Research
I develop graph-driven multi-agent systems that enable efficient, adaptive, and distributed LLM reasoning across edge–cloud environments. My research integrates graph intelligence, multi-agent collaboration, and efficient distributed systems.
Research Areas
Structured Reasoning
Graph-based reasoning methods organize LLM inference into structured, interpretable workflows. By representing reasoning steps as nodes in a directed graph, we can guide multi-agent systems through complex problem-solving with explicit dependencies and subject-based decomposition.
Related: S-DAG (AAAI 2026)
Collaborative Multi-Agent Systems
Multi-agent LLM systems coordinate specialized agents to solve problems beyond the capability of any single model. My work focuses on heterogeneous agent collaboration, where agents with different capabilities and roles communicate through structured protocols.
Related: S-DAG (AAAI 2026)
Efficient Distributed Inference
Edge–cloud collaboration enables efficient LLM inference by distributing computation across heterogeneous devices. I design adaptive scheduling and token-efficient strategies that minimize latency while respecting resource constraints at the edge.
Related: HybridFlow (ICML 2026)
Adaptive Learning
Continual and personalized learning in multi-agent settings allows systems to adapt to new tasks, user preferences, and changing environments without retraining from scratch. This includes federated personalization and online adaptation strategies.
Active Projects
- S-DAG Framework — Subject-based DAG for multi-agent reasoning. Accepted at AAAI 2026
- HybridFlow — Adaptive edge-cloud LLM inference scheduling. Accepted at ICML 2026
Collaborations
I am fortunate to work with my advisor Prof. Wanyu Lin and collaborators at The Hong Kong Polytechnic University.