Kalanit Grill-Spector, Ph.D.

Professor in Psychology

Department of Psychology

School of Humanities and Sciences at Stanford University

The future of human vision: a cognitive computational neuroanatomical approach to study the human visual system

fMRI and computational modeling have transformed our understanding of the human brain. In the visual system, modeling population receptive fields (pRF) led to discoveries of multiple maps of pRF eccentricity, polar angle, and size and revealed both maps and clustered representations of visual categories. However, it is unknown how functional representations are scaffolded by brain structure, what is the nature of temporal computations, and what kind of constraints generate functional maps. Here, I will describe how innovations in imaging technologies such as quantitative and diffusion MRI, as well as computational modeling, referred together as ¬cognitive computational neuroanatomy, have significantly advanced understanding of the human ventral visual stream that is involved in visual categorization. First, I will describe empirical data revealing the interplay between cytoarchitecture, white matter connections, and category representations. Findings suggest that eccentricity and cytoarchitecture are basic organizing principles. Second, I will describe a new empirical and computational framework to estimate the spatiotemporal population receptive field (st-pRF) of each voxel in visual degrees and milliseconds, revealing increasing spatial and temporal integration windows across visual streams. Finally, I will describe a new topographic deep neural network (TDANN) that enables testing which constraints predict both cortical responses and spatial organization. We find that a single unified principle ¬– self supervised training together with a spatial loss that encourages units on the simulated cortical sheet to have correlated responses – predicts both functional responses and maps in the visual system– from V1 to high-level regions.

Faculty Host: Dr. Kristina Nielsen