Dr Fred Roosta-Khorasani and his research team design, analyse and implement novel optimisation algorithms for training modern machine learning problems.

The underlying approaches draw from randomised techniques, statistics, non-convex non-linear analysis, geometry, and parallel/distributed computing.

Dr Fred Roosta is also interested in using novel ML models for scientific discovery, in particular the design of innovative physicsaware deep learning models and the development of their statistical learning theory.

Project members

Professor Fred Roosta

School of Mathematics and Physics