Roger Grosse

Associate Professor and Schwartz Reisman Chair in Technology and Society at University of TorontoMember of Technical Staff, Alignment Science at AnthropicFounding Member at Vector Institute

Grosse co-created K-FAC, a method that speeds neural-network training by using a practical approximation to second-order optimization. He now applies that understanding to alignment at Anthropic, tracing surprising model behavior back to influential training examples and testing models for deception.