today local_bar

Zhihang Yuan

Researcher, ByteDance Inc.

2 papers at NeurIPS 2025

Homepage· OpenReview· Semantic Scholar· Google Scholar

Poster Session 4

Thursday, December 4, 2025 · 4:30 PM → 7:30 PM

Exhibit Hall C,D,E

SAFEx: Analyzing Vulnerabilities of MoE-Based LLMs via Stable Safety-critical Expert Identification

#3818 · ZhengLin Lai, Mengyao Liao, Bingzhe Wu, Dong Xu, Zebin Zhao, Zhihang Yuan, Chao Fan, Jianqiang Li

We identified a safety issue in the MoE architecture and designed experiments to demonstrate it.

R2R: Efficiently Navigating Divergent Reasoning Paths with Small-Large Model Token Routing

#5317 · Tianyu Fu, Yi Ge, Yichen You, Enshu Liu, Zhihang Yuan, Guohao Dai, Shengen Yan, Huazhong Yang, Yu Wang

We train a token-level neural router to let SLM following LLM reasoning paths by replacing only divergent tokens