AlignedCut: Visual Concepts Discovery on Brain-Guided Universal Feature Space
This work addresses the challenge of understanding visual information processing in AI and neuroscience, offering insights into how deep networks form visual concepts, which is foundational but incremental in bridging these fields.
The paper tackled the problem of connecting visual data, deep networks, and brain activity by creating a universal channel alignment using brain fMRI response prediction as the training objective, resulting in the discovery of shared feature channels across models that correspond to brain regions and produce semantically meaningful object segments without supervision.
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.