ICSC and Spoke 2 - Where Are We Now?

An API for training Physics-Informed Kolmogorov-Arnold Networks using Nvidia Modulus SYM

Not scheduled

20m

Conference Room (Physics Dept and INFN, Catania)

Innovation Grants

Speaker

Alessandro Bombini (Istituto Nazionale di Fisica Nucleare)

Description

In this contribution we discuss the novel neural network architecture, based on the Kolmogorov-Arnold Representation Theorem, dubbed "Kolmogorov-Arnold Network" (KAN), its variants (like the Chebyshev-KAN, the Jacobi-KAN, the FastKAN), and its applications to numerical resolution of PDEs via the Physics-Informed Neural Network (PINN) framework.

We discuss our implementation of the API for using seamlessly these network architectures within a widely adopted open source package for PINN, Nvidia Modulus SYM.

Presentation materials

Slides: Spoke2 Meeting: Towards virtual painting recolouring using Vision Transformer on X-Ray Fluorescence datacubes