Sompolinsky, Haim
26 publications
ICLR
2025
When Narrower Is Better: The Narrow Width Limit of Bayesian Parallel Branching Neural Networks
NeurIPS
2024
Dissecting the Interplay of Attention Paths in a Statistical Mechanics Theory of Transformers
NeurIPSW
2024
Diverse Capability and Scaling of Diffusion and Auto-Regressive Models When Learning Abstract Rules