Choose timezone
Your profile timezone:
CD Technical Meeting (ML6): GPs for Physics-informed Manifold Learning and EFIT Shape Inference from MAST images
→
Europe/London
,
Description
Machine Learning, Uncertainty Quantification and Data Science
Challenges in Physics-Informed Manifold Learning for Gaussian Processes: This talk introduces the motivation for Physics-Informed Gaussian Processes (GPs), focusing on how constraining manifold learning with differential equations impacts GP performance. Key concepts include Gaussian Processes, nonlinear dimensionality reduction (including VAEs), and Physics-Informed Neural Networks (PINNs).
Computer Vision for EFIT Shape Inference in MAST: This project explores the use of camera data from the MAST tokamak to infer plasma shape parameters, with the goal of assessing its potential as a complementary diagnostic. A computer vision pipeline is developed to extract plasma boundary features from experimental video frames, which are then used to predict equilibrium shape parameters using a Gaussian process model. The approach demonstrates the feasibility of deriving meaningful shape information from visual data alone, offering a novel, non-invasive route for plasma diagnostics.
