CATT
We introduce CATT, a controlled benchmark for language-model data attribution. We identify where existing gradient-based methods break down and show how two targeted fixes substantially improve attribution.
Steven Yu
I'm an undergraduate student studying Computer Science at Northeastern University, where I conduct research in the Bau Lab with Prof. David Bau. My research interests include data attribution, influence functions, and training dynamics in language models.
I'm also currently an AI research intern at the U.S. Air Force Research Laboratory, working on ML-based analysis of persuasive narratives in media.
selected works
experience
2025–present
Undergraduate Researcher · Bau Lab, Northeastern University
Working on language-model data attribution and influence functions.
2026–present
AI Research Intern · U.S. Air Force Research Laboratory
Developing ML and LLM-based methods for detecting and classifying persuasion in media.