PGML - Research project

The aim of this project is to develop a theoretical framework, as generic as possible, to model the integration of constraints from theoretical knowledge of physics in various classes of machine learning algorithms. A side effect of this approach is that it could contribute to develop the interpretability of the learned models since they will be based partly on known physics theories. One could also think to extend this integration of constraints in machine learning algorithms to the injection of ethical knowledge, constraints on learning bias, etc. Such a project is intended to propose a win-win approach for the fields of computer science and physics. Unlike traditional approaches where computer science uses physics for its algorithms or physics uses computer science to discover underlying models hidden in data, the project aims to enable both fields to benefit from its outcomes.
On a longer term, this project could lead to an even more ambitious project called Machine-Assisted Scientific Discovery, where Physics and Machine Learning would work together on computer-assisted scientific discovery.