Lei Wang

DCVD proposes a dual-channel cross-modal fusion framework that jointly detects software vulnerabilities and precisely localizes the vulnerable lines, outperforming existing state-of-the-art methods.

AccLock proposes a passive, zero-involvement user authentication system that uses unique biometric features from in-ear accelerometers (BCG signals) to achieve secure and unobtrusive identity verification.

The paper introduces a 'replication-first' paradigm for LLM behavioral benchmarking, demonstrating that this rigorous approach uncovers significant, non-obvious performance drops between successive model versions, such as a notable decline in advice-restraint for GPT-5.

SkillC introduces a Contrastive Skill Credit Assignment (CSCA) framework to enable LLM agents to autonomously internalize skills during training, significantly outperforming existing methods without requiring runtime skill access.

The paper identifies and demonstrates that post-conclusion continuation in answer-correct long-CoT traces is harmful during LLM fine-tuning, proposing a method to cut this continuation.

The paper proposes a sequence-alignment framework using Soft Dynamic Time Warping to evaluate audio-driven talking-head generation, demonstrating that this approach provides more robust and fair comparisons than traditional frame-wise metrics.