Nathan Srebro

Learning to Answer from Correct Demonstrations

Positive Distribution Shift as a Framework for Understanding Tractable Learning

Learning single index models via harmonic decomposition

Temperature is All You Need for Generalization in Langevin Dynamics and other Markov Processes

PENCIL: Long Thoughts with Short Memory

Weak-to-Strong Generalization Even in Random Feature Networks, Provably

Shift is Good: Mismatched Data Mixing Improves Test Performance