NeurIPS 2025 Explorer
Concepts
Authors
Glossary
johnsanterre.github.io
Samet Oymak
4 papers
University of Michigan - Ann Arbor
Attention with Trained Embeddings Provably Selects Important Tokens
BREAD: Branched Rollouts from Expert Anchors Bridge SFT & RL for Reasoning
Extrapolation by Association: Length Generalization Transfer In Transformers
Theoretical Insights into In-context Learning with Unlabeled Data