MS1-11 - Data driven and Physics informed methods for solving Partial Differential Equations

S. Yadav (Computational and Data Sciences, Indian Institute of Science) *, I. Sood

Data driven techniques offer new insights in age-old problem of numerically solving intractable Partial Differential Equations(PDE). What sets them apart from other conventional techniques such as Finite element method is the massive parallelization and the universal non-linear approximation capabilities. This mini-symposium aims to provide a platform for disseminating knowledge about integration of data driven techniques and conventional solvers for PDEs. The presenters will be able to discuss supervised, semi-supervised and unsupervised techniques for approximating solutions for time-dependent, convection-dominated PDEs and turbulent flows. Further, a special focus will be on the development of software packages integrating conventional PDE solvers with general Artificial Intelligence pipeline. The speakers will present about the following topics: • Recurrent Neural Networks for time dependent problems • Variational Auto-Encoder based Super Resolution for turbulent flows • Convolutional Neural Networks for solving Partial Differential Equations • Principal Component analysis for inverse problems • Deep learning based PDE solvers