Songchen Ma

A training-free framework that reformulates failure recovery as test-time scaling over causal histories, which decomposes recovery into three coupled decisions: when to revise the causal history, where to recover a reliable history prefix, and which history configuration best supports subsequent execution.

LLMs can now compress their KV cache more effectively by dynamically synthesizing soft tokens tailored to the input, preserving crucial context that's otherwise lost with static methods.

Multi-chiplet architectures can unlock significant speedups and memory savings for low-batch MoE inference by dynamically scheduling expert computations across high-bandwidth die-to-die links.