Wei Li

This paper proposes a flexible framework called Sofia for reliable Synthetic Song Detection (SSD) using music-intrinsic attributes and an adaptive Mixture-of-Experts module.

This paper presents video-SALMONN-R$^3$, an end-to-end video-LLM that enables re-watch through reinforcement learning, improving question answering performance with a two-stage paradigm.

This paper introduces a provably sound runtime meter for KV-cache quantization to ensure compression does not damage the served request in real-time.

The paper introduces DrugHarness, an evolvable agent that composes scientific operations and adapts to changing evidence and constraints for computational drug discovery.

This paper improves kernel performance on GPUs by accounting for physical non-idealities in hardware resources.