Differentiable Physics (for Deep Learning), Overview Talk by Nils Thuerey скачать в хорошем качестве

Differentiable Physics (for Deep Learning), Overview Talk by Nils Thuerey

In this talk Nils explains recent research works that shows how to employ differentiable PDE solvers for deep learning. A central aim here is to improve the outcomes of numerical simulations. Results for reducing the numerical error in iterative solvers, and long-term control of physical systems are demonstrated. Source code is available at: https://github.com/tum-pbs/PhiFlow The two research papers discussed in the talk can be found at: https://ge.in.tum.de/publications/202... & https://arxiv.org/pdf/2007.00016 https://ge.in.tum.de/publications/202... & https://arxiv.org/pdf/2001.07457 Addendum: the missing video at 33:45 is https://ge.in.tum.de/download/2020-ic... , it shows a much more "intuitive" solution compared to the single-case optimization.