I am an undergraduate researcher at Hubei University’s Manchester Metropolitan Joint Institute and an incoming M.Eng. student in Communication Engineering at East China Normal University’s School of Excellent Engineers, starting in September 2027. I am currently a research intern in Prof. Lin Yang’s group at The Hong Kong Polytechnic University and an Algorithm Research Intern at Kingsoft AI Product Center.
My research focuses on reliable learning and decision-making, spanning foundation models, reinforcement-learning agents, multimodal perception, and safety-critical control. At PolyU, I develop algorithms for medical AI, including large language models for nursing and diabetes care, together with evaluation frameworks for these models. At Kingsoft, I work on foundation-model pretraining and game-playing agents.
I develop DanKS, an open-source GuanDan AI system, and awesome-latex-skills, a collection of academic workflows for AI agents. I am also a Kaggle Competitions Expert, ranked in the top 1%. Previously, I interned at Z.ai and TeleAI.
The manuscripts below are under review. Public preprints are linked where available; other full texts are not yet public. Select an overview figure to enlarge it.
Human Motion Generation
Submitted to ICLR 2027 · Under review
MoPhys combines multi-scale time–frequency modeling with physics-inspired expert collaboration to refine text-generated 3D human motion using supervision from real motion.
Decision-making
Submitted to AAAI 2027 · Under review · Project / Code
A structure-aware candidate ranker and candidate-conditioned actor–critic for long-horizon decisions in large combinatorial action spaces.
Multi-Agent Learning
Submitted to AAAI 2027 · Under review
Multiple agents turn interaction evidence into reusable, human-readable strategy factors without retraining the underlying policies.
Medical Vision
Submitted to npj Digital Medicine · Under review
BiTraQ-DN recovers missed lesion hypotheses from neighboring frames and verifies them against residual target-frame memory.
LLM Safety
Submitted to ICASSP 2027 · Under review
SafeTab isolates risky table spans using structural, semantic, and rule-based evidence while preserving information needed for reasoning.
Efficient LLM Inference
Submitted to ICASSP 2027 · Under review
A memory-state-aware difficulty estimator and calibrated router that selects the least costly model within a quality tolerance.
Safe Control
Submitted to ICRA 2027 · Under review · arXiv:2609.27463 ↗
A distributed controller that decouples trajectory communication from locally sensed safety constraints under packet loss.
Safe Control
Submitted to ACC 2027 · Under review · Sole author
Event-triggered nominal planning with per-sample safety projection for fast and reliable distributed voltage regulation.
REINFORCEMENT LEARNING · GAME AI
A GuanDan AI codebase spanning three generations: structured retrieval, learned candidate selection, and memory-aware PPO policies.
GitHub statistics checked 2026-10-01.
RESEARCH TOOLING · AGENT WORKFLOWS
Five reusable agent skills for common academic workflows, combining structured instructions, reference knowledge, and verification guardrails.
GitHub statistics checked 2026-10-01.
School of Nursing, The Hong Kong Polytechnic University.
Algorithm research and development for medical AI, including large language models for nursing and diabetes care and evaluation frameworks for these models.
Foundation-model pretraining, card-game strategy AI, and distributed training on Ascend NPU clusters.
In-vehicle document question answering, supervised fine-tuning, and scalable inference.
Multimodal public-safety analytics, detection and segmentation, and Qwen-VL fine-tuning.
M.Eng. in Communication Engineering
School of Excellent Engineers
Manchester Metropolitan Joint Institute
B.Eng. in Electronic Information Engineering
Average: 85/100
Expected UK degree classification: First-Class Honours