skip to content

Researcher: Jingwei Liang

Non-smooth optimisation has become ubiquitous in many fields including inverse problems, image science and machine learning. The need for fast optimisation methods is becoming increasingly strong due to the demands of real-world problems. However, traditional acceleration approaches have reached a bottleneck and cannot provide satisfactory results, while a universal generic acceleration framework for non-smooth optimisation remains an open problem. In this project, we study the geometry of non-smooth optimisation problems and an acceleration framework which can automatically adapt to the geometry of the problem. The project has the potential to render algorithms which can, for the very first time, achieve optimal performance.