About
I am a Ph.D. student at The University of Hong Kong (HKU), advised by Prof. Hongyang Du. I joined the Network Intelligence and Computing Ecosystem (NICE) Lab in June 2025. Previously, I completed my master's studies at the Institute of Computing Technology, Chinese Academy of Sciences (ICT, CAS), advised by Prof. Zhenyu Li. I completed my undergraduate studies at Hunan University.
My research lies at the intersection of large language models and networked systems. I focus on building reliable and efficient LLM agents that can adapt to changing network and execution conditions. My recent work explores network-aware tool routing, reliable agent orchestration, and selective on-policy distillation for LLM post-training. More broadly, I am interested in combining learning algorithms with system design to improve the practical performance of intelligent systems.
News
HACO has been accepted to NeurIPS 2026.
Started my Ph.D. at HKU and joined the NICE Lab.
SAROS has been accepted to APNet 2024.
Selected Publications
View All →SpatialCast: Reading the 3D World via Cast Shadows
Enhan Li, Jinghao Zhou, Jiaze Zhang, Hongyang Du
arXiv preprint arXiv:2611.22657 · 2026
A cast-shadow benchmark for diagnosing spatial reasoning in multimodal LLMs, plus GRPO post-training on its data.
CROP: Task Relevance via Counterfactuals for Selective On-Policy Distillation
Enhan Li, Junhao He, Hongyang Du
arXiv preprint arXiv:2608.13387 · 2026
Allocating token-level supervision using counterfactual task relevance for selective on-policy distillation.
HACO: Hedged Agent Computing for Reliable LLM Systems
Enhan Li, Hongyang Du
Advances in Neural Information Processing Systems (NeurIPS, accepted) · 2026
Reliable agent execution through adaptive hedge selection under changing network and service conditions.