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

Forum topic · 小凯 · 2026-05-26

Summary

A new framework called BrainCause moves beyond activation-based brain mapping to causally validate visual representations in the human brain. Existing methods localize coarse functional regions (e.g., faces, places) via activation maximization, but strong activation alone does not prove a region represents a concept—responses may be driven by correlated visual or semantic cues. BrainCause, developed by Yuval Golbari, Navve Wasserman, and Matias Cosarinsky (arXiv:2505.21397), combines generative models with image-to-fMRI encoding models to synthesize controlled stimuli, including counterfactually edited images that remove the target concept while preserving other content, plus images with distractor concepts. It then searches for representations that respond specifically to the target concept rather than correlated alternatives, and proposes follow-up fMRI experiments. The method successfully recovers known functional localizations and identifies new candidate representations for dozens of concepts, validated on both predicted and measured fMRI data. Crucially, the authors show that without causal verification, many localizations would be false positives—demonstrating that activation alone is insufficient evidence of neural representation.

Field: Computer Vision / Neuroscience Authors: Yuval Golbari, Navve Wasserman, Matias Cosarinsky arXiv: 2505.21397

Original Abstract

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, the framework constructs targeted stimulus sets comprising concept images, counterfactual edits that remove the target concept while preserving other image content, and images carrying candidate correlated distractors. An image-to-fMRI encoding model then predicts brain responses, and the framework searches for representations that respond specifically to the target concept rather than to correlated alternatives. BrainCause returns verified candidate representations and proposes follow-up fMRI experiments to further test or extend its findings. The approach successfully recovers known functional localizations and identifies new candidate representations for dozens of concepts, validated on both predicted and measured fMRI data. Crucially, the authors show that without causal verification, a large fraction of localizations would be false positives—confirming that activation alone is insufficient evidence of representation.

Key Takeaways

Problem: Activation maximization can localize brain regions that respond strongly to a concept, but strong activation may reflect correlated visual or semantic cues rather than true representation.
Solution: BrainCause combines generative models and image-to-fMRI brain models to run targeted causal tests with controlled and counterfactual stimuli.
Results: Recovered known functional regions (faces, places) and uncovered new candidate representations for dozens of concepts, verified against measured fMRI data.
Implication: Many activation-based localizations are false positives; causal verification is essential for claiming neural representation.

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

Summary

Original Abstract

Key Takeaways

Tags