Idea

We are developing machine learning models (namely neural networks) that gaurantee constraint statisfaction. So-called physics-informed neural networks strive to achieve a similar outcome, but do not gaurantee constraint during training or inference and have weights that are difficult to tune. We instead focus on using computationally efficient projection methods that overcome these limitations.

Location

University of Waterloo
Engineering 6, Room 5008
200 University Avenue West
Waterloo, Ontario N2L 3G1
Canada