Yiming Yang

Recent Research

·        LLM-Based Agents for Problem Solving: Research spans instruction tuning, demonstration-guided control, retrieval-augmented reasoning, and reinforcement learning—enabling adaptive agents for complex tasks such as math problem solving, combinatorial search, code synthesis and revision, and PDE solving.

·        Combinatorial Optimization and AI for Science: Developed learning-based solvers for NP-hard problems using diffusion models, Langevin dynamics, and Fourier neural operators; contributed to neural PDE solving, scientific workload scheduling, and code-generation-driven scientific computing.

·        Scalable Oversight and Learning Beyond Human Supervision: Advanced reinforcement learning–based alignment frameworks that scale supervision using self-play, instructible reward models, and principle-driven fine-tuning—enabling controllable and generalizable LLMs that go beyond human-written data and guidance.

Publications (list)

Teaching

Last updated: June 7, 2025