Judy Hanwen Shen

I am a PhD student in the Theory Group in the Computer Science Department at Stanford University. I am fortunate to be advised by Omer Reingold. I work towards trustworthy machine learning through understanding data.

I am an advocate for PhD student well-being. I led the most recent PhD Climate Survey in the Stanford Computer Science Department. I also write about various aspects of graduate student life in my blog. My non-research aspirations include qualifying for the Boston Marathon, learning to play the cello, and cultivating my own supply of fresh kale.

email: jhshen [at] stanford [dot] edu

The Data Addition Dilemma

JH Shen, Inioluwa Deborah Raji, Irene Y. Chen
Machine Learning For Health Care 2024 (MLHC)
Adding data is not always better, we formalize a practical data accumulation model to explain why.

Paper Code

Multigroup Robustness

with Lunja Hu and Charlotte Peale (αβ)
International Conference on Machine Learning 2024 (ICML)
When data comes from different sources, robustness guarantees for subgroups should depend on how much corruption occurs within that subgroup.

Paper Code

Unifying Corroborative and Contributive Attributions in Large Language Models

Teddi Worledge*, JH Shen*, Nicole Meister, Caleb Winston, Carlos Guestrin
IEEE Conference on Secure and Trustworthy Machine Learning 2024 (SaTML)
Modern LLM applications require both factuality and training data attributions, we introduce a unified model for both.

Paper

Dissenting Explanations: Leveraging Disagreement to Reduce Model Overreliance

with Omer Reingold and Aditi Talati (αβ)
AAAI Conference on Artificial Intelligence 2024 (AAAI)
When explanations can argue both for and against a model decision, humans make better decisions.