Xiaoxiao Shaun Li

I am a tenured Associate Professor in the Electrical and Computer Engineering Department and an Associate Member in the Computer Science Department at the University of British Columbia (UBC). I am also a faculty member at Vector Institute and an adjunct Professor at School of Medicine, Yale University. I am honored to be named as a Canada CIFAR AI Chair and a Canada Research Chair (Tier II) in Responsible AI. I am currently a Visiting Research Faculty at Google, where I work on GenAI and world model representation learning.

Before joining UBC, I was a postdoc working with Prof. Kai Li and Prof. Olga Troyanskaya at Princeton University. I obtained my Ph.D. degree from Yale University, where I was fortunate to be advised by Prof. James Duncan. I obtained my B.S. (honors degree) from Chu Kochen College, Zhejiang University, China, in June 2015.

I aim to narrow the gap between AI research and its applications by building the next generation of trustworthy and efficient AI systems — spanning the full stack from representation learning and multimodal perception to the reasoning, reinforcement learning, and agentic layers built on top. Please visit our group website Trusted and Efficient AI (TEA) Lab to learn about our up-to-date research projects and meet with my students.

Research Interests

My research currently spans the following themes; the pointers link to representative papers (see the full list on the publications page).

• Reinforcement Learning, Reward Modeling & Policy Optimization — understanding and stabilizing group-relative RL (GRPO/RLVR): exploration–exploitation steering, negative-gradient effects, reward hacking, and policy collapse.
  → Token Hidden Reward, On GRPO Collapse in Search-R1, GSS.
• Agentic Systems, Planning & Optimization — compound/agentic AI: budget-aware value-tree search, multi-agent→single-agent distillation, and optimizing multi-component pipelines.
  → Budget-Aware Value Tree Search, AgentArk, Textual Equilibrium Propagation.
• Data-Centric AI: Valuation, Curation & Synthetic Data — data valuation and knowledge-grounded curation for SFT/RLHF and generative models.
  → GMValuator, MedReason, Efficient Forward-Only Data Valuation.
• Evaluation, Robustness & Verification for Deployment — reward/evaluation frameworks, attention-distraction diagnostics for LVLMs, self-verification reliability, fairness and robustness benchmarks.
  → When RAG Hurts, Verification Mirage, RVCBench, FairMedFM.
• Optimization & Distributed (Large-Scale) Learning — federated/distributed learning, model merging, delta-parameter pruning, and convergence analysis.
  → Can Textual Gradient Work in FL?, DARE, Local Superior Soups.
• Vision & Multimodal Perception — vision-language and multimodal representation learning, graph-prompt learning, robust learning on noisy/missing signals, and high-order relational modeling.
  → LoFi, Learning High-Order Relationships of Brain Regions, MMGPL.

Honors & Awards

• 2026 — UBC Killam Research Fellowship; Amazon Research Award
• 2025 — Amazon Research Award; NVIDIA Academic Program Award; Google Gemini Research Award; NeurIPS Outstanding Area Chair; ICML AI for Math Workshop Best Paper Award; 3rd Place, Reasoning Dataset Competition
• 2024 — Canada Research Chair (Tier II) in Responsible AI
• 2023 — Canada CIFAR AI Chair; ICLR Outstanding Reviewer
• 2022 — Meta Research Award; Google Cloud Research Innovator Award
• 2021/22 — NVIDIA Academic Program Award
• 2018–2020 — MICCAI Student Participation and Travel Award; NeurIPS Student Travel (NIH-funded) Award; IPMI Conference Scholarship; OHBM Merit Abstract Award; ISBI NIH-funded Travel Award; Yale Advanced Graduate Leadership Fellowship