We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Mathematics > Numerical Analysis

Title: Solving inverse-PDE problems with physics-aware neural networks

Abstract: We propose a novel composite framework that enables finding unknown fields in the context of inverse problems for partial differential equations (PDEs). We blend the high expressibility of deep neural networks as universal function estimators with the accuracy and reliability of existing numerical algorithms for partial differential equations. Our design brings together techniques of computational mathematics, machine learning and pattern recognition under one umbrella to seamlessly incorporate any domain-specific knowledge and insights through modeling. The network is explicitly aware of the governing physics through a hard-coded PDE solver stage; this subsequently focuses the computational load to only the discovery of the hidden fields. In addition, techniques of pattern recognition and surface reconstruction are used to further represent the unknown fields in a straightforward fashion. Most importantly, our inverse-PDE solver allows effortless integration of domain-specific knowledge about the physics of underlying fields, such as symmetries and proper basis functions. We call this approach Blended Inverse-PDE Networks (hereby dubbed BIPDE-Nets) and demonstrate its applicability on recovering the variable diffusion coefficient in Poisson problems in one and two spatial dimensions. We also show that this approach is robust to noise.
Comments: 18 pages, 8 figures, submitted to SIAM Journal on Scientific Computing
Subjects: Numerical Analysis (math.NA); Machine Learning (cs.LG); Computational Physics (physics.comp-ph)
Cite as: arXiv:2001.03608 [math.NA]
  (or arXiv:2001.03608v1 [math.NA] for this version)

Submission history

From: Samira Pakravan [view email]
[v1] Fri, 10 Jan 2020 18:46:50 GMT (2552kb,D)
[v2] Mon, 6 Apr 2020 06:16:24 GMT (4632kb,D)
[v3] Wed, 18 Nov 2020 21:47:28 GMT (5909kb,D)

Link back to: arXiv, form interface, contact.