Representations in Deep Learning and the Brain

Speaker: Ambuj K Singh, Distinguished Professor of Computer Science at the University of California, Santa Barbara

Date: 18 March 2026

YouTube link: https://youtu.be/z3AoggYtiMo

The talk on ‘Representations in Deep Learning and the Brain’ by Ambuj K Singh was organised jointly with the Department of Computational and Data Sciences (CDS) at IISc. The attendees were from different departments in IISc, namely, the Centre for Neuroscience (CNS), Institute Mathematics Initiative (IMI), Department of Computer Science and Automation (CSA), and Department of Computational and Data Sciences (CDS). A summary of the talk is provided below.

Recent progress in brain reconstruction has shown that visual and semantic information can be decoded from fMRI into high-level representation spaces. In this talk, Ambuj Singh outlined a broader research direction centred on representation alignment as a unifying principle for brain modelling.

First, he discussed ongoing efforts to construct a shared, aligned brain representation space that maps multiple subjects into a common geometry. Rather than treating each subject or dataset independently, this framework aims to formalise subject-agnostic alignment and improve data efficiency through structured adapters.

Second, he described extensions beyond fMRI to M/EEG, exploring whether heterogeneous modalities can be integrated into the same representational space despite differences in spatial and temporal resolution. This raises foundational questions about what aspects of neural geometry are modality-invariant.

Finally, he broadened the perspective to task-dependent representations and comparisons with modern AI systems, including vision and language models. By studying representation structure, alignment, and geometry across biological and artificial systems, the aim is to better understand how task demands shape internal spaces and where human and machine representations diverge.