Hugging Face Daily Papers · June 3, 2026 · 4 min read

From Activation to Causality: Discovery of Causal Visual Representations in the Human Brain

Mirrored from Hugging Face Daily Papers for archival readability. Support the source by reading on the original site.

Like Read original ↗

<video src=\"https://cdn-uploads.huggingface.co/production/uploads/6492c419702103104f9450c4/h8fw5hFf5Or7eBgYmgDEg.mp4\" controls=\"\" class=\"max-w-full!\"></video></p>","updatedAt":"2026-06-03T12:01:22.859Z","author":{"_id":"6492c419702103104f9450c4","avatarUrl":"/avatars/d6700e4bb1b9f172096ea31ba15a83b2.svg","fullname":"navve wasserman","name":"navvew","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":1,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.46098804473876953},"editors":["navvew"],"editorAvatarUrls":["/avatars/d6700e4bb1b9f172096ea31ba15a83b2.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2605.23895","authors":[{"_id":"6a1d847b808ddbc3c7d438a8","name":"Yuval Golbari","hidden":false},{"_id":"6a1d847b808ddbc3c7d438a9","name":"Navve Wasserman","hidden":false},{"_id":"6a1d847b808ddbc3c7d438aa","name":"Matias Cosarinsky","hidden":false},{"_id":"6a1d847b808ddbc3c7d438ab","name":"Roman Beliy","hidden":false},{"_id":"6a1d847b808ddbc3c7d438ac","name":"Aude Oliva","hidden":false},{"_id":"6a1d847b808ddbc3c7d438ad","name":"Antonio Torralba","hidden":false},{"_id":"6a1d847b808ddbc3c7d438ae","name":"Michal Irani","hidden":false},{"_id":"6a1d847b808ddbc3c7d438af","name":"Tamar Rott Shaham","hidden":false}],"publishedAt":"2026-05-22T00:00:00.000Z","submittedOnDailyAt":"2026-06-03T00:00:00.000Z","title":"From Activation to Causality: Discovery of Causal Visual Representations in the Human Brain","submittedOnDailyBy":{"_id":"6492c419702103104f9450c4","avatarUrl":"/avatars/d6700e4bb1b9f172096ea31ba15a83b2.svg","isPro":false,"fullname":"navve wasserman","user":"navvew","type":"user","name":"navvew"},"summary":"Identifying which brain regions represent a visual concept in the human brain is a central challenge in neuroscience. Existing approaches have localized coarse functional regions (e.g., faces, places) through activation maximization, identifying regions that activate strongly for a target concept relative to other concepts. Yet strong activation alone does not establish that a region represents the concept itself, as responses may instead be driven by correlated visual or semantic cues. We introduce BrainCause, an automated framework that combines generative and brain models to synthesize controlled stimuli and validate neural representations through targeted causal testing. Given a query specifying a concept of interest, our framework constructs targeted stimulus sets comprising concept images, counterfactual edits that remove the target concept while preserving other image content, and images with candidate correlated distractors. It then uses an image-to-fMRI encoding model to predict brain responses and searches for representations that respond specifically to the target concept over correlated alternatives. BrainCause returns validated candidate representations and proposes follow-up fMRI experiments to further test or extend its discoveries. Our approach successfully recovers known functional localizations and identifies new candidate representations across dozens of concepts, validated on both predicted and measured fMRI data. Critically, we show that without causal validation, a large fraction of localizations would be false positives, confirming that activation alone is insufficient evidence of representation.","upvotes":17,"discussionId":"6a1d847b808ddbc3c7d438b0","projectPage":"https://yuvalgol123.github.io/BrainCause/","ai_summary":"BrainCause framework uses generative and brain models to identify valid neural representations through causal testing, demonstrating that activation alone is insufficient for confirming concept representation.","ai_keywords":["activation maximization","image-to-fMRI encoding model","causal testing","functional localizations","concept representation","brain models","generative models"],"ai_summary_model":"Qwen/Qwen2.5-Coder-32B-Instruct","organization":{"_id":"6a0b110e0f1cfb3293dae517","name":"BrainCause","fullname":"BrainCause","avatar":"https://www.gravatar.com/avatar/e799531d41bb47accbfb7fd7261307e4?d=retro&size=100"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"6492c419702103104f9450c4","avatarUrl":"/avatars/d6700e4bb1b9f172096ea31ba15a83b2.svg","isPro":false,"fullname":"navve wasserman","user":"navvew","type":"user"},{"_id":"693ab3a7a053d510053283ac","avatarUrl":"/avatars/00d27390171ccdd58aa1f03c6672a6be.svg","isPro":false,"fullname":"Noor Athamnah","user":"NoorAth","type":"user"},{"_id":"6698688dd723c35c0ea9f16d","avatarUrl":"/avatars/0748274382d0814bb52b9491bb0a329c.svg","isPro":false,"fullname":"Roman Beliy","user":"RomanBeliy","type":"user"},{"_id":"67c5d126bf57d282996fe700","avatarUrl":"/avatars/8b7b3eaa57a3e04109f5cb071ef3b5ef.svg","isPro":false,"fullname":"Jonathan Kogman","user":"jkogman","type":"user"},{"_id":"6622956521adda2fcf9daf0e","avatarUrl":"/avatars/7cb9c2ebddad57269d803608decad516.svg","isPro":false,"fullname":"Peter Sampodrias","user":"petersampodrias","type":"user"},{"_id":"6735088b22d14a01ae17501f","avatarUrl":"/avatars/23d2eb2bb833dcf7a05434b499fedd5e.svg","isPro":false,"fullname":"Matias Cosarinsky","user":"mcosarinsky","type":"user"},{"_id":"687f44e793eb81b0684b4eee","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/uDef9M8sQxTweYvwcVs07.png","isPro":false,"fullname":"Yuval Golbari","user":"yuvalgolbari","type":"user"},{"_id":"6a2026d0fea3f60347fa7bf9","avatarUrl":"/avatars/e7f38c2cbad5ae9182ff404702524478.svg","isPro":false,"fullname":"Shahak Oren","user":"ShahakOren","type":"user"},{"_id":"678bfa754dd9bb4f2b8c8c0b","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/aUXxZHGHhTgLYKocSE4Jg.png","isPro":false,"fullname":"Itamar Fruchter","user":"PitzF","type":"user"},{"_id":"62b3e85bcbd2a402fc7804b1","avatarUrl":"/avatars/63125ce8a1e20b8c6e836f223d24284f.svg","isPro":false,"fullname":"noam rotstein","user":"noamrot","type":"user"},{"_id":"6745d69eef06fb82fc7d7f33","avatarUrl":"/avatars/c1b8d041cbebc50f71a8a2460c3c3c00.svg","isPro":false,"fullname":"Spiegel","user":"orspiegel","type":"user"},{"_id":"62f16fea335ce25164bf7ef0","avatarUrl":"/avatars/b50b633c70cbd1c66ec2fc556552638f.svg","isPro":false,"fullname":"David Novikov","user":"DavidNovikov","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"6a0b110e0f1cfb3293dae517","name":"BrainCause","fullname":"BrainCause","avatar":"https://www.gravatar.com/avatar/e799531d41bb47accbfb7fd7261307e4?d=retro&size=100"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2605/2605.23895.md"}">

Papers

arxiv:2605.23895

From Activation to Causality: Discovery of Causal Visual Representations in the Human Brain

Published on May 22

· Submitted by

navve wasserman on Jun 3

BrainCause

Upvote

Authors:

Abstract

BrainCause framework uses generative and brain models to identify valid neural representations through causal testing, demonstrating that activation alone is insufficient for confirming concept representation.

Generated by Qwen/Qwen2.5-Coder-32B-Instruct

Identifying which brain regions represent a visual concept in the human brain is a central challenge in neuroscience. Existing approaches have localized coarse functional regions (e.g., faces, places) through activation maximization, identifying regions that activate strongly for a target concept relative to other concepts. Yet strong activation alone does not establish that a region represents the concept itself, as responses may instead be driven by correlated visual or semantic cues. We introduce BrainCause, an automated framework that combines generative and brain models to synthesize controlled stimuli and validate neural representations through targeted causal testing. Given a query specifying a concept of interest, our framework constructs targeted stimulus sets comprising concept images, counterfactual edits that remove the target concept while preserving other image content, and images with candidate correlated distractors. It then uses an image-to-fMRI encoding model to predict brain responses and searches for representations that respond specifically to the target concept over correlated alternatives. BrainCause returns validated candidate representations and proposes follow-up fMRI experiments to further test or extend its discoveries. Our approach successfully recovers known functional localizations and identifies new candidate representations across dozens of concepts, validated on both predicted and measured fMRI data. Critically, we show that without causal validation, a large fraction of localizations would be false positives, confirming that activation alone is insufficient evidence of representation.

View arXiv page View PDF Project page Add to collection