Yizhou Chi

Yizhou Chi 池一舟

I am a PhD candidate in Computer Science at the University of Cambridge, where I work with Andreas Vlachos on building language agents that collaborate safely and effectively with people. I study how we can give models the tools to interpret subtle intent, reason through multi-step tasks, and stay grounded in long-term conversations.

Before Cambridge, I completed my M.S. in EECS and B.A. in Computer Science & Cognitive Science at UC Berkeley under the guidance of Dan Klein. Those years sparked my fascination with language understanding, structured reasoning, and the human side of AI systems.

Research focus

• Designing interactive language agents that can clarify uncertainty, self-correct, and remain aligned with user goals over time.
• Developing reasoning frameworks that help AI build faithful world models and explain their decisions.
• Studying how people and models co-adapt in conversational settings, with an emphasis on safety, transparency, and evaluation.

Publications

• SELA: Tree-Search Enhanced LLM Agents for Automated Machine Learning
  Authors: Yizhou Chi, Yizhang Lin, Sirui Hong, Duyi Pan, Yaying Fei, Guanghao Mei, Bangbang Liu, Tianqi Pang, Jacky Kwok, Ceyao Zhang, Bang Liu, Chenglin Wu
  We frame AutoML pipeline design as a tree-search problem, enabling LLM agents guided by Monte Carlo Tree Search to plan experiments, reuse past insights, and converge on strong configurations.
  [paper] · [code]

• THOUGHTSCULPT: Reasoning with Intermediate Revision and Search
  Authors: Yizhou Chi, Kevin Yang, Dan Klein
  Findings of NAACL 2024
  We demonstrate how decomposing candidate solutions and revising them via MCTS yields state-of-the-art structured reasoning while keeping the search space inspectable.
  [paper] · [code]

• CLARINET: Augmenting Language Models to Ask Clarification Questions for Retrieval
  Authors: Yizhou Chi, Jessy Lin, Kevin Lin, Dan Klein
  By conditioning on retrieval distributions, CLARINET teaches LLMs to ask targeted clarification questions that reduce ambiguity and consistently outperform heuristic baselines.
  [paper]

• Feature Selection of High Dimensional Data by Adaptive Potential Particle Swarm Optimization
  Authors: Xingyue Huang, Yizhou Chi, Yu Zhou
  IEEE CEC 2019
  We adapt particle swarm optimization with ReliefF-driven filtering to handle small-sample, high-dimensional classification tasks more robustly than standard PSO variants.
  [paper]

Projects & community work

• Among Agents — ACL 2024 Wordplay Workshop
  A text-based social deduction environment for probing how language agents reason under partial information and social pressure.
  [paper] · [code]

• Interactive Flock Simulator
  A real-time flocking visualizer that pairs physics-based animation with intuitive controls.
  [demo]