Junze (Tony) Ye
Junze (Tony) Ye
I am passionate about facilitating the diffusion of generative AI under resource constraints, where efficiency is a first-order concern. Concretely, I work on the data aspect of LLM post-training and AI agents, including on-policy data curation and curriculum design for SFT-RL, as well as model evaluations. My research aims to combine engineering pragmatics with a formal perspective from applied probability and sequential decision-making to optimize these learning systems for reliability and usefulness.
Faculty Advisors
Research Interests
- Applied Probability
- Sequential decision making
- Data-centric AI (Active learning / Curriculum learning)
- Post-Training
Publications
Working Papers
Finalist for INFORMS HAS Student Paper Competition. ArXiv preprint: https://arxiv.org/abs/2512.19691 Code and data release: https://github.com/junzeye/validate-medcalc-labels
Accepted by ICML 2026 workshop: RL from World Feedback (RLxF). TL;DR: When finetuning LLM agents, spending teacher labels on broader student-context coverage can be more effective than spending it on longer or more heavily filtered teacher completions.
A poster version was presented at CS 329A's poster session on December 12, 2025.