Bridging Machine and Human Cognition: From Prediction to Modulation

Join us with Zoom

Abstract: NeuroAI, where data and methodologies from AI and neuroscience converge, has gained accelerating traction across academia and industry over the past decade. Its intersection with vision and cognitive science has been particularly fruitful: it has driven the rise of modern AI vision systems, advanced new fields like "mind-reading" from recorded brain activity, and provided powerful tools to probe the visual system itself. In this talk, I will present how today's machine-executable models of the ventral visual stream are now accurate enough to design visually delivered interventions that causally modulate primate neural population activity and behavior. Using leading ANN vision models, we guide pixel-level perturbations and predict high-level neural and behavioral responses at the edge of the natural image regime. We then empirically validate their predictions through neural recordings in the inferior temporal (IT) cortex of nonhuman primates and complementary human psychophysics. Together, these results highlight the role of contemporary models of primate visual cognition in advancing our mechanistic understanding of vision and their promise for model-driven modulation of neural and behavioral states.

Bio: Guy is a Research Scientist at Meta Reality Labs in New York, specializing in neural interfaces. Previously, he was a Postdoctoral Researcher in the DiCarlo Lab at MIT, focusing on the intersection between human and machine vision. His recent thrusts aim to harness models of primate visual cognition for beneficial neural and behavioral modulation. During his Ph.D., advised by Prof. Michal Irani, Guy pioneered methods for reconstructing visual experience from brain activity. He holds a Ph.D. in Computer Science and an M.Sc. in Physics from the Weizmann Institute of Science, and a B.Sc. in Physics and EECS from the Hebrew University of Jerusalem. Guy was awarded the Fulbright Postdoctoral Fellowship in 2021.