Mark Schmidt

Implicit Bias of Spectral Descent and Muon on Multiclass Separable Data

ReMA: Learning to Meta-Think for LLMs with Multi-agent Reinforcement Learning

BlockLLM: Memory-Efficient Adaptation of LLMs by Selecting and Optimizing the Right Coordinate Blocks

Heavy-Tailed Class Imbalance and Why Adam Outperforms Gradient Descent on Language Models