STRUCTURE PROBING NEURAL NETWORK DEFLATION

G. U. Yiqi, Chunmei Wang, Haizhao Yang

Research output: Contribution to journalArticlepeer-review

Abstract

Deep learning is a powerful tool for solving nonlinear differential equations, but usually, only the solution corresponding to the flattest local minimizer can be found due to the implicit regularization of stochastic gradient descent. This paper proposes Structure Probing Neural Network Deflation (SP-NND) to make deep learning capable of identifying multiple solutions that are ubiquitous and important in nonlinear physical models. First, we introduce deflation operators built with known solutions to make known solutions no longer local minimizers of the optimization energy landscape. Second, to facilitate the convergence to the desired local minimizer, a structure probing technique is proposed to obtain an initial guess close to the desired local minimizer. Together with neural network structures carefully designed in this paper, the new regularized optimization can converge to new solutions efficiently. Due to the mesh-free nature of deep learning, SP-NND is capable of solving high-dimensional problems on complicated domains with multiple solutions, while existing methods focus on merely one or two-dimensional regular domains and are more expensive than SP-NND in operation counts. Numerical experiments also demonstrate that SP-NND could find more solutions than exiting methods.

Original languageEnglish
JournalUnknown Journal
StatePublished - Jul 7 2020

Keywords

  • Convergence
  • Deep Residual Method
  • High Dimension
  • Neural Networks Deflation
  • Nonlinear Differential Equations
  • Structure Probing

Fingerprint Dive into the research topics of 'STRUCTURE PROBING NEURAL NETWORK DEFLATION'. Together they form a unique fingerprint.

Cite this