SIAM Journal on Scientific Computing· 2021Q1
Understanding and Mitigating Gradient Flow Pathologies in Physics-Informed Neural Networks
- 1,697citations
- Q1SCImago
- 2021year
Short summary
This paper identifies and quantifies pathological gradient flows in Physics-Informed Neural Networks (PINNs) that hinder training, particularly for stiff dynamics, and proposes a novel adaptive activation function that significantly improves convergence and accuracy.
AI-generated from the title and abstract; the full text is not read.
Key points
- Identifies pathological gradient flows as a key limitation in PINN training, particularly for stiff dynamics.
- Quantifies the impact of these pathologies on convergence and accuracy.
- Proposes a novel adaptive activation function to mitigate gradient flow issues.
- Demonstrates significant improvements in convergence and accuracy using the proposed method on benchmark problems.
AI-generated from the title and abstract; the full text is not read.
Abstract
Related DatabasesWeb of Science You must be logged in with an active subscription to view this.Article DataHistorySubmitted: 12 February 2020Accepted: 25 May 2021Published online: 09 September 2021Keywordsdeep learning, differential equations, optimization, stiff dynamics, computational physicsAMS Subject Headings68T99, 65M99, 68U20Publication DataISSN (print): 1064-8275ISSN (online): 1095-7197Publisher: Society for Industrial and Applied MathematicsCODEN: sjoce3
The authors' abstract, as published at the source. SIAM Journal on Scientific Computing, 2021 · DOI ↗
Continue with a free account
Ask the paper: 3 free questions a day about this paper; save it, get its citation, new summaries every day for your field. Takeaways are Premium.
Continue free on the webSign in with Google or Apple; no card needed. You come back to this paper.
On your phone:
Field: Statistical and Nonlinear Physics
Statistical and Nonlinear PhysicsPhysics and Astronomy