Junze (Tony) Ye

PhD Student, Operations, Information & Technology

PhD Program Office Graduate School of Business Stanford University 655 Knight Way Stanford, CA 94305

Junze (Tony) Ye

I work on the data aspect of LLM post-training and AI agents, including on-policy data curation, SFT-RL recipes, and evaluation signal quality. My research aims to bring a perspective from applied probability and sequential decision-making to analyze and optimize these machine learning systems.

Faculty Advisors

Mohsen Bayati

Research Interests

Applied Probability
Sequential decision making
Data
Post-Training

Publications

ScoreFusion: Fusing Score-based Generative Models via Kullback-Leibler Barycenters

AISTATS 2025 Oral

Working Papers

Scalable Stewardship of an LLM-Assisted Clinical Benchmark with Physician Oversight

- ArXiv preprint: https://arxiv.org/abs/2512.19691 - Code and data release: https://github.com/junzeye/validate-medcalc-labels

A Few Teacher Steps Go a Long Way: Cost-Efficient On-Policy Data Augmentation for Agent Post-Training

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.

LLM-Gated DAgger: Interactive Imitation Learning for Finetuning QA Agents

PDF

A poster version was presented at CS 329A's poster session on December 12, 2025.