DM-Diaz commited on
Commit
e2d5888
Β·
verified Β·
1 Parent(s): db31445

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +97 -4
README.md CHANGED
@@ -7,6 +7,7 @@ tags:
7
  - fmri
8
  - encoding-model
9
  - ridge-regression
 
10
  - natural-scenes-dataset
11
  - simclr
12
  - resnet18
@@ -18,7 +19,7 @@ tags:
18
 
19
  # VEDB SimCLR ResNet-18 β€” NSD Voxelwise Encoding Models
20
 
21
- This repository contains **subject-specific voxelwise encoding-model fits** from:
22
 
23
  **Diaz, D. M., & Henderson, M. M. (2026). *Eccentricity-Constrained CNN Training Reveals Adaptive Information Coding Around the Visual Field.* Proceedings of the 9th Conference on Cognitive Computational Neuroscience.**
24
 
@@ -59,11 +60,12 @@ The four pretrained models were trained on complementary versions of the same so
59
  | Component | Status |
60
  | --- | --- |
61
  | Subject-specific encoding-model fits | Available |
 
62
  | Model card | Available |
63
  | Feature-extraction / model-fitting code | Forthcoming |
64
  | Analysis code | Forthcoming |
65
  | VEDB imagery | Full dataset not redistributed; available via [Databrary](https://www.databrary.org/volume/1612) |
66
- | NSD stimuli and fMRI data | Not redistributed; repository contains encoding-model fits only; see [NSD](https://www.naturalscenesdataset.org/) |
67
 
68
  The feature-extraction, model-fitting, and analysis code is currently being consolidated and documented and will be linked here when released.
69
 
@@ -74,9 +76,12 @@ baseline/
74
  fovea-gaze/
75
  periph/
76
  periph-nf/
 
 
 
77
  ```
78
 
79
- Each folder contains voxelwise encoding-model fits for **NSD subjects S1–S8**.
80
 
81
  Example:
82
 
@@ -182,6 +187,92 @@ These internal identifiers are preserved in fields such as `model`, `model1`, `m
182
 
183
  The `features_file_list`, `features_file_list1`, and `features_file_list2` fields also preserve the original analysis paths for provenance. These paths refer to the computing environment used for the original study and are not expected to resolve outside that environment.
184
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
185
  ## Loading a Fit
186
 
187
  ```python
@@ -205,7 +296,9 @@ voxel_nc = fit["voxel_nc"]
205
 
206
  ## Scope of the Released Artifacts
207
 
208
- The `.npy` files contain the fitted voxelwise encoding models, prediction metrics, ridge-penalty information, and voxel metadata used in the study.
 
 
209
 
210
  Reproducing voxel predictions for new images also requires the corresponding pretrained ResNet-18 checkpoint and the feature-extraction, spatial-pooling, PCA, concatenation, and normalization procedures used during fitting. These steps will be documented in the accompanying public code release.
211
 
 
7
  - fmri
8
  - encoding-model
9
  - ridge-regression
10
+ - variance-partitioning
11
  - natural-scenes-dataset
12
  - simclr
13
  - resnet18
 
19
 
20
  # VEDB SimCLR ResNet-18 β€” NSD Voxelwise Encoding Models
21
 
22
+ This repository contains **subject-specific voxelwise encoding-model fits and the variance-partitioning fits reported in the associated study** from:
23
 
24
  **Diaz, D. M., & Henderson, M. M. (2026). *Eccentricity-Constrained CNN Training Reveals Adaptive Information Coding Around the Visual Field.* Proceedings of the 9th Conference on Cognitive Computational Neuroscience.**
25
 
 
60
  | Component | Status |
61
  | --- | --- |
62
  | Subject-specific encoding-model fits | Available |
63
+ | Variance-partitioning fits reported in the paper | Available |
64
  | Model card | Available |
65
  | Feature-extraction / model-fitting code | Forthcoming |
66
  | Analysis code | Forthcoming |
67
  | VEDB imagery | Full dataset not redistributed; available via [Databrary](https://www.databrary.org/volume/1612) |
68
+ | NSD stimuli and fMRI data | Not redistributed; repository contains derived encoding-model and variance-partitioning fits only; see [NSD](https://www.naturalscenesdataset.org/) |
69
 
70
  The feature-extraction, model-fitting, and analysis code is currently being consolidated and documented and will be linked here when released.
71
 
 
76
  fovea-gaze/
77
  periph/
78
  periph-nf/
79
+ variance-partitioning/
80
+ β”œβ”€β”€ fovea-gaze-vs-periph/
81
+ └── periph-vs-periph-nf/
82
  ```
83
 
84
+ Each model-condition folder contains voxelwise encoding-model fits for **NSD subjects S1–S8**. The `variance-partitioning/` directory contains the two model comparisons reported in Figure 4C–D of the paper, with one fit for each of the same eight NSD participants.
85
 
86
  Example:
87
 
 
187
 
188
  The `features_file_list`, `features_file_list1`, and `features_file_list2` fields also preserve the original analysis paths for provenance. These paths refer to the computing environment used for the original study and are not expected to resolve outside that environment.
189
 
190
+ ## Variance-Partitioning Fits
191
+
192
+ The repository also contains the subject-specific variance-partitioning fits reported in the associated paper:
193
+
194
+ - **Fovea-Gaze vs. Periph** β€” Figure 4C
195
+ - **Periph vs. Periph-NF** β€” Figure 4D
196
+
197
+ For each comparison and NSD participant, three voxelwise encoding models were fit using (1) features from model A alone, (2) features from model B alone, and (3) the concatenated feature spaces from both models. Unique variance was computed from held-out voxelwise `RΒ²` using the equations reported in the paper:
198
+
199
+ $$
200
+ R^2_{\mathrm{unique},A}
201
+ =
202
+ R^2_{\mathrm{combined}}
203
+ -
204
+ R^2_{B\text{-only}}
205
+ $$
206
+
207
+ $$
208
+ R^2_{\mathrm{unique},B}
209
+ =
210
+ R^2_{\mathrm{combined}}
211
+ -
212
+ R^2_{A\text{-only}}
213
+ $$
214
+
215
+ Thus, the unique variance attributed to one model is the variance explained by the combined representation minus the variance explained by the other model alone.
216
+
217
+ Variance-partitioning files are stored under:
218
+
219
+ ```text
220
+ variance-partitioning/
221
+ β”œβ”€β”€ fovea-gaze-vs-periph/
222
+ β”‚ β”œβ”€β”€ NSD_S1_varpart_Fovea-Gaze_vs_Periph.npy
223
+ β”‚ β”œβ”€β”€ ...
224
+ β”‚ └── NSD_S8_varpart_Fovea-Gaze_vs_Periph.npy
225
+ └── periph-vs-periph-nf/
226
+ β”œβ”€β”€ NSD_S1_varpart_Periph_vs_Periph-NF.npy
227
+ β”œβ”€β”€ ...
228
+ └── NSD_S8_varpart_Periph_vs_Periph-NF.npy
229
+ ```
230
+
231
+ Each variance-partitioning file contains a Python dictionary with:
232
+
233
+ - `subject`
234
+ - `model1`
235
+ - `model2`
236
+ - `features_file_list1`
237
+ - `features_file_list2`
238
+ - `lambdas`
239
+ - `voxel_mask`
240
+ - `voxel_index`
241
+ - `voxel_nc`
242
+ - `brain_nii_shape`
243
+ - `weights_varpart`
244
+ - `r2_varpart`
245
+ - `corr_varpart`
246
+ - `best_lambda_inds_varpart`
247
+
248
+ The `weights_varpart`, `r2_varpart`, `corr_varpart`, and `best_lambda_inds_varpart` fields are dictionaries with entries for model A alone, model B alone, and the combined feature space. Their exact keys retain the original internal model identifiers listed above.
249
+
250
+ Because the six retained layers contribute 200 PCA components each, a single-model variance-partition fit contains **1,200 model features plus one intercept**, giving `weights_varpart` matrices with 1,201 rows for the model-only fits. The combined fit contains 2,400 model features plus one intercept, giving 2,401 rows.
251
+
252
+ Example:
253
+
254
+ ```python
255
+ import numpy as np
256
+
257
+ fit = np.load(
258
+ "variance-partitioning/fovea-gaze-vs-periph/"
259
+ "NSD_S1_varpart_Fovea-Gaze_vs_Periph.npy",
260
+ allow_pickle=True
261
+ ).item()
262
+
263
+ r2 = fit["r2_varpart"]
264
+
265
+ model_a = fit["model1"]
266
+ model_b = fit["model2"]
267
+
268
+ r2_a_only = r2[f"{model_a}-only"]
269
+ r2_b_only = r2[f"{model_b}-only"]
270
+ r2_combined = r2["combined"]
271
+
272
+ r2_unique_a = r2_combined - r2_b_only
273
+ r2_unique_b = r2_combined - r2_a_only
274
+ ```
275
+
276
  ## Loading a Fit
277
 
278
  ```python
 
296
 
297
  ## Scope of the Released Artifacts
298
 
299
+ The repository contains **32 subject-specific encoding-model fits** (4 model conditions Γ— 8 NSD participants) and **16 variance-partitioning fits** corresponding to the two variance-partition analyses reported in the paper (2 model comparisons Γ— 8 participants), for a total of **48 fitted `.npy` artifacts**.
300
+
301
+ The released files contain fitted voxelwise weights, held-out prediction metrics, ridge-penalty information, and voxel metadata used in the study. The variance-partitioning artifacts additionally contain the model-A-only, model-B-only, and combined fits used to compute unique variance.
302
 
303
  Reproducing voxel predictions for new images also requires the corresponding pretrained ResNet-18 checkpoint and the feature-extraction, spatial-pooling, PCA, concatenation, and normalization procedures used during fitting. These steps will be documented in the accompanying public code release.
304