Choose timezone
Your profile timezone:
B&A Seminars
Differentiable Beam Dynamics Simulation Codes and their Applications
by
→
America/Chicago
Description
Abstract: Differentiable simulation frameworks enable the computation of derivatives of simulation outputs with respect to input parameters via automatic differentiation (AD). This capability allows for efficient gradient-based optimization and seamless integration with modern machine learning and AI methods. In accelerator physics, such tools can greatly reduce the computational cost of solving high-dimensional optimization problems.
This talk will introduce the core principles of differentiable simulation and show the existing differentiable accelerator simulation packages. Several ML/AI application examples using the PyTorch-based Cheetah framework will be presented, including lattice optimization, data-driven model calibration, and phase space reconstruction.