Siyuan Ma

Preserving skill-level attention structures in MLLMs can dramatically reduce forgetting while adapting to new tasks without relying on replay mechanisms.

LLM judges in healthcare show promise for scalable evaluation, but their reliability swings wildly across tasks, demanding careful design and validation before trusting their verdicts.

Compute-in-memory's promise of energy efficiency hits a power delivery wall, requiring careful consideration of current demand patterns that can lead to voltage droop and thermal hotspots.