known datasets such as MNIST,\nFashion, and Cifar10 show that our algorithm can discover various competitive\nmodels compared with the state-of-the-art approach. In addition, our approach\nalso introduces multi-objective trade-off solutions for both accuracy and\nuncertainty metrics for users to make better decisions.\n", "date": "2022-11-20", "categories": []}
{"id": "2210.04971", "title": "Multi-step Planning for Automated Hyperparameter Optimization with\n  OptFormer", "abstract": "  As machine learning permeates more industries and models become more\nexpensive and time consuming to train, the need for efficient automated\nhyperparameter optimization (HPO) has never been more pressing. Multi-step\nplanning based approaches to hyperparameter optimization promise improved\nefficiency over myopic alternatives by more effectively balancing out\nexploration and exploitation. However, the potential of these approaches has\nnot been fully realized due to their technical complexity and computational\nintensity. In this work, we leverage recent advances in Transformer-based,\nnatural-language-interfaced hyperparameter optimization to circumvent these\nbarriers. We build on top of the recently proposed OptFormer which casts both\nhyperparameter suggestion and target function approximation as autoregressive\ngeneration thus making planning via rollouts simple and efficient. We conduct\nextensive exploration of different strategies for performing multi-step\nplanning on top of the OptFormer model to highlight its potential for use in\nconstructing non-myopic HPO strategies.\n", "date": "2022-10-10", "categories": []}
{"id": "2211.04148", "title": "The Technological Emergence of AutoML: A Survey of Performant Software\n  and Applications in the Context of Industry", "abstract": "  With most technical fields, there exists a delay between fundamental academic\nresearch and practical industrial uptake. Whilst some sciences have robust and\nwell-established processes for commercialisation, such as the pharmaceutical\npractice of regimented drug trials, other fields face transitory periods in\nwhich fundamental academic advancements diffuse gradually into the space of\ncommerce and industry. For the still relatively young field of\nAutomated/Autonomous Machine Learning (AutoML/AutonoML), that transitory period\nis under way, spurred on by a burgeoning interest from broader society. Yet, to\ndate, little research has been undertaken to assess the current state of this\ndissemination and its uptake. Thus, this review makes two primary contributions\nto knowledge around this topic. Firstly, it provides the most up-to-date and\ncomprehensive survey of existing AutoML tools, both open-source and commercial.\nSecondly, it motivates and outlines a framework for assessing whether an AutoML\nsolution designed for real-world application is 'performant'; this framework\nextends beyond the limitations of typical academic criteria, considering a\nvariety of stakeholder needs and the human-computer interactions required to\nservice them. Thus, additionally supported by an extensive assessment and\ncomparison of academic and commercial case-studies, this review evaluates\nmainstream engagement with AutoML in the early 2020s, identifying obstacles and\nopportunities for accelerating future uptake.\n", "date": "2022-11-08", "categories": []}
{"id": "2211.04362", "title": "Hyperparameter optimization in deep multi-target prediction", "abstract": "  As a result of the ever increasing complexity of configuring and fine-tuning\nmachine learning models, the field of automated machine learning (AutoML) has\nemerged over the past decade. However, software implementations like Auto-WEKA\nand Auto-sklearn typically focus on classical machine learning (ML) tasks such\nas classification and regression. Our work can be seen as the first attempt at\noffering a single AutoML framework for most problem settings that fall under\nthe umbrella of multi-target prediction, which includes popular ML settings\nsuch as multi-label classification, multivariate regression, multi-task\nlearning, dyadic prediction, matrix completion, and zero-shot learning.\nAutomated problem selection and model configuration are achieved by extending\nDeepMTP, a general deep learning framework for MTP problem settings, with\npopular hyperparameter optimization (HPO) methods. Our extensive benchmarking\nacross different datasets and MTP problem settings identifies cases where\nspecific HPO methods outperform others.\n", "date": "2022-11-08", "categories": []}
{"id": "2211.00376", "title": "Automated Imbalanced Learning", "abstract": "  Automated Machine Learning has grown very successful in automating the\ntime-consuming, iterative tasks of machine learning model development. However,\ncurrent methods struggle when the data is imbalanced. Since many real-world\ndatasets are naturally imbalanced, and improper handling of this issue can lead\nto quite useless models, this issue should be handled carefully. This paper\nfirst introduces a new benchmark to study how different AutoML methods are\naffected by label imbalance. Second, we propose strategies to better deal with\nimbalance and integrate them into an existing AutoML framework. Finally, we\npresent a systematic study which evaluates the impact of these strategies and\nfind that their inclusion in AutoML systems significantly increases their\nrobustness against label imbalance.\n", "date": "2022-11-01", "categories": []}
{"id": "2210.14687", "title": "Which is the best model for my data?", "abstract": "  In this paper, we tackle the problem of selecting the optimal model for a\ngiven structured pattern classification dataset. In this context, a model can\nbe understood as a classifier and a hyperparameter configuration. The proposed\nmeta-learning approach purely relies on machine learning and involves four\nmajor steps. Firstly, we present a concise collection of 62 meta-features that\naddress the problem of information cancellation when aggregation measure values\ninvolving positive and negative measurements. Secondly, we describe two\ndifferent approaches for synthetic data generation intending to enlarge the\ntraining data. Thirdly, we fit a set of pre-defined classification models for\neach classification problem while optimizing their hyperparameters using grid\nsearch. The goal is to create a meta-dataset such that each row denotes a\nmultilabel instance describing a specific problem. The features of these\nmeta-instances denote the statistical properties of the generated datasets,\nwhile the labels encode the grid search results as binary vectors such that\nbest-performing models are positively labeled. Finally, we tackle the model\nselection problem with several multilabel classifiers, including a\nConvolutional Neural Network designed to handle tabular data. The simulation\nresults show that our meta-learning approach can correctly predict an optimal\nmodel for 91% of the synthetic datasets and for 87% of the real-world datasets.\nFurthermore, we noticed that most meta-classifiers produced better results when\nusing our meta-features. Overall, our proposal differs from other meta-learning\napproaches since it tackles the algorithm selection and hyperparameter tuning\nproblems in a single step. Toward the end, we perform a feature importance\nanalysis to determine which statistical features drive the model selection\nmechanism.\n", "date": "2022-10-26", "categories": []}
{"id": "2210.11836", "title": "Structural Kernel Search via Bayesian Optimization and Symbolical\n  Optimal Transport", "abstract": "  Despite recent advances in automated machine learning, model selection is\nstill a complex and computationally intensive process. For Gaussian processes\n(GPs), selecting the kernel is a crucial task, often done manually by the\nexpert. Additionally, evaluating the model selection criteria for Gaussian\nprocesses typically scales cubically in the sample size, rendering kernel\nsearch particularly computationally expensive. We propose a novel, efficient\nsearch method through a general, structured kernel space. Previous methods\nsolved this task via Bayesian optimization and relied on measuring the distance\nbetween GP's directly in function space to construct a kernel-kernel. We\npresent an alternative approach by defining a kernel-kernel over the symbolic\nrepresentation of the statistical hypothesis that is associated with a kernel.\nWe empirically show that this leads to a computationally more efficient way of\nsearching through a discrete kernel space.\n", "date": "2022-10-21", "categories": []}
{"id": "2210.09084", "title": "Multi-Agent Automated Machine Learning", "abstract": "  In this paper, we propose multi-agent automated machine learning (MA2ML) with\nthe aim to effectively handle joint optimization of modules in automated\nmachine learning (AutoML). MA2ML takes each machine learning module, such as\ndata augmentation (AUG), neural architecture search (NAS), or hyper-parameters\n(HPO), as an agent and the final performance as the reward, to formulate a\nmulti-agent reinforcement learning problem. MA2ML explicitly assigns credit to\neach agent according to its marginal contribution to enhance cooperation among\nmodules, and incorporates off-policy learning to improve search efficiency.\nTheoretically, MA2ML guarantees monotonic improvement of joint optimization.\nExtensive experiments show that MA2ML yields the state-of-the-art top-1\naccuracy on ImageNet under constraints of computational cost, e.g.,\n$79.7\\%/80.5\\%$ with FLOPs fewer than 600M/800M. Extensive ablation studies\nverify the benefits of credit assignment and off-policy learning of MA2ML.\n", "date": "2022-10-17", "categories": []}
{"id": "2210.06835", "title": "Multi-agent Dynamic Algorithm Configuration", "abstract": "  Automated algorithm configuration relieves users from tedious,\ntrial-and-error tuning tasks. A popular algorithm configuration tuning paradigm\nis dynamic algorithm configuration (DAC), in which an agent learns dynamic\nconfiguration policies across instances by reinforcement learning (RL).\nHowever, in many complex algorithms, there may exist different types of\nconfiguration hyperparameters, and such heterogeneity may bring difficulties\nfor classic DAC which uses a single-agent RL policy. In this paper, we aim to\naddress this issue and propose multi-agent DAC (MA-DAC), with one agent working\nfor one type of configuration hyperparameter. MA-DAC formulates the dynamic\nconfiguration of a complex algorithm with multiple types of hyperparameters as\na contextual multi-agent Markov decision process and solves it by a cooperative\nmulti-agent RL (MARL) algorithm. To instantiate, we apply MA-DAC to a\nwell-known optimization algorithm for multi-objective optimization problems.\nExperimental results show the effectiveness of MA-DAC in not only achieving\nsuperior performance compared with other configuration tuning approaches based\non heuristic rules, multi-armed bandits, and single-agent RL, but also being\ncapable of generalizing to different problem classes. Furthermore, we release\nthe environments in this paper as a benchmark for testing MARL algorithms, with\nthe hope of facilitating the application of MARL.\n", "date": "2022-10-13", "categories": []}
{"id": "2210.04728", "title": "PyHopper -- Hyperparameter optimization", "abstract": "  Hyperparameter tuning is a fundamental aspect of machine learning research.\nSetting up the infrastructure for systematic optimization of hyperparameters\ncan take a significant amount of time. Here, we present PyHopper, a black-box\noptimization platform designed to streamline the hyperparameter tuning workflow\nof machine learning researchers. PyHopper's goal is to integrate with existing\ncode with minimal effort and run the optimization process with minimal\nnecessary manual oversight. With simplicity as the primary theme, PyHopper is\npowered by a single robust Markov-chain Monte-Carlo optimization algorithm that\nscales to millions of dimensions. Compared to existing tuning packages,\nfocusing on a single algorithm frees the user from having to decide between\nseveral algorithms and makes PyHopper easily customizable. PyHopper is publicly\navailable under the Apache-2.0 license at https://github.com/PyHopper/PyHopper.\n", "date": "2022-10-10", "categories": []}
{"id": "2210.03324", "title": "AutoML for Climate Change: A Call to Action", "abstract": "  The challenge that climate change poses to humanity has spurred a rapidly\ndeveloping field of artificial intelligence research focused on climate change\napplications. The climate change AI (CCAI) community works on a diverse,\nchallenging set of problems which often involve physics-constrained ML or\nheterogeneous spatiotemporal data. It would be desirable to use automated\nmachine learning (AutoML) techniques to automatically find high-performing\narchitectures and hyperparameters for a given dataset. In this work, we\nbenchmark popular AutoML libraries on three high-leverage CCAI applications:\nclimate modeling, wind power forecasting, and catalyst discovery. We find that\nout-of-the-box AutoML libraries currently fail to meaningfully surpass the\nperformance of human-designed CCAI models. However, we also identify a few key\nweaknesses, which stem from the fact that most AutoML techniques are tailored\nto computer vision and NLP applications. For example, while dozens of search\nspaces have been designed for image and language data, none have been designed\nfor spatiotemporal data. Addressing these key weaknesses can lead to the\ndiscovery of novel architectures that yield substantial performance gains\nacross numerous CCAI applications. Therefore, we present a call to action to\nthe AutoML community, since there are a number of concrete, promising\ndirections for future work in the space of AutoML for CCAI. We release our code\nand a list of resources at\nhttps://github.com/climate-change-automl/climate-change-automl.\n", "date": "2022-10-07", "categories": []}
{"id": "2209.14598", "title": "Dynamic Surrogate Switching: Sample-Efficient Search for Factorization\n  Machine Configurations in Online Recommendations", "abstract": "  Hyperparameter optimization is the process of identifying the appropriate\nhyperparameter configuration of a given machine learning model with regard to a\ngiven learning task. For smaller data sets, an exhaustive search is possible;\nHowever, when the data size and model complexity increase, the number of\nconfiguration evaluations becomes the main computational bottleneck. A\npromising paradigm for tackling this type of problem is surrogate-based\noptimization. The main idea underlying this paradigm considers an incrementally\nupdated model of the relation between the hyperparameter space and the output\n(target) space; the data for this model are obtained by evaluating the main\nlearning engine, which is, for example, a factorization machine-based model. By\nlearning to approximate the hyperparameter-target relation, the surrogate\n(machine learning) model can be used to score large amounts of hyperparameter\nconfigurations, exploring parts of the configuration space beyond the reach of\ndirect machine learning engine evaluation. Commonly, a surrogate is selected\nprior to optimization initialization and remains the same during the search. We\ninvestigated whether dynamic switching of surrogates during the optimization\nitself is a sensible idea of practical relevance for selecting the most\nappropriate factorization machine-based models for large-scale online\nrecommendation. We conducted benchmarks on data sets containing hundreds of\nmillions of instances against established baselines such as Random Forest- and\nGaussian process-based surrogates. The results indicate that surrogate\nswitching can offer good performance while considering fewer learning engine\nevaluations.\n", "date": "2022-09-29", "categories": []}
{"id": "2209.13575", "title": "Efficient Non-Parametric Optimizer Search for Diverse Tasks", "abstract": "  Efficient and automated design of optimizers plays a crucial role in\nfull-stack AutoML systems. However, prior methods in optimizer search are often\nlimited by their scalability, generability, or sample efficiency. With the goal\nof democratizing research and application of optimizer search, we present the\nfirst efficient, scalable and generalizable framework that can directly search\non the tasks of interest. We first observe that optimizer updates are\nfundamentally mathematical expressions applied to the gradient. Inspired by the\ninnate tree structure of the underlying math expressions, we re-arrange the\nspace of optimizers into a super-tree, where each path encodes an optimizer.\nThis way, optimizer search can be naturally formulated as a path-finding\nproblem, allowing a variety of well-established tree traversal methods to be\nused as the search algorithm. We adopt an adaptation of the Monte Carlo method\nto tree search, equipped with rejection sampling and equivalent-form detection\nthat leverage the characteristics of optimizer update rules to further boost\nthe sample efficiency. We provide a diverse set of tasks to benchmark our\nalgorithm and demonstrate that, with only 128 evaluations, the proposed\nframework can discover optimizers that surpass both human-designed counterparts\nand prior optimizer search methods.\n", "date": "2022-09-27", "categories": []}
{"id": "2209.08018", "title": "IoT Data Analytics in Dynamic Environments: From An Automated Machine\n  Learning Perspective", "abstract": "  With the wide spread of sensors and smart devices in recent years, the data\ngeneration speed of the Internet of Things (IoT) systems has increased\ndramatically. In IoT systems, massive volumes of data must be processed,\ntransformed, and analyzed on a frequent basis to enable various IoT services\nand functionalities. Machine Learning (ML) approaches have shown their capacity\nfor IoT data analytics. However, applying ML models to IoT data analytics tasks\nstill faces many difficulties and challenges, specifically, effective model\nselection, design/tuning, and updating, which have brought massive demand for\nexperienced data scientists. Additionally, the dynamic nature of IoT data may\nintroduce concept drift issues, causing model performance degradation. To\nreduce human efforts, Automated Machine Learning (AutoML) has become a popular\nfield that aims to automatically select, construct, tune, and update machine\nlearning models to achieve the best performance on specified tasks. In this\npaper, we conduct a review of existing methods in the model selection, tuning,\nand updating procedures in the area of AutoML in order to identify and\nsummarize the optimal solutions for every step of applying ML algorithms to IoT\ndata analytics. To justify our findings and help industrial users and\nresearchers better implement AutoML approaches, a case study of applying AutoML\nto IoT anomaly detection problems is conducted in this work. Lastly, we discuss\nand classify the challenges and research directions for this domain.\n", "date": "2022-09-16", "categories": []}
{"id": "2209.02991", "title": "Auto-TransRL: Autonomous Composition of Vision Pipelines for Robotic\n  Perception", "abstract": "  Creating a vision pipeline for different datasets to solve a computer vision\ntask is a complex and time consuming process. Currently, these pipelines are\ndeveloped with the help of domain experts. Moreover, there is no systematic\nstructure to construct a vision pipeline apart from relying on experience,\ntrial and error or using template-based approaches. As the search space for\nchoosing suitable algorithms for achieving a particular vision task is large,\nhuman exploration for finding a good solution requires time and effort. To\naddress the following issues, we propose a dynamic and data-driven way to\nidentify an appropriate set of algorithms that would be fit for building the\nvision pipeline in order to achieve the goal task. We introduce a Transformer\nArchitecture complemented with Deep Reinforcement Learning to recommend\nalgorithms that can be incorporated at different stages of the vision workflow.\nThis system is both robust and adaptive to dynamic changes in the environment.\nExperimental results further show that our method also generalizes well to\nrecommend algorithms that have not been used while training and hence\nalleviates the need of retraining the system on a new set of algorithms\nintroduced during test time.\n", "date": "2022-09-07", "categories": []}
{"id": "2207.06028", "title": "High Per Parameter: A Large-Scale Study of Hyperparameter Tuning for\n  Machine Learning Algorithms", "abstract": "  Hyperparameters in machine learning (ML) have received a fair amount of\nattention, and hyperparameter tuning has come to be regarded as an important\nstep in the ML pipeline. But just how useful is said tuning? While\nsmaller-scale experiments have been previously conducted, herein we carry out a\nlarge-scale investigation, specifically, one involving 26 ML algorithms, 250\ndatasets (regression and both binary and multinomial classification), 6 score\nmetrics, and 28,857,600 algorithm runs. Analyzing the results we conclude that\nfor many ML algorithms we should not expect considerable gains from\nhyperparameter tuning on average, however, there may be some datasets for which\ndefault hyperparameters perform poorly, this latter being truer for some\nalgorithms than others. By defining a single hp_score value, which combines an\nalgorithm's accumulated statistics, we are able to rank the 26 ML algorithms\nfrom those expected to gain the most from hyperparameter tuning to those\nexpected to gain the least. We believe such a study may serve ML practitioners\nat large.\n", "date": "2022-07-13", "categories": []}
{"id": "2208.14046", "title": "A Deep Neural Networks ensemble workflow from hyperparameter search to\n  inference leveraging GPU clusters", "abstract": "  Automated Machine Learning with ensembling (or AutoML with ensembling) seeks\nto automatically build ensembles of Deep Neural Networks (DNNs) to achieve\nqualitative predictions. Ensemble of DNNs are well known to avoid over-fitting\nbut they are memory and time consuming approaches. Therefore, an ideal AutoML\nwould produce in one single run time different ensembles regarding accuracy and\ninference speed. While previous works on AutoML focus to search for the best\nmodel to maximize its generalization ability, we rather propose a new AutoML to\nbuild a larger library of accurate and diverse individual models to then\nconstruct ensembles. First, our extensive benchmarks show asynchronous\nHyperband is an efficient and robust way to build a large number of diverse\nmodels to combine them. Then, a new ensemble selection method based on a\nmulti-objective greedy algorithm is proposed to generate accurate ensembles by\ncontrolling their computing cost. Finally, we propose a novel algorithm to\noptimize the inference of the DNNs ensemble in a GPU cluster based on\nallocation optimization. The produced AutoML with ensemble method shows robust\nresults on two datasets using efficiently GPU clusters during both the training\nphase and the inference phase.\n", "date": "2022-08-30", "categories": []}
{"id": "2208.13116", "title": "An Empirical Study on the Usage of Automated Machine Learning Tools", "abstract": "  The popularity of automated machine learning (AutoML) tools in different\ndomains has increased over the past few years. Machine learning (ML)\npractitioners use AutoML tools to automate and optimize the process of feature\nengineering, model training, and hyperparameter optimization and so on. Recent\nwork performed qualitative studies on practitioners' experiences of using\nAutoML tools and compared different AutoML tools based on their performance and\nprovided features, but none of the existing work studied the practices of using\nAutoML tools in real-world projects at a large scale. Therefore, we conducted\nan empirical study to understand how ML practitioners use AutoML tools in their\nprojects. To this end, we examined the top 10 most used AutoML tools and their\nrespective usages in a large number of open-source project repositories hosted\non GitHub. The results of our study show 1) which AutoML tools are mostly used\nby ML practitioners and 2) the characteristics of the repositories that use\nthese AutoML tools. Also, we identified the purpose of using AutoML tools (e.g.\nmodel parameter sampling, search space management, model\nevaluation/error-analysis, Data/ feature transformation, and data labeling) and\nthe stages of the ML pipeline (e.g. feature engineering) where AutoML tools are\nused. Finally, we report how often AutoML tools are used together in the same\nsource code files. We hope our results can help ML practitioners learn about\ndifferent AutoML tools and their usages, so that they can pick the right tool\nfor their purposes. Besides, AutoML tool developers can benefit from our\nfindings to gain insight into the usages of their tools and improve their tools\nto better fit the users' usages and needs.\n", "date": "2022-08-28", "categories": []}
{"id": "2208.11792", "title": "A Survey of Open Source Automation Tools for Data Science Predictions", "abstract": "  We present an expository overview of technical and cultural challenges to the\ndevelopment and adoption of automation at various stages in the data science\nprediction lifecycle, restricting focus to supervised learning with structured\ndatasets. In addition, we review popular open source Python tools implementing\ncommon solution patterns for the automation challenges and highlight gaps where\nwe feel progress still demands to be made.\n", "date": "2022-08-24", "categories": []}
{"id": "2207.07012", "title": "AutoML-Based Drought Forecast with Meteorological Variables", "abstract": "  A precise forecast for droughts is of considerable value to scientific\nresearch, agriculture, and water resource management. With emerging\ndevelopments of data-driven approaches for hydro-climate modeling, this paper\ninvestigates an AutoML-based framework to forecast droughts in the U.S.\nCompared with commonly-used temporal deep learning models, the AutoML model can\nachieve comparable performance with less training data and time. As deep\nlearning models are becoming popular for Earth system modeling, this paper aims\nto bring more efforts to AutoML-based methods, and the use of them as benchmark\nbaselines for more complex deep learning models.\n", "date": "2022-06-09", "categories": []}
{"id": "2208.10498", "title": "Design Automation for Fast, Lightweight, and Effective Deep Learning\n  Models: A Survey", "abstract": "  Deep learning technologies have demonstrated remarkable effectiveness in a\nwide range of tasks, and deep learning holds the potential to advance a\nmultitude of applications, including in edge computing, where deep models are\ndeployed on edge devices to enable instant data processing and response. A key\nchallenge is that while the application of deep models often incurs substantial\nmemory and computational costs, edge devices typically offer only very limited\nstorage and computational capabilities that may vary substantially across\ndevices. These characteristics make it difficult to build deep learning\nsolutions that unleash the potential of edge devices while complying with their\nconstraints. A promising approach to addressing this challenge is to automate\nthe design of effective deep learning models that are lightweight, require only\na little storage, and incur only low computational overheads. This survey\noffers comprehensive coverage of studies of design automation techniques for\ndeep learning models targeting edge computing. It offers an overview and\ncomparison of key metrics that are used commonly to quantify the proficiency of\nmodels in terms of effectiveness, lightness, and computational costs. The\nsurvey then proceeds to cover three categories of the state-of-the-art of deep\nmodel design automation techniques: automated neural architecture search,\nautomated model compression, and joint automated design and compression.\nFinally, the survey covers open issues and directions for future research.\n", "date": "2022-08-22", "categories": []}
{"id": "2208.04376", "title": "On Taking Advantage of Opportunistic Meta-knowledge to Reduce\n  Configuration Spaces for Automated Machine Learning", "abstract": "  The automated machine learning (AutoML) process can require searching through\ncomplex configuration spaces of not only machine learning (ML) components and\ntheir hyperparameters but also ways of composing them together, i.e. forming ML\npipelines. Optimisation efficiency and the model accuracy attainable for a\nfixed time budget suffer if this pipeline configuration space is excessively\nlarge. A key research question is whether it is both possible and practical to\npreemptively avoid costly evaluations of poorly performing ML pipelines by\nleveraging their historical performance for various ML tasks, i.e.\nmeta-knowledge. The previous experience comes in the form of\nclassifier/regressor accuracy rankings derived from either (1) a substantial\nbut non-exhaustive number of pipeline evaluations made during historical AutoML\nruns, i.e. 'opportunistic' meta-knowledge, or (2) comprehensive cross-validated\nevaluations of classifiers/regressors with default hyperparameters, i.e.\n'systematic' meta-knowledge. Numerous experiments with the AutoWeka4MCPS\npackage suggest that (1) opportunistic/systematic meta-knowledge can improve ML\noutcomes, typically in line with how relevant that meta-knowledge is, and (2)\nconfiguration-space culling is optimal when it is neither too conservative nor\ntoo radical. However, the utility and impact of meta-knowledge depend\ncritically on numerous facets of its generation and exploitation, warranting\nextensive analysis; these are often overlooked/underappreciated within AutoML\nand meta-learning literature. In particular, we observe strong sensitivity to\nthe `challenge' of a dataset, i.e. whether specificity in choosing a predictor\nleads to significantly better performance. Ultimately, identifying `difficult'\ndatasets, thus defined, is crucial to both generating informative\nmeta-knowledge bases and understanding optimal search-space reduction\nstrategies.\n", "date": "2022-08-08", "categories": []}
{"id": "2208.03320", "title": "HPO: We won't get fooled again", "abstract": "  Hyperparameter optimization (HPO) is a well-studied research field. However,\nthe effects and interactions of the components in an HPO pipeline are not yet\nwell investigated. Then, we ask ourselves: can the landscape of HPO be biased\nby the pipeline used to evaluate individual configurations? To address this\nquestion, we proposed to analyze the effect of the HPO pipeline on HPO problems\nusing fitness landscape analysis. Particularly, we studied the DS-2019 HPO\nbenchmark data set, looking for patterns that could indicate evaluation\npipeline malfunction, and relate them to HPO performance. Our main findings\nare: (i) In most instances, large groups of diverse hyperparameters (i.e.,\nmultiple configurations) yield the same ill performance, most likely associated\nwith majority class prediction models; (ii) in these cases, a worsened\ncorrelation between the observed fitness and average fitness in the\nneighborhood is observed, potentially making harder the deployment of\nlocal-search based HPO strategies. Finally, we concluded that the HPO pipeline\ndefinition might negatively affect the HPO landscape.\n", "date": "2022-08-04", "categories": []}
{"id": "2208.00220", "title": "HPO X ELA: Investigating Hyperparameter Optimization Landscapes by Means\n  of Exploratory Landscape Analysis", "abstract": "  Hyperparameter optimization (HPO) is a key component of machine learning\nmodels for achieving peak predictive performance. While numerous methods and\nalgorithms for HPO have been proposed over the last years, little progress has\nbeen made in illuminating and examining the actual structure of these black-box\noptimization problems. Exploratory landscape analysis (ELA) subsumes a set of\ntechniques that can be used to gain knowledge about properties of unknown\noptimization problems. In this paper, we evaluate the performance of five\ndifferent black-box optimizers on 30 HPO problems, which consist of two-,\nthree- and five-dimensional continuous search spaces of the XGBoost learner\ntrained on 10 different data sets. This is contrasted with the performance of\nthe same optimizers evaluated on 360 problem instances from the black-box\noptimization benchmark (BBOB). We then compute ELA features on the HPO and BBOB\nproblems and examine similarities and differences. A cluster analysis of the\nHPO and BBOB problems in ELA feature space allows us to identify how the HPO\nproblems compare to the BBOB problems on a structural meta-level. We identify a\nsubset of BBOB problems that are close to the HPO problems in ELA feature space\nand show that optimizer performance is comparably similar on these two sets of\nbenchmark problems. We highlight open challenges of ELA for HPO and discuss\npotential directions of future research and applications.\n", "date": "2022-07-30", "categories": []}
{"id": "2207.10351", "title": "Auto Machine Learning for Medical Image Analysis by Unifying the Search\n  on Data Augmentation and Neural Architecture", "abstract": "  Automated data augmentation, which aims at engineering augmentation policy\nautomatically, recently draw a growing research interest. Many previous\nauto-augmentation methods utilized a Density Matching strategy by evaluating\npolicies in terms of the test-time augmentation performance. In this paper, we\ntheoretically and empirically demonstrated the inconsistency between the train\nand validation set of small-scale medical image datasets, referred to as\nin-domain sampling bias. Next, we demonstrated that the in-domain sampling bias\nmight cause the inefficiency of Density Matching. To address the problem, an\nimproved augmentation search strategy, named Augmented Density Matching, was\nproposed by randomly sampling policies from a prior distribution for training.\nMoreover, an efficient automatical machine learning(AutoML) algorithm was\nproposed by unifying the search on data augmentation and neural architecture.\nExperimental results indicated that the proposed methods outperformed\nstate-of-the-art approaches on MedMNIST, a pioneering benchmark designed for\nAutoML in medical image analysis.\n", "date": "2022-07-21", "categories": []}
{"id": "2206.03130", "title": "Towards Meta-learned Algorithm Selection using Implicit Fidelity\n  Information", "abstract": "  Automatically selecting the best performing algorithm for a given dataset or\nranking multiple algorithms by their expected performance supports users in\ndeveloping new machine learning applications. Most approaches for this problem\nrely on pre-computed dataset meta-features and landmarking performances to\ncapture the salient topology of the datasets and those topologies that the\nalgorithms attend to. Landmarking usually exploits cheap algorithms not\nnecessarily in the pool of candidate algorithms to get inexpensive\napproximations of the topology. While somewhat indicative, hand-crafted dataset\nmeta-features and landmarks are likely insufficient descriptors, strongly\ndepending on the alignment of the topologies that the landmarks and the\ncandidate algorithms search for. We propose IMFAS, a method to exploit\nmulti-fidelity landmarking information directly from the candidate algorithms\nin the form of non-parametrically non-myopic meta-learned learning curves via\nLSTMs in a few-shot setting during testing. Using this mechanism, IMFAS jointly\nlearns the topology of the datasets and the inductive biases of the candidate\nalgorithms, without the need to expensively train them to convergence. Our\napproach produces informative landmarks, easily enriched by arbitrary\nmeta-features at a low computational cost, capable of producing the desired\nranking using cheaper fidelities. We additionally show that IMFAS is able to\nbeat Successive Halving with at most 50% of the fidelity sequence during test\ntime.\n", "date": "2022-06-07", "categories": []}
{"id": "2207.06216", "title": "Goal-Oriented Sensitivity Analysis of Hyperparameters in Deep Learning", "abstract": "  Tackling new machine learning problems with neural networks always means\noptimizing numerous hyperparameters that define their structure and strongly\nimpact their performances. In this work, we study the use of goal-oriented\nsensitivity analysis, based on the Hilbert-Schmidt Independence Criterion\n(HSIC), for hyperparameter analysis and optimization. Hyperparameters live in\nspaces that are often complex and awkward. They can be of different natures\n(categorical, discrete, boolean, continuous), interact, and have\ninter-dependencies. All this makes it non-trivial to perform classical\nsensitivity analysis. We alleviate these difficulties to obtain a robust\nanalysis index that is able to quantify hyperparameters' relative impact on a\nneural network's final error. This valuable tool allows us to better understand\nhyperparameters and to make hyperparameter optimization more interpretable. We\nillustrate the benefits of this knowledge in the context of hyperparameter\noptimization and derive an HSIC-based optimization algorithm that we apply on\nMNIST and Cifar, classical machine learning data sets, but also on the\napproximation of Runge function and Bateman equations solution, of interest for\nscientific machine learning. This method yields neural networks that are both\ncompetitive and cost-effective.\n", "date": "2022-07-13", "categories": []}
{"id": "2206.03493", "title": "DeepCAVE: An Interactive Analysis Tool for Automated Machine Learning", "abstract": "  Automated Machine Learning (AutoML) is used more than ever before to support\nusers in determining efficient hyperparameters, neural architectures, or even\nfull machine learning pipelines. However, users tend to mistrust the\noptimization process and its results due to a lack of transparency, making\nmanual tuning still widespread. We introduce DeepCAVE, an interactive framework\nto analyze and monitor state-of-the-art optimization procedures for AutoML\neasily and ad hoc. By aiming for full and accessible transparency, DeepCAVE\nbuilds a bridge between users and AutoML and contributes to establishing trust.\nOur framework's modular and easy-to-extend nature provides users with\nautomatically generated text, tables, and graphic visualizations. We show the\nvalue of DeepCAVE in an exemplary use-case of outlier detection, in which our\nframework makes it easy to identify problems, compare multiple runs and\ninterpret optimization processes. The package is freely available on GitHub\nhttps://github.com/automl/DeepCAVE.\n", "date": "2022-06-07", "categories": []}
{"id": "2207.04010", "title": "MACFE: A Meta-learning and Causality Based Feature Engineering Framework", "abstract": "  Feature engineering has become one of the most important steps to improve\nmodel prediction performance, and to produce quality datasets. However, this\nprocess requires non-trivial domain-knowledge which involves a time-consuming\nprocess. Thereby, automating such process has become an active area of research\nand of interest in industrial applications. In this paper, a novel method,\ncalled Meta-learning and Causality Based Feature Engineering (MACFE), is\nproposed; our method is based on the use of meta-learning, feature distribution\nencoding, and causality feature selection. In MACFE, meta-learning is used to\nfind the best transformations, then the search is accelerated by pre-selecting\n\"original\" features given their causal relevance. Experimental evaluations on\npopular classification datasets show that MACFE can improve the prediction\nperformance across eight classifiers, outperforms the current state-of-the-art\nmethods in average by at least 6.54%, and obtains an improvement of 2.71% over\nthe best previous works.\n", "date": "2022-07-08", "categories": []}
{"id": "2207.01269", "title": "DiffML: End-to-end Differentiable ML Pipelines", "abstract": "  In this paper, we present our vision of differentiable ML pipelines called\nDiffML to automate the construction of ML pipelines in an end-to-end fashion.\nThe idea is that DiffML allows to jointly train not just the ML model itself\nbut also the entire pipeline including data preprocessing steps, e.g., data\ncleaning, feature selection, etc. Our core idea is to formulate all pipeline\nsteps in a differentiable way such that the entire pipeline can be trained\nusing backpropagation. However, this is a non-trivial problem and opens up many\nnew research questions. To show the feasibility of this direction, we\ndemonstrate initial ideas and a general principle of how typical preprocessing\nsteps such as data cleaning, feature selection and dataset selection can be\nformulated as differentiable programs and jointly learned with the ML model.\nMoreover, we discuss a research roadmap and core challenges that have to be\nsystematically tackled to enable fully differentiable ML pipelines.\n", "date": "2022-07-04", "categories": []}
{"id": "2206.08476", "title": "Zero-Shot AutoML with Pretrained Models", "abstract": "  Given a new dataset D and a low compute budget, how should we choose a\npre-trained model to fine-tune to D, and set the fine-tuning hyperparameters\nwithout risking overfitting, particularly if D is small? Here, we extend\nautomated machine learning (AutoML) to best make these choices. Our\ndomain-independent meta-learning approach learns a zero-shot surrogate model\nwhich, at test time, allows to select the right deep learning (DL) pipeline\n(including the pre-trained model and fine-tuning hyperparameters) for a new\ndataset D given only trivial meta-features describing D such as image\nresolution or the number of classes. To train this zero-shot model, we collect\nperformance data for many DL pipelines on a large collection of datasets and\nmeta-train on this data to minimize a pairwise ranking objective. We evaluate\nour approach under the strict time limit of the vision track of the ChaLearn\nAutoDL challenge benchmark, clearly outperforming all challenge contenders.\n", "date": "2022-06-16", "categories": []}
{"id": "2206.09423", "title": "Efficient End-to-End AutoML via Scalable Search Space Decomposition", "abstract": "  End-to-end AutoML has attracted intensive interests from both academia and\nindustry which automatically searches for ML pipelines in a space induced by\nfeature engineering, algorithm/model selection, and hyper-parameter tuning.\nExisting AutoML systems, however, suffer from scalability issues when applying\nto application domains with large, high-dimensional search spaces. We present\nVolcanoML, a scalable and extensible framework that facilitates systematic\nexploration of large AutoML search spaces. VolcanoML introduces and implements\nbasic building blocks that decompose a large search space into smaller ones,\nand allows users to utilize these building blocks to compose an execution plan\nfor the AutoML problem at hand. VolcanoML further supports a Volcano-style\nexecution model -- akin to the one supported by modern database systems -- to\nexecute the plan constructed. Our evaluation demonstrates that, not only does\nVolcanoML raise the level of expressiveness for search space decomposition in\nAutoML, it also leads to actual findings of decomposition strategies that are\nsignificantly more efficient than the ones employed by state-of-the-art AutoML\nsystems such as auto-sklearn. This paper is the extended version of the initial\nVolcanoML paper appeared in VLDB 2021.\n", "date": "2022-06-19", "categories": []}
{"id": "2206.12002", "title": "STREAMLINE: A Simple, Transparent, End-To-End Automated Machine Learning\n  Pipeline Facilitating Data Analysis and Algorithm Comparison", "abstract": "  Machine learning (ML) offers powerful methods for detecting and modeling\nassociations often in data with large feature spaces and complex associations.\nMany useful tools/packages (e.g. scikit-learn) have been developed to make the\nvarious elements of data handling, processing, modeling, and interpretation\naccessible. However, it is not trivial for most investigators to assemble these\nelements into a rigorous, replicatable, unbiased, and effective data analysis\npipeline. Automated machine learning (AutoML) seeks to address these issues by\nsimplifying the process of ML analysis for all. Here, we introduce STREAMLINE,\na simple, transparent, end-to-end AutoML pipeline designed as a framework to\neasily conduct rigorous ML modeling and analysis (limited initially to binary\nclassification). STREAMLINE is specifically designed to compare performance\nbetween datasets, ML algorithms, and other AutoML tools. It is unique among\nother autoML tools by offering a fully transparent and consistent baseline of\ncomparison using a carefully designed series of pipeline elements including:\n(1) exploratory analysis, (2) basic data cleaning, (3) cross validation\npartitioning, (4) data scaling and imputation, (5) filter-based feature\nimportance estimation, (6) collective feature selection, (7) ML modeling with\n`Optuna' hyperparameter optimization across 15 established algorithms\n(including less well-known Genetic Programming and rule-based ML), (8)\nevaluation across 16 classification metrics, (9) model feature importance\nestimation, (10) statistical significance comparisons, and (11) automatically\nexporting all results, plots, a PDF summary report, and models that can be\neasily applied to replication data.\n", "date": "2022-06-23", "categories": []}
{"id": "2606.21641", "title": "When Is an LLM Worth It for Hyperparameter Optimization? A Budget-Matched Study on Tabular Data Finds the Warm-Start Is a Default Configuration, Not the Model", "abstract": "Large language models (LLMs) have been proposed as hyperparameter-optimization (HPO) advisors that \"warm-start\" search from prior knowledge, proposing strong configurations in very few evaluations. We test that claim under a budget-matched, multi-seed protocol on eight PMLB tabular benchmarks, comparing an LLM advisor (LLM-OptFlow) against four classical baselines (random search, Optuna-TPE, Gaussian-process Bayesian optimization, and successive halving) over one shared search space, with paired tests and bootstrap 95% CIs across 8 x 5 = 40 (task, seed) units. The finding is cautionary. The advisor's strong first point is not an LLM output at all: like prior LLM-HPO systems the loop is seeded with a fixed default configuration, evaluated before any model call, which alone reaches 88.7% mean best-CV, identical to within 0.01 pp across all seven advisor models tested. The LLM's own proposals add only +0.40 pp of cross-validation accuracy over that seed and nothing on held-out test (LLM-Default = -0.01 pp, p = 0.92). When the same seed is granted to classical search, the apparent lead collapses: against seeded random search it leads by +0.20 pp at 2 evaluations, is tied by 5, and is behind by 12 (-0.37 pp). Without the seed, classical search ties the advisor by 12 evaluations and beats it by 40 (+0.6 to +0.8 pp, p <= 1e-4). Two LLM-specific behaviors survive: a single-task exploration failure (vehicle), and a rule-based confidence filter that removes ~33% of wasted compute without changing accuracy. The recommendation is deflationary: on tabular HPO, seed classical search with a sensible default; an LLM advisor adds no measurable generalization benefit and is overtaken within a handful of evaluations. We release the harness and a script that reproduces every statistic.", "date": "2026-06-19", "categories": []}
{"id": "2606.23739", "title": "Systematic Exploration of 4-Expert Heterogeneous Mixture-of-Experts via Automated Pipeline Search", "abstract": "We present an automated large-scale search pipeline for heterogeneous 4-Expert Mixture-of-Experts (MoE4) architectures within the LEMUR neural network dataset ecosystem. Building on a hand-crafted heterogeneous MoE reference model, we replace manual design with a deterministic code-assembly generator that systematically combines base architecture families drawn from the LEMUR database into MoE4 ensembles, each governed by a convolutional gating network with temperature scaling, mixup augmentation, and cosine-annealed learning rate scheduling. Over a 28-day campaign on an NVIDIA RTX 4090, the pipeline generated 4,463 candidate models across 197 batches, of which 1,021 were evaluated successfully. A critical finding emerged from the campaign: due to alphabetical enumeration via itertools.combinations, the entire explored search space (4.8% of the theoretical 23,751 possible 4-family combinations) is anchored to a single family, AirNet. We characterise this coverage bias precisely, identify the root cause in the generator, and propose a stratified random sampling fix. Within the AirNet anchored scope, ShuffleNet and MobileNetV3 consistently co-produce the highest-accuracy ensembles (mean accuracy up to 0.632), while FractalNet and MNASNet are identified as low-yield families warranting exclusion in future campaigns. The pipeline, analysis artefacts, and corrected generator are released as part of the open-source NNGPT project at https://github.com/ABrain-One/nn-gpt", "date": "2026-06-21", "categories": []}
{"id": "2606.22731", "title": "Closed-loop Auto Research for Molecular Property Prediction: Discovering and Certifying Generalizable Improvements", "abstract": "Closed-loop Auto Research extends automated machine learning from fixed-dataset fitting to changing the research workflow, with language-model agents editing representations and model code and acquiring external evidence. Molecular property prediction spans many small endpoints. We ask whether this action space yields improvements generalizing beyond the validation signal selecting them.\n  We isolate three Auto Research axes, features, models, and external evidence, under a file-level ablation lock attributing each gain to one axis over a strong baseline. Across 36 endpoints in three benchmark suites we score each selected configuration once on a held-out test whose labels the search never read. A routed pipeline taking each endpoint's best validation axis reaches positive held-out gains of 0.013, 0.011, and 0.042, the transferable axis differing by suite, data on TDC, model on Polaris, feature and model on MoleculeNet. The largest model-search gain falls from 0.041 on validation to 0.003 on test, while curated data reaches 0.022 but negative 0.019 on test, two non-transfer signatures. Curated external data raises held-out CYP2C9-substrate performance by 0.17 and half-life by 0.08, admitted through a contamination filter rejecting same-source files overlapping 64 to 89 percent of test structures, necessary but not sufficient for transfer. A matched-trial automated machine learning control did not reproduce the agent's code-level model intervention, reaching 0.006 against 0.042, and the pipeline stays competitive with an 84M-parameter pretrained 3D model on the shared training split.\n  The experiments stay within molecular property prediction, but separating discovery from held-out certification is a domain-agnostic lesson for any closed-loop system optimising a proxy for a held-out quantity.", "date": "2026-06-22", "categories": []}
{"id": "2606.23208", "title": "Leveraging AutoML for Sustainable Deep Learning: A Multi-Objective HPO Approach on Deep Shift Neural Networks", "abstract": "Deep Learning (DL) has advanced various fields by extracting complex patterns from large datasets. However, the computational demands of DL models pose environmental and resource challenges. Deep Shift Neural Networks (DSNNs) present a solution by leveraging shift operations to reduce computational complexity at inference. Compared to common DNNs, DSNNs are still less well understood and less well optimized. By leveraging AutoML techniques, we provide valuable insights into the potential of DSNNs and how to design them in a better way. We focus on image classification, a core task in computer vision, especially in low-resource environments. Since we consider complementary objectives such as accuracy and energy consumption, we combine state-of-the-art multi-fidelity (MF) hyperparameter optimization (HPO) with multi-objective optimization to find a set of Pareto optimal trade-offs on how to design DSNNs. Our approach led to significantly better configurations of DSNNs regarding loss and emissions compared to default DSNNs. This includes simultaneously increasing performance by about 20% and reducing emissions, in some cases by more than 60%. Investigating the behavior of quantized networks in terms of both emissions and accuracy, our experiments reveal surprising model-specific trade-offs, yielding the greatest energy savings. For example, in contrast to common expectations, quantizing smaller portions of the network with low precision can be optimal with respect to energy consumption while retaining or improving performance. We corroborated these findings across multiple backbone architectures, highlighting important nuances in quantization strategies and offering an automated approach to balancing energy efficiency and model performance.", "date": "2026-06-22", "categories": []}
{"id": "2606.23299", "title": "GRIMIP: A General Framework for Instance-Specific Configuration of MIP Solvers Using LLMs", "abstract": "Configuring the hyperparameters of Mixed-integer programming (MIP) solvers is a high-dimensional, instance-dependent optimization problem where suboptimal settings can degrade solving time by orders of magnitude. Default configurations are often suboptimal, while traditional tuning methods either suffer from the ``cold-start'' problem and inefficient search or heavily rely on expert experience. This paper introduces \\textbf{GRIMIP} (\\textbf{\\underline{G}}eneral \\textbf{\\underline{R}}easoning for \\textbf{\\underline{I}}nstance-specific \\textbf{\\underline{MIP}} configuration), a novel hybrid intelligence framework that synergistically integrates the semantic reasoning capabilities of Large Language Models (LLMs) with the sample-efficient search of Bayesian Optimization (BO). GRIMIP enables the LLM to function as a complete probabilistic surrogate within the BO loop, significantly improving performance and reducing sampling and evaluation costs. On seven benchmarks including MIPLIB, GRIMIP achieves over 40\\% reduction in Primal-Dual Integral on hard instances, outperforming SMAC and other LLM-assisted BO methods. By granting LLMs sufficient autonomy, GRIMIP combines the expert-level reasoning of LLMs with the efficient search of BO, achieving state-of-the-art performance.", "date": "2026-06-22", "categories": []}
{"id": "2606.24679", "title": "FlowPipe: LLM-Enhanced Conditional Generative Flow Networks for Data Preparation Pipeline Construction", "abstract": "Data preparation pipelines improve data quality in machine learning by transforming raw tables into learning-ready data through sequential cleaning and feature transformation operators. However, automatically constructing such pipelines is computationally difficult because operator sequences are combinatorial and end-to-end evaluation is expensive. Existing state-of-the-art (SOTA) Multi-DQN methods still face three key limitations: decoupled value estimators weaken long-horizon credit assignment, dataset context is only weakly injected into the policy, and exploration is inefficient in a sparse search space with many invalid states. To address these issues, we propose FlowPipe, a unified framework that formulates pipeline synthesis as conditional probabilistic flow generation over a directed acyclic graph. FlowPipe uses Conditional Generative Flow Networks (C-GFlowNets) with a Trajectory Balance objective to connect terminal validation rewards with early pipeline decisions. It further introduces Deep Semantic Modulation through Feature-wise Linear Modulation (FiLM), allowing LLM-derived logical priors to condition the policy's internal activations according to dataset semantics. In addition, FlowPipe incorporates failure awareness into the flow objective to avoid invalid states and concentrate search on high-potential regions. Experiments on two benchmark suites with 74 real-world datasets show that FlowPipe outperforms SOTA baselines, improving accuracy by 11.96% on average and achieving 12.5x faster training convergence. Source code is available at https://github.com/KunyuNi/FlowPipe.", "date": "2026-06-23", "categories": []}
{"id": "2606.25601", "title": "Statistically Valid Hyperparameter Selection: From Tuning to Guarantees", "abstract": "Hyperparameter selection is a critical step in the deployment of modern artificial intelligence systems, given the need to tune degrees of freedom such as inference-time parameters, implementation-level settings, and thresholds driving decision rules. Despite its practical importance, hyperparameter selection is typically performed using best-effort empirical methods such as grid search or Bayesian optimization, which provide no formal statistical guarantees on reliability or safety.\n  This monograph presents a unified statistical framework for reliable hyperparameter selection, centered on the learn-then-test (LTT) paradigm, which formulates the problem as multiple hypothesis testing over a candidate set of hyperparameters. The framework enables the selection of hyperparameters that provably satisfy application-specific reliability requirements -- such as bounds on average risk, quantile risk, or information-theoretic constraints -- with explicit, finite-sample control of error probabilities. The supporting statistical machinery, namely p-values, e-values, and concentration inequalities, is developed from first principles in a dedicated appendix.", "date": "2026-06-24", "categories": []}
{"id": "2606.30911", "title": "Why Solve It Twice? Hierarchical Accumulation of Skills for Transfer-Efficient ML Engineering", "abstract": "ML engineering agents waste compute rediscovering known techniques because every competition is a cold start. We present HASTE, a hierarchical multi-agent system that organizes cross-competition knowledge into three scope tiers (global, domain, and competition-specific), each coupled to a matching agent level. An orchestrator coordinates domain specialists and promotes learning between tiers via LLM-driven abstraction. A controlled ablation provides evidence for scoped loading: holding a 159-skill inventory constant across 8 competitions, tiered loading achieves a 100% medal rate while flat loading reaches only 62.5%, the same medal rate as loading no skills, and consumes 2x the output tokens. On the full MLE-Bench Lite benchmark (22 Kaggle competitions), HASTE reaches a medal rate of 77.3% using Claude Sonnet 4.6 at 12h per competition; this is a single-seed campaign result, and multi-seed replication is the priority follow-up. In a cold-start run, the system begins with no accumulated skills. In warm-start runs, it reloads skills learned from earlier competitions, using only global and domain-level skills for transfer across competitions. Warm starts use 52% fewer refinement iterations, and the fraction of proposed changes kept by the agent rises from 42% at low inventory to 85% once 50+ skills are available. These results suggest that better knowledge organization can partly substitute for model strength and compute budget in ML-engineering agents.", "date": "2026-06-29", "categories": []}