Half-Inverse Gradients for Physical Deep Learning

Patrick Schnell, Philipp Holl, Nils Thuerey

Recent works in deep learning have shown that integrating differentiable physics simulators into the training process can greatly improve the quality of results. Although this comb…