1 h 11 min
Seminar, Hosted by MIT Center for Brains, Minds and Machines — Quest | CBMM Seminar Series
A Fruitful Reciprocity: The Neuroscience-AI Connection
Stanford University
Recording
Abstract
NeuroAI uses techniques from artificial intelligence to model brain data, but the connection can run in both directions. Under "AI driving neuroscience", Dan Yamins presents a candidate universal principle of functional organisation in the brain, based on recent advances in self-supervised learning, that explains both fine detail and large-scale structure in the visual system and perhaps beyond. Under "neuroscience guiding AI", he presents a cognitively grounded computational theory of perception that yields robust learning algorithms for real-world scene understanding. Together, these lines of work suggest that neural networks optimised for cognitively informed tasks offer a unified framework for understanding the brain and for improving AI.