{"paper":{"title":"Feature Visualization Recovers Known Cortical Selectivity from TRIBE v2","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Feature visualization via gradient ascent on a brain encoder recovers the known progression of selectivity from V1 to V4 and distinctive patterns for MT, FFA and PPA.","cross_cats":["cs.LG"],"primary_cat":"q-bio.NC","authors_text":"Brinnae Bent, Stuart Bladon","submitted_at":"2026-05-13T00:55:45Z","abstract_excerpt":"Brain encoder models predict cortical fMRI responses from the internal activations of pretrained vision and language networks, and are typically evaluated by held-out prediction accuracy. This is a useful signal for training but a poor one for interpretation: it tells us an encoder fits the data without telling us whether it has internalized the functional organization of the brain. We propose feature visualization -- gradient ascent on the encoder's predicted activation for a target region of interest (ROI) -- as a complementary interpretability technique, and apply it to TRIBE v2 composed wi"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"The probe recovers a visible progression of increasing spatial scale and feature complexity across V1 to V4, matching the ventral-stream hierarchy. It also produces three distinctive downstream regimes: radial frozen-motion streaks for MT, face-like features for FFA, and consistent rectilinear line patterns for PPA. Optimized FFA stimuli drive the predicted region ~4x as much as a natural face photograph.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That gradient-ascent optimization on the encoder's output for a target ROI reveals the functional organization the model has internalized, rather than being dominated by optimization artifacts or adversarial effects unrelated to biological selectivity.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Feature visualization on TRIBE v2 brain encoders recovers the known ventral visual hierarchy from V1 to V4 and produces distinctive patterns for MT, FFA, and PPA, with optimized stimuli driving ~4x higher activation than natural images.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Feature visualization via gradient ascent on a brain encoder recovers the known progression of selectivity from V1 to V4 and distinctive patterns for MT, FFA and PPA.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"082dd20bce0be21af97a66dd43c1c4f89696b1401a0c17a94d47eff93ff06643"},"source":{"id":"2605.13904","kind":"arxiv","version":1},"verdict":{"id":"561c7444-d418-46ce-87ab-cf9513dfaf59","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T02:55:25.341664Z","strongest_claim":"The probe recovers a visible progression of increasing spatial scale and feature complexity across V1 to V4, matching the ventral-stream hierarchy. It also produces three distinctive downstream regimes: radial frozen-motion streaks for MT, face-like features for FFA, and consistent rectilinear line patterns for PPA. Optimized FFA stimuli drive the predicted region ~4x as much as a natural face photograph.","one_line_summary":"Feature visualization on TRIBE v2 brain encoders recovers the known ventral visual hierarchy from V1 to V4 and produces distinctive patterns for MT, FFA, and PPA, with optimized stimuli driving ~4x higher activation than natural images.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That gradient-ascent optimization on the encoder's output for a target ROI reveals the functional organization the model has internalized, rather than being dominated by optimization artifacts or adversarial effects unrelated to biological selectivity.","pith_extraction_headline":"Feature visualization via gradient ascent on a brain encoder recovers the known progression of selectivity from V1 to V4 and distinctive patterns for MT, FFA and PPA."},"references":{"count":20,"sample":[{"doi":"","year":null,"title":"V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning","work_id":"a9c28401-f16a-4933-89f0-788e2f94e52b","ref_index":1,"cited_arxiv_id":"2506.09985","is_internal_anchor":true},{"doi":"10.1126/science.aav9436","year":null,"title":"Preprint at bioRxiv 461525","work_id":"adcae95f-e62d-44fb-9450-484d1c7f4a78","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1146/annurev.neuro.26.041002","year":null,"title":"doi: 10.1146/annurev.neuro.26.041002. 131052. d’Ascoli, S., Rapin, J., Benchetrit, Y ., Banville, H., and King, J.-R. TRIBE: TRImodal brain encoder for whole- brain fMRI response prediction.arXiv prep","work_id":"0274a6a5-2708-4e24-b26d-b5fd87528fab","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.48550/arxiv.2507.22229","year":2025,"title":"arXiv preprint arXiv:2507.22229 , year=","work_id":"8e8c13a0-820d-474f-a9eb-df5342f906c0","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1038/33402","year":null,"title":"Erhan, D., Bengio, Y ., Courville, A., and Vincent, P","work_id":"fbf260ca-ccfc-4d17-a5b6-f122a681d32f","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":20,"snapshot_sha256":"864005cc646bb2dcacbcaa95c0198924cdf6d773ce5d70cade53dbc615cba710","internal_anchors":4},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}