Coverage: all plotted cells have seeds 0, 1 and 2.
Figures
Downloads
per_seed_results.csv and aggregate_seed_results.csv contain the plotted values. Existing per-task bootstrap confidence intervals are carried through in the per-seed CSV when present; the bar error bars are seed variability.
Aggregate Table
| Direction | Dataset | Metric | Mechanism | n | Direct | Grounded mean +/- SD | Delta mean +/- SD | Directionality |
|---|---|---|---|---|---|---|---|---|
| BGE -> RepTok | COCO-1000 | CLIP similarity | Dynamic + sigmoid keep | 3 | 0.28647 | 0.291675 +/- 0.000239 | +0.00520567 +/- 0.000239 | higher is better |
| BGE -> RepTok | COCO-1000 | CLIP similarity | Dynamic + softmax keep | 3 | 0.28647 | 0.290629 +/- 0.000851 | +0.00415929 +/- 0.000851 | higher is better |
| BGE -> RepTok | COCO-1000 | CLIP similarity | Pinned + sigmoid keep | 3 | 0.28647 | 0.290556 +/- 0.000201 | +0.0040868 +/- 0.000201 | higher is better |
| BGE -> RepTok | COCO-1000 | CLIP similarity | Pinned + softmax keep | 3 | 0.28647 | 0.289881 +/- 9.28e-05 | +0.00341102 +/- 9.28e-05 | higher is better |
| BGE -> RepTok | COCO-1000 | FID | Dynamic + sigmoid keep | 3 | 75.6146 | 69.2706 +/- 0.324 | -6.34405 +/- 0.324 | lower is better |
| BGE -> RepTok | COCO-1000 | FID | Dynamic + softmax keep | 3 | 75.6146 | 69.6307 +/- 0.802 | -5.98393 +/- 0.802 | lower is better |
| BGE -> RepTok | COCO-1000 | FID | Pinned + sigmoid keep | 3 | 75.6146 | 70.3902 +/- 0.464 | -5.22444 +/- 0.464 | lower is better |
| BGE -> RepTok | COCO-1000 | FID | Pinned + softmax keep | 3 | 75.6146 | 70.3137 +/- 0.855 | -5.30097 +/- 0.855 | lower is better |
| BGE -> RepTok | COCO-1000 | LPIPS | Dynamic + sigmoid keep | 3 | 0.697183 | 0.69602 +/- 0.000304 | -0.00116261 +/- 0.000304 | lower is better |
| BGE -> RepTok | COCO-1000 | LPIPS | Dynamic + softmax keep | 3 | 0.697183 | 0.695985 +/- 0.000241 | -0.00119813 +/- 0.000241 | lower is better |
| BGE -> RepTok | COCO-1000 | LPIPS | Pinned + sigmoid keep | 3 | 0.697183 | 0.696239 +/- 0.000229 | -0.000943979 +/- 0.000229 | lower is better |
| BGE -> RepTok | COCO-1000 | LPIPS | Pinned + softmax keep | 3 | 0.697183 | 0.695971 +/- 0.000247 | -0.00121226 +/- 0.000247 | lower is better |
| BGE -> RepTok | ImageNet-1000 | CLIP similarity | Dynamic + sigmoid keep | 3 | 0.324203 | 0.326046 +/- 0.00014 | +0.00184281 +/- 0.00014 | higher is better |
| BGE -> RepTok | ImageNet-1000 | CLIP similarity | Dynamic + softmax keep | 3 | 0.324203 | 0.325806 +/- 0.000292 | +0.00160271 +/- 0.000292 | higher is better |
| BGE -> RepTok | ImageNet-1000 | CLIP similarity | Pinned + sigmoid keep | 3 | 0.324203 | 0.326087 +/- 0.000423 | +0.00188404 +/- 0.000423 | higher is better |
| BGE -> RepTok | ImageNet-1000 | CLIP similarity | Pinned + softmax keep | 3 | 0.324203 | 0.325258 +/- 0.00035 | +0.00105523 +/- 0.00035 | higher is better |
| BGE -> RepTok | ImageNet-1000 | FID | Dynamic + sigmoid keep | 3 | 51.64 | 50.6868 +/- 0.125 | -0.953152 +/- 0.125 | lower is better |
| BGE -> RepTok | ImageNet-1000 | FID | Dynamic + softmax keep | 3 | 51.64 | 50.7774 +/- 0.16 | -0.862517 +/- 0.16 | lower is better |
| BGE -> RepTok | ImageNet-1000 | FID | Pinned + sigmoid keep | 3 | 51.64 | 50.7923 +/- 0.177 | -0.847673 +/- 0.177 | lower is better |
| BGE -> RepTok | ImageNet-1000 | FID | Pinned + softmax keep | 3 | 51.64 | 50.9739 +/- 0.204 | -0.666083 +/- 0.204 | lower is better |
| BGE -> RepTok | ImageNet-1000 | LPIPS | Dynamic + sigmoid keep | 3 | 0.647536 | 0.646692 +/- 0.000103 | -0.000843028 +/- 0.000103 | lower is better |
| BGE -> RepTok | ImageNet-1000 | LPIPS | Dynamic + softmax keep | 3 | 0.647536 | 0.647548 +/- 0.000159 | +1.2219e-05 +/- 0.000159 | lower is better |
| BGE -> RepTok | ImageNet-1000 | LPIPS | Pinned + sigmoid keep | 3 | 0.647536 | 0.647344 +/- 0.000104 | -0.00019151 +/- 0.000104 | lower is better |
| BGE -> RepTok | ImageNet-1000 | LPIPS | Pinned + softmax keep | 3 | 0.647536 | 0.647318 +/- 0.000157 | -0.000217537 +/- 0.000157 | lower is better |
| DINO -> symbolic | ImageNet-50k | Top-1 accuracy | Dynamic + sigmoid keep | 3 | 0.836 | 0.836027 +/- 6.43e-05 | +2.66433e-05 +/- 6.43e-05 | higher is better |
| DINO -> symbolic | ImageNet-50k | Top-1 accuracy | Dynamic + softmax keep | 3 | 0.836 | 0.799627 +/- 0.0413 | -0.0363734 +/- 0.0413 | higher is better |
| DINO -> symbolic | ImageNet-50k | Top-1 accuracy | Pinned + sigmoid keep | 3 | 0.836 | 0.836053 +/- 9.24e-05 | +5.33263e-05 +/- 9.24e-05 | higher is better |
| DINO -> symbolic | ImageNet-50k | Top-1 accuracy | Pinned + softmax keep | 3 | 0.836 | 0.81902 +/- 0.017 | -0.01698 +/- 0.017 | higher is better |
| RepTok -> SONAR | ImageNet-50k | Corpus BLEU-4 | Dynamic + sigmoid keep | 3 | 11.8117 | 12.3648 +/- 0.423 | +0.553028 +/- 0.423 | higher is better |
| RepTok -> SONAR | ImageNet-50k | Corpus BLEU-4 | Dynamic + softmax keep | 3 | 11.8117 | 11.5165 +/- 1.31 | -0.295276 +/- 1.31 | higher is better |
| RepTok -> SONAR | ImageNet-50k | Corpus BLEU-4 | Pinned + sigmoid keep | 3 | 11.8117 | 11.2626 +/- 0.0643 | -0.549174 +/- 0.0643 | higher is better |
| RepTok -> SONAR | ImageNet-50k | Corpus BLEU-4 | Pinned + softmax keep | 3 | 11.8117 | 10.0338 +/- 0.623 | -1.77798 +/- 0.623 | higher is better |
| Symbolic -> RepTok | ImageNet 1000 x 10 | ResNet target-class accuracy | Dynamic + sigmoid keep | 3 | 0.9525 | 0.957567 +/- 0.00177 | +0.00506669 +/- 0.00177 | higher is better |
| Symbolic -> RepTok | ImageNet 1000 x 10 | ResNet target-class accuracy | Dynamic + softmax keep | 3 | 0.9499 | 0.9546 +/- 0.00329 | +0.00469995 +/- 0.00329 | higher is better |
| Symbolic -> RepTok | ImageNet 1000 x 10 | ResNet target-class accuracy | Pinned + sigmoid keep | 3 | 0.9459 | 0.953833 +/- 0.0016 | +0.0079333 +/- 0.0016 | higher is better |
| Symbolic -> RepTok | ImageNet 1000 x 10 | ResNet target-class accuracy | Pinned + softmax keep | 3 | 0.9517 | 0.9554 +/- 0.00151 | +0.00370002 +/- 0.00151 | higher is better |
| Symbolic -> RepTok | ImageNet 1000 x 10 | YOLO IoU | Dynamic + sigmoid keep | 3 | 0.460004 | 0.458865 +/- 0.00115 | -0.00113982 +/- 0.00115 | higher is better |
| Symbolic -> RepTok | ImageNet 1000 x 10 | YOLO IoU | Dynamic + softmax keep | 3 | 0.459704 | 0.448074 +/- 0.00171 | -0.0116296 +/- 0.00171 | higher is better |
| Symbolic -> RepTok | ImageNet 1000 x 10 | YOLO IoU | Pinned + sigmoid keep | 3 | 0.462827 | 0.454263 +/- 0.00166 | -0.00856329 +/- 0.00166 | higher is better |
| Symbolic -> RepTok | ImageNet 1000 x 10 | YOLO IoU | Pinned + softmax keep | 3 | 0.460617 | 0.456822 +/- 0.00312 | -0.0037945 +/- 0.00312 | higher is better |
| Symbolic -> RepTok | ImageNet 1000 x 10 | YOLO detection rate | Dynamic + sigmoid keep | 3 | 0.9847 | 0.9822 +/- 0.000889 | -0.00249996 +/- 0.000889 | higher is better |
| Symbolic -> RepTok | ImageNet 1000 x 10 | YOLO detection rate | Dynamic + softmax keep | 3 | 0.986 | 0.978367 +/- 0.000379 | -0.00763335 +/- 0.000379 | higher is better |
| Symbolic -> RepTok | ImageNet 1000 x 10 | YOLO detection rate | Pinned + sigmoid keep | 3 | 0.9817 | 0.9781 +/- 0.000819 | -0.0036 +/- 0.000819 | higher is better |
| Symbolic -> RepTok | ImageNet 1000 x 10 | YOLO detection rate | Pinned + softmax keep | 3 | 0.9816 | 0.9801 +/- 0.00122 | -0.00150005 +/- 0.00122 | higher is better |