Novel Machine Learning Approach Solves Forward and Inverse Problems

Strategic Insights -

Researchers from the University of Texas Austin have proposed a new model-constrained Tikhonov autoencoder neural network framework, called TAEN, capable of learning both forward and inverse surrogate models using a single arbitrary observational sample. This framework, TAEN, is designed to address the challenge of addressing scarce data regimes and overfitting issues in machine learning approaches. The research team, led by Hai Van Nguyen, has validated the TAEN approach through extensive numerical experiments on two challenging inverse problems: 2D heat conductivity inversion and initial condition reconstruction for time-dependent 2D Navier-Stokes equations.

Key Takeaways:

  • The TAEN framework is capable of learning both forward and inverse surrogate models using a single arbitrary observational sample.
  • TAEN addresses the challenge of addressing scarce data regimes and overfitting issues in machine learning approaches.
  • The research team developed comprehensive theoretical foundations, including forward and inverse inference error bounds for the proposed approach for linear cases.
  • The TAEN approach delivers accuracy comparable to traditional Tikhonov solvers and numerical forward solvers for both inverse and forward problems, respectively.
  • TAEN provides orders of magnitude computational speedups.
  • The research has been peer-reviewed and published in the journal "Computer Methods In Applied Mechanics and Engineering".

Statistics:

  • The research team conducted extensive numerical experiments on two challenging inverse problems: 2D heat conductivity inversion and initial condition reconstruction for time-dependent 2D Navier-Stokes equations.
  • The TAEN approach achieved accuracy comparable to traditional Tikhonov solvers and numerical forward solvers for both inverse and forward problems, respectively.
  • The TAEN approach delivers orders of magnitude computational speedups.

Sources:

  • Taen: a Model-constrained Tikhonov Autoencoder Network for Forward and Inverse Problems. Computer Methods In Applied Mechanics and Engineering, 2025;446.
  • NewsRx. Researchers from University of Texas Austin Report Details of New Studies and Findings in the Area of Machine Learning (Taen: a Model-constrained Tikhonov Autoencoder Network for Forward and Inverse Problems). Information Technology Newsweekly. November 4, 2025; p 787.