Login

The Computing Division Technical Meetings are a platform for

  • presenting Pinboard papers under review for journal or conference publication,
  • inviting speakers who are current or prospective UKAEA collaborators at external organisations,
  • presenting work done in PhD projects funded by or co-supervised by UKAEA,
  • presenting work done during summer placements or other secondments to UKAEA.

 

If you would like to invite a speaker on a topic that would be of interest to one or more Units within the Computing Division, but is not currently collaborating on a UKAEA project, please consider nominating them for a Computing Division Cross-Disciplinary Seminar.

These meetings are normally recorded. Recordings of past meetings can be found here:

CD Technical Meetings Archive on UKAEA Sharepoint

CD Technical Meeting (ML1): Calibrated Physics-Informed Uncertainty Quantification for Neural PDE Solvers

→ Europe/London
Vignesh Gopakumar (UKAEA)
Description
Machine Learning, Uncertainty Quantification and Data Science 

 

Join Meeting Now (Teams)
Paper: Calibrated Physics-Informed Uncertainty Quantification
    • 1
      Talk: Calibrated Physics-Informed Uncertainty Quantification for Neural PDE Solvers

      Abstract:

      Neural PDEs offer efficient alternatives to computationally expensive numerical PDE solvers for simulating complex physical systems. However, their lack of robust uncertainty quantification (UQ) limits deployment in critical applications. We introduce a model-agnostic, physics-informed conformal prediction (CP) framework that provides guaranteed uncertainty estimates without requiring labelled data. By utilising a physics-based approach, we are able to quantify and calibrate the model's inconsistencies with the PDE rather than the uncertainty arising from the data. Our approach uses convolutional layers as finite-difference stencils and leverages physics residual errors as nonconformity scores, enabling data-free UQ with marginal and joint coverage guarantees across prediction domains for a range of complex PDEs. We further validate the efficacy of our method on neural PDE models for plasma modelling and shot design in fusion reactors.

      Speaker: Vignesh Gopakumar (UKAEA)
    • 2
      Q&A
      Speaker: Vignesh Gopakumar (UKAEA)
Powered by Indico v3.3.7