From 205a0a593c86bdda3c7f73a9c6eaeacde10a7ea6 Mon Sep 17 00:00:00 2001 From: Yuan Yin Date: Thu, 21 Nov 2024 21:48:43 +0100 Subject: [PATCH] Create publication page for GEPS --- _publications/2024_geps.md | 76 ++++++++++++++++++++++++++++++++++++++ 1 file changed, 76 insertions(+) create mode 100644 _publications/2024_geps.md diff --git a/_publications/2024_geps.md b/_publications/2024_geps.md new file mode 100644 index 0000000..0bb1175 --- /dev/null +++ b/_publications/2024_geps.md @@ -0,0 +1,76 @@ +--- +layout: publication +title: "GEPS: Boosting Generalization in Parametric PDE Neural Solvers through Adaptive Conditioning" +image: +hide: true +category: [generalization, dynamics, prediction] +authors: Armand Kassaï Koupaï, Jorge Mifsut-Benet, Yuan Yin, Jean-Noël Vittaut, Patrick Gallinari +venue: NeurIPS +venue_long: Conference on Neural Information Processing Systems +year: 2024 +month: 12 +code_url: https://github.com/itsakk/geps +paper_url: https://arxiv.org/abs/2410.23889 +blog_url: https://geps-project.github.io +slides_url: +bib_url: +permalink: /publications/geps/ +--- + +

{{page.title}}

+ + + +

Armand Kassaï Koupaï    Jorge Mifsut-Benet    Yuan Yin    Jean-Noël Vittaut    Patrick Gallinari + + +

{{page.venue}} {{page.year}}

+ +
+

+ {% if page.paper_url %} + Paper   + {% endif %} + {% if page.code_url %} + Code    + {% endif %} + {% if page.blog_url %} + Blog    + {% endif %} + {% if page.slides_url %} + Slides   + {% endif %} + {% if page.bib_url %} + BibTeX   + {% endif %} +

+
+ + +
+ project teaser +
+ + +
+ +

Abstract

+ +

Solving parametric partial differential equations (PDEs) presents significant challenges for data-driven methods due to the sensitivity of spatio-temporal dynamics to variations in PDE parameters. Machine learning approaches often struggle to capture this variability. To address this, data-driven approaches learn parametric PDEs by sampling a very large variety of trajectories with varying PDE parameters. We first show that incorporating conditioning mechanisms for learning parametric PDEs is essential and that among them, \textit{adaptive conditioning}, allows stronger generalization. As existing adaptive conditioning methods do not scale well with respect to the number of PDE parameters, we propose GEPS, a simple adaptation mechanism to boost GEneralization in Pde Solvers via a first-order optimization and low-rank rapid adaptation of a small set of context parameters. We demonstrate the versatility of our approach for both fully data-driven and for physics-aware neural solvers. Validation performed on a whole range of spatio-temporal forecasting problems demonstrates excellent performance for generalizing to unseen conditions including initial conditions, PDE coefficients, forcing terms and solution domain.

+ +
+
+ +

BibTeX

+ + 
+@inproceedings{kassai2024geps,
+  title={GEPS: Boosting Generalization in Parametric PDE Neural Solvers through Adaptive Conditioning},
+  author={Kassaï Koupaï, Armand and Mifsut Benet, Jorge and Vittaut, Jean-Noël and Gallinari, Patrick},
+  booktitle={38th Conference on Neural Information Processing Systems (NeurIPS 2024)},
+  year={2024}
+}
+
+ + +