Alexander Modell

Alexander Modell is a Research Fellow in Statistics and Machine Learning at the Department of Mathematics, Imperial College London, holding the Chapman Research Fellowship. His work focuses on AI safety through interpretability research, studying how large language models represent information and reason about the world.

He earned his PhD from the University of Bristol with the thesis "Spectral embedding of large graphs and dynamic networks" under Patrick Rubin-Delanchy's supervision, followed by postdoctoral research with Nick Heard on the NeST programme.

Modell's research bridges AI interpretability and neuroscience, using reproducible neural network experiments to uncover knowledge representation mechanisms. He investigates mathematical foundations of machine learning with emphasis on clustering, manifold learning, and analysis of dynamic networks. His work aims to detect AI misalignment pre-deployment through "artificial neurosurgeries" to modify system behavior.

Recent publications reveal strong trends in representation learning for large language models and dynamic networks, combining statistical thresholding, spectral methods, and geometric approaches to address AI safety challenges and network analysis problems.

Scientific recognition includes:

• Chapman Research Fellowship
• Notable Paper Award at AISTATS 2023 (top 32 of 496 accepted papers)

Modell actively contributes to the research community through invited talks at ISNPS, PopNets, and NeurIPS workshops, with grant support from his Chapman Fellowship enabling foundational work in AI interpretability and statistical machine learning theory.