Yufei (Celina) Chen

Hi! I’m Yufei (Celina) Chen, an incoming master’s student at the University of Toronto, where I recently completed my Computer Science Specialist degree. I’m also a researcher at the Vector Institute.

I am broadly interested in machine learning, with particular focus on:

• Agentic security
• Data Privacy
• Mechanistic interpretability

Current work

I am a researcher advised by Professor Nicolas Papernot, examining how models can be adapted under data scarcity and sensitivity, particularly when retraining is constrained by privacy or uneven data availability. One line of this work studies differentially private data mixing, where mixtures of public datasets are learned to compensate for limited private data without consuming additional privacy budget.

I am also drawn to interpretability and representation learning, and how internal representations shape model behavior after training. I have investigated model merging through representation alignment, studying when and how aligning internal feature spaces improves mergeability while preserving task-specific structure. A brief summary of that work is available here.

Education

• University of Toronto
  B.Sc. (Undergraduate), Computer Science Specialist
  2023–2026

This website documents my research, projects, and academic activities.