Sekh Mainul Islam

About me

Hi, I am a third-year PhD student at the CopeNLU group, Department of Computer Science, University of Copenhagen. I am also affiliated with the Pioneer Centre for Artificial Intelligence. I work with my PhD supervisors Prof. Isabelle Augenstein and Prof. Pepa Atanasova on Interpreting how LLMs encode factual and personalization concepts.

My research interest in Mechanistic Interpretability focuses on how LLMs encode different knowledge interactions into their model space and how different components conceptually associate with their biased behaviours. Recently, I started working on understanding how Mixture-of-Experts LLMs encode personalized concepts into distinct routing paths, offering better interpretability than dense LLMs.

News