A Novel Tensor Product-Based Neural Network for Solving Partial Differential Equations

2026-05-28 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Mathematics & Computational Sciences · Depth: Expert, quick

Summary

The Tensor Product Network (TPNet) is a novel neural architecture designed for efficient and accurate function approximation and solving Partial Differential Equations (PDEs). This network explicitly constructs solutions as a linear combination of basis functions, with coefficients determined by a direct least-squares solve, thereby eliminating the need for traditional gradient-based training. Key methodological contributions include an efficient tensor-product scheme that generates multi-dimensional basis functions from subnetwork outputs, significantly reducing model complexity and parameter count while maintaining expressivity. TPNet also incorporates a block time-marching strategy for improved computational efficiency in long-time simulations and a linear reformulation strategy to handle nonlinear PDEs by treating known nonlinear terms as sources. This structured design and deterministic fitting enable TPNet to achieve superior accuracy and shorter training times compared to conventional neural network solvers like Physics-Informed Neural Networks (PINNs).

Key takeaway

For Machine Learning Engineers and Research Scientists developing PDE solvers, TPNet offers a compelling alternative to gradient-based methods. You should consider TPNet for applications requiring high accuracy and faster training times, especially when dealing with complex multi-dimensional problems or long-time simulations. Its deterministic least-squares fitting and structured tensor-product design can significantly reduce computational overhead compared to traditional Physics-Informed Neural Networks (PINNs).

Key insights

TPNet solves PDEs and approximates functions via a tensor-product basis and direct least-squares, bypassing gradient training.

Principles

• Direct least-squares fitting enhances efficiency.
• Tensor-product basis reduces model complexity.
• Structured design outperforms iterative optimization.

Method

TPNet constructs solutions as a linear combination of tensor-product basis functions, with coefficients found via direct least-squares. It uses block time-marching and linear reformulation for nonlinear PDEs.

In practice

• Apply TPNet for faster PDE solving.
• Use tensor-product basis to reduce parameters.
• Reformulate nonlinear PDEs as linear problems.

Topics

• Tensor Product Network
• Partial Differential Equations
• Neural Network Solvers
• Least-Squares Optimization
• Function Approximation
• Model Complexity

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• Physics-Informed Neural Networks: A Didactic Derivation of the Complete Training Cycle — This guide demystifies PINN training by exposing the full algebraic cycle, from forward pass to gradient descent.
• Toward AI-Driven Digital Twins for Metropolitan Floods: A Conditional Latent Dynamics Network Surrogate of the Shallow Water Equations — CLDNet significantly accelerates metropolitan flood forecasting by using a latent neural ODE and coordinate-based decoding.
• Learning functional components of PDEs from data using neural networks — Neural networks can approximate unknown functional components within PDEs by embedding and training them with data.
• Laplace Approximation for Bayesian Tensor Network Kernel Machines — LA-TNKM provides scalable, principled uncertainty estimates for tensor network kernel machines via Laplace approximation.
• Lightweight Geometric Adaptation for Training Physics-Informed Neural Networks — A curvature-aware optimization framework improves PINN training by adaptively correcting first-order optimizers using secant information.
• Striding Across Reynolds Numbers: Representation Geometry in Neural PDE Generalisation — Representation geometry and multi-scale representations are critical for cross-Reynolds generalisation in neural PDE solvers.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.