Differential Equations as Neural Network Representations || Oct 31, 2025 скачать в хорошем качестве

Differential Equations as Neural Network Representations || Oct 31, 2025

Speaker, institute & title 1) Adeel Pervez, Institute of Science and Technology Austria (ISTA), Differential Equations as Neural Network Representations Abstract: In my talk I look at the synthesis of neural networks and differential equations. Such models have important applications in the modeling of spatio-temporal dynamics and in inverse problems in science. Past approaches have considered network-in-solver approaches which plug neural networks into classical solvers. I present our work on a solver-in-network approach which allows neural network representations that are differential equations. Embedding specialized, parallel and differentiable solvers in neural networks allows greater flexibility for dynamical modeling. As an application of these models, I discuss our work on the discovery of governing differential equations (ODEs and PDEs) from data and other applications.