About Me

Hi, I am Yuanming Shao. I am a second-year master’s student at the University of Pennsylvania, in the Computer and Information Science (CIS) Department, where I am fortunate to work with Ryan Marcus and Zack Ives. Before this, I received my bachelor’s degree in Computer Science from ShanghaiTech University.

I’m actively looking for research collaborations and happy to chat. Feel free to reach out!

Email / Google Scholar / GitHub / LinkedIn

Research Interests

I am broadly interested in efficient, interpretable, and scalable machine learning, particularly generative models and machine learning systems. I aim to better understand how models work and develop simple, effective methods.

I’m interested in three broad questions:

• Generative Modeling: What principles underlie generative models, and how can they inform better methods for sampling and scientific applications?
• Efficiency: How can we accelerate training and inference while preserving model quality under limited computational resources?
• Machine Learning & Systems: How can machine learning improve system performance, and how can systems techniques make AI applications more scalable and reliable?

Publications

Data Canvas: A Provenance-Guided Harness for Agentic Data Engineering
COLM’26
Zixuan Yi, Yuanming Shao, Shaun Wallace, Zachary Ives, Ryan Marcus
[paper] [slides]

When an agent goes wrong, can we identify the responsible step and repair only what it affected?

Agent workflows are difficult to inspect and repair when their outputs come from long, opaque execution traces. Data Canvas structures agent execution into semantic operators and tracks fine-grained provenance, allowing feedback to be traced to the responsible computation, propagated to related outputs, and repaired by replaying only the affected parts of the workflow.

IGAMT: Privacy-Preserving Electronic Health Record Synthesization with Heterogeneity and Irregularity.
AAAI’24
Wenjie Wang, Pengfei Tang, Jian Lou, Yuanming Shao, Lance Waller, Yi-an Ko, Li Xiong [code] [paper] [poster]

How can generative models capture heterogeneous, irregular sequences while preserving privacy?

We present IGAMT, the Imitative Generative Adversarial Mixed-embedding Transformer, for differentially private data synthesis. It combines Transformer-based representations with adversarial learning to capture heterogeneous features, missingness patterns, and irregular observation times. An Imitator network improves the privacy-utility trade-off. Experiments on electronic health records demonstrate the quality of the generated data and their utility for downstream tasks.

Teaching Experience

CIS 4190/5190 Applied Machine Learning, University of Pennsylvania
Teaching Assistant, Fall 2025, Spring 2026

CS 182 Introduction to Machine Learning, ShanghaiTech University
Teaching Assistant, Fall 2023

Miscellaneous

I’m always happy to chat — feel free to drop me an email about research or anything else on your mind. I’ve been lucky to be surrounded by wonderful friends, collaborators, and mentors, and I try to keep an optimistic outlook on life. And if you ever just need someone to talk to, feel free to reach out.