Frequency Shift Physics-Informed Extreme Learning Machine for Solving High-Frequency Partial Differential Equations

2026-07-02 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Mathematics & Computational Sciences · Depth: Expert, quick

Summary

The Frequency Shift Physics-Informed Extreme Learning Machine (FS-PIELM) framework is introduced to overcome spectral bias in solving high-frequency partial differential equations (PDEs) using physics-informed machine learning. This method employs an additive weight initialization mechanism that translates the mean of the Gaussian weight distribution while maintaining a fixed variance of unity, avoiding variance amplification seen in scaling-based approaches. Two variants, FS-PIELM-L and FS-PIELM-G, assign independent or grouped frequency magnitudes to neurons, respectively. Theoretical analysis confirms the framework's bounded frequency variance, approaching unity regardless of target frequency. Preserving the computational efficiency of extreme learning machines, FS-PIELM requires only a single linear solve. Experiments across seven benchmark problems, including Helmholtz and wave equations, demonstrated that the linear variant achieved superior accuracy in six cases, showing improvements of one to nearly five orders of magnitude over existing PIELM methods. Code and data will be publicly available at https://github.com/xgxgnpu/Physics-informed-vibe-coding/tree/main/FS-PIELM, published on 2026-07-02.

Key takeaway

For Machine Learning Engineers developing physics-informed models for high-frequency partial differential equations, you should consider adopting the FS-PIELM framework. Its additive weight initialization significantly improves accuracy by one to five orders of magnitude over existing PIELM variants, effectively mitigating spectral bias. This approach maintains computational efficiency, requiring only a single linear solve, making it a powerful tool for complex scientific computing tasks.

Key insights

FS-PIELM uses an additive weight initialization to mitigate spectral bias in high-frequency PDE solutions.

Principles

• Additive weight initialization prevents variance amplification.
• Bounded frequency variance improves high-frequency learning.
• Preserving ELM efficiency allows fast solutions.

Method

FS-PIELM initializes weights by translating the mean of a Gaussian distribution, keeping variance at unity. It offers linear (FS-PIELM-L) and grouped (FS-PIELM-G) neuron frequency assignments.

In practice

• Apply FS-PIELM-L for high-accuracy PDE solutions.
• Use FS-PIELM for computationally efficient physics-informed models.
• Explore FS-PIELM-G for robustness in complex geometries.

Topics

• Physics-Informed Machine Learning
• Partial Differential Equations
• Extreme Learning Machines
• Spectral Bias
• Weight Initialization
• High-Frequency PDEs

Code references

• xgxgnpu/Physics-informed-vibe-coding

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

Related on AIssential

• Learning universal approximations for partial differential equations with Physics-Informed Broad Learning System — PIBLS offers a backpropagation-free, least-squares optimization approach for PDEs, achieving faster, more accurate solutions than PINNs.
• LFNO: Bridging Laplace and Fourier via Transient-Steady Decomposition — LFNO unifies Laplace and Fourier operators to model dynamical systems across transient and steady-state regimes.
• Engineering Hybrid Physics-Informed Neural Networks for Next-Generation Electricity Systems: A State-of-the-Art Review — Physics-informed machine learning integrates domain physics into models, enhancing accuracy and efficiency for data-scarce electricity systems.
• A Machine Learning Approach to the Nirenberg Problem — A physics-informed neural network can effectively solve the Nirenberg problem and distinguish between realisable and non-realisable curvatures.
• JA-SIREN: Deterministic Initialization for Sinusoidal Networks via Spectral Matching — Deterministic initialization via spectral matching for sinusoidal networks significantly enhances INR reproducibility and performance.
• Physics-Guided Dimension Reduction for Simulation-Free Operator Learning of Stiff Differential--Algebraic Systems — An extended Newton implicit layer enables simulation-free, accurate operator learning for stiff DAEs by enforcing constraints and reducing dimensions.

Open in AIssential →

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