Edward Berman

New convolutions improve neural networks for solving PDEs on surfaces

Researchers have identified and addressed smoothness errors in neural network models used for solving partial differential equations over surfaces. Traditional graph neural networks can suffer from oversmoothing, where …