AlignedCut: Visual Concepts Discovery on Brain-Guided Universal Feature Space

TL;DR

This paper introduces AlignedCut, a method that aligns deep network features with brain responses to discover shared visual concepts and analyze information processing across models.

Contribution

It proposes a brain-guided universal feature space that reveals common channels, visual concepts, and layer-wise processing in deep networks without supervision.

Findings

Shared feature channels across models correspond to brain regions.

Semantic object segments emerge from channel clusters.

The method enables precise comparison of network layer processing.

Abstract

We study the intriguing connection between visual data, deep networks, and the brain. Our method creates a universal channel alignment by using brain voxel fMRI response prediction as the training objective. We discover that deep networks, trained with different objectives, share common feature channels across various models. These channels can be clustered into recurring sets, corresponding to distinct brain regions, indicating the formation of visual concepts. Tracing the clusters of channel responses onto the images, we see semantically meaningful object segments emerge, even without any supervised decoder. Furthermore, the universal feature alignment and the clustering of channels produce a picture and quantification of how visual information is processed through the different network layers, which produces precise comparisons between the networks.