Yaowenqi (Howard) Liu
Master of Computer Science, University of Illinois Urbana-Champaign
I am a Master of Computer Science student at the University of Illinois Urbana-Champaign (UIUC), where I also completed my undergraduate degree in Computer Science & Statistics. I am a research intern working with Professor Tong Zhang.
My research is on making LLM agents reliable from verifiable feedback like Lean 4, retrieved evidence, and uncertainty quantification.
Making LLM agents reliable. I study this with verifiable feedback such as retrieved evidence, uncertainty estimates, formal languages, and reward-guided training.
Agents for automated research in machine learning: empirical work grounded in code execution and retrieved literature, and theoretical work checked with Lean 4.
GPA 3.9/4.0
GPA 3.9/4.0
* Co-first author.
GUIDE: Towards Scalable Advising for Research Ideas
An LLM-based system that evaluates the novelty and significance of ML papers. It retrieves evidence from a 200K-paper database, and its 7B reviewer model is trained with RLHF to correct the optimistic bias of LLM-generated reviews.
Optimal Aggregation of LLM and PRM Signals for Efficient Test-Time Scaling
A theoretical framework showing the optimal test-time scaling strategy is a weighted majority vote over LLM and PRM signals, with calibrated weights that penalize low-quality responses.
AgentSPEX: An Agent SPecification and EXecution Language
A YAML-based language for specifying LLM-agent workflows with explicit control flow, typed steps, parallelism and reusable submodules, paired with a customizable agent harness.