From Activation to Causality: BrainCause Discovers Causal Visual Representations in the Human Brain

Paper Overview

• Field: CV
• Authors: Yuval Golbari, Navve Wasserman, Matias Cosarinsky
• arXiv: 2505.21397

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.

The authors introduce BrainCause, an automated framework that combines generative and brain models to synthesize controlled stimuli and validate neural representations through targeted causal testing.

How it works

• Given a query specifying a concept of interest, the framework constructs targeted stimulus sets comprising:
• concept images,
• counterfactual edits that remove the target concept while preserving the rest of the image content,
• images containing candidate correlated distractors.
• An image-to-fMRI encoding model predicts brain responses to these stimuli.
• The framework searches for representations that respond specifically to the target concept rather than to correlated alternatives.

Findings

• BrainCause returns validated candidate representations and proposes follow-up fMRI experiments to further test or extend its findings.
• The method successfully recovers known functional localizations and identifies new candidate representations for dozens of concepts, validated on both predicted and measured fMRI data.
• Critically, without causal verification, a large fraction of localizations would be false positives — confirming that activation alone is insufficient evidence of representation.

Original Abstract (excerpt)

> 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...

*Auto-collected on 2026-05-26.*