Theses

📚 Available Theses

When a model unlearns a concept in English, can it still produce that concept if prompted in another language? This thesis empirically tests the cross-lingual robustness of machine unlearning methods on LLMs.

Fairness-Aware Machine Unlearning

Can selectively forgetting training data improve or harm model fairness? This thesis investigates the interplay between machine unlearning and fairness metrics across multiple domains using the ERASURE framework.

ERASURE: Improving Usability and Community Adoption of a Machine Unlearning Framework

A thesis focused on debugging, refactoring, and documenting the ERASURE machine unlearning framework, lowering the skill floor so that researchers can adopt it without reading the source code.