DM-Diaz commited on
Commit
900bd65
·
verified ·
1 Parent(s): bcace9d

Upload config.json

Browse files
Files changed (1) hide show
  1. config.json +141 -0
config.json ADDED
@@ -0,0 +1,141 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "format_version": 1,
3
+ "repository_type": "fitted_voxelwise_encoding_models",
4
+ "artifact_name": "VEDB SimCLR ResNet-18 — NSD Voxelwise Encoding Models",
5
+ "framework": {
6
+ "feature_extractor": "pytorch",
7
+ "encoding_model_fitting": "pytorch",
8
+ "artifact_serialization": "numpy"
9
+ },
10
+ "paper": {
11
+ "title": "Eccentricity-Constrained CNN Training Reveals Adaptive Information Coding Around the Visual Field",
12
+ "authors": [
13
+ "Dylan M. Diaz",
14
+ "Margaret M. Henderson"
15
+ ],
16
+ "year": 2026,
17
+ "venue": "Proceedings of the 9th Conference on Cognitive Computational Neuroscience",
18
+ "doi": "10.32470/0416gfsq",
19
+ "arxiv": "2607.19316"
20
+ },
21
+ "upstream_models": {
22
+ "architecture": "resnet18",
23
+ "pretraining_objective": "simclr",
24
+ "pretraining_dataset": "Visual Experience Dataset (VEDB)",
25
+ "conditions": [
26
+ "Baseline",
27
+ "Fovea-Gaze",
28
+ "Periph",
29
+ "Periph-NF"
30
+ ],
31
+ "collection": "DM-Diaz/eccentricity-constrained-simclr-models-vedb"
32
+ },
33
+ "neural_dataset": {
34
+ "name": "Natural Scenes Dataset (NSD)",
35
+ "modality": "7T fMRI",
36
+ "subjects": [
37
+ "S1",
38
+ "S2",
39
+ "S3",
40
+ "S4",
41
+ "S5",
42
+ "S6",
43
+ "S7",
44
+ "S8"
45
+ ],
46
+ "held_out_evaluation_images": 1000,
47
+ "held_out_images_description": "NSD images shared across participants",
48
+ "raw_data_redistributed": false
49
+ },
50
+ "input_preprocessing": {
51
+ "input_resolution": [
52
+ 224,
53
+ 224
54
+ ],
55
+ "nsd_visual_field_transform_reapplied": false,
56
+ "rescale": "uint8 / 255.0",
57
+ "normalization": {
58
+ "name": "ImageNet",
59
+ "mean": [
60
+ 0.485,
61
+ 0.456,
62
+ 0.406
63
+ ],
64
+ "std": [
65
+ 0.229,
66
+ 0.224,
67
+ 0.225
68
+ ]
69
+ }
70
+ },
71
+ "feature_extraction": {
72
+ "layers": [
73
+ "conv1",
74
+ "layer1.1",
75
+ "layer2.1",
76
+ "layer3.1",
77
+ "layer4.1",
78
+ "avgpool"
79
+ ],
80
+ "convolutional_spatial_reduction": {
81
+ "method": "adaptive_average_pooling",
82
+ "target_pre_pca_features_per_layer": 5000
83
+ },
84
+ "pca": {
85
+ "components_per_layer": 200,
86
+ "fit_separately_by_subject": true,
87
+ "fit_separately_by_model_condition": true,
88
+ "fit_separately_by_layer": true,
89
+ "fit_scope": "full subject-specific feature matrix before encoding-model train/holdout partitioning"
90
+ },
91
+ "layer_features_concatenated": true
92
+ },
93
+ "encoding_model": {
94
+ "type": "voxelwise_ridge_regression",
95
+ "regularization": "L2",
96
+ "candidate_lambda_count": 20,
97
+ "lambda_selection": "nested_holdout",
98
+ "selection_scope": "independently_per_voxel",
99
+ "feature_normalization": {
100
+ "method": "z_score",
101
+ "statistics_fit_on": "training_plus_nested_holdout",
102
+ "final_held_out_evaluation_excluded": true
103
+ },
104
+ "intercept": {
105
+ "included": true,
106
+ "implementation": "column_of_ones_appended_to_feature_matrix",
107
+ "saved_weight_location": "final_row_of_weights"
108
+ },
109
+ "evaluation_metrics": [
110
+ "r2",
111
+ "corr"
112
+ ]
113
+ },
114
+ "artifact": {
115
+ "file_format": ".npy",
116
+ "serialization": "numpy_saved_python_dictionary",
117
+ "load_with_allow_pickle": true,
118
+ "fit_fields": [
119
+ "subject",
120
+ "model",
121
+ "features_file_list",
122
+ "lambdas",
123
+ "voxel_mask",
124
+ "voxel_index",
125
+ "voxel_nc",
126
+ "brain_nii_shape",
127
+ "weights",
128
+ "r2",
129
+ "corr",
130
+ "best_lambda_inds"
131
+ ],
132
+ "contains_fitted_voxelwise_weights": true,
133
+ "contains_held_out_metrics": true,
134
+ "contains_raw_nsd_stimuli": false,
135
+ "contains_raw_fmri_data": false,
136
+ "contains_fitted_pca_transforms": false,
137
+ "contains_feature_normalization_statistics": false,
138
+ "turnkey_new_image_to_voxel_prediction": false
139
+ },
140
+ "artifact_scope_note": "The released files are fitted encoding-model artifacts. Exact prediction for new images additionally requires the corresponding pretrained ResNet-18 checkpoint and the original feature-extraction, spatial-pooling, PCA, concatenation, and normalization procedures."
141
+ }