ZICQ
中 Log in / Sign up
Newsroom Research & Papers #Data-Free Training #Physics-Informed Neural Networks #Level-Set Interface Advection #arXiv #Data-Scarce

arXiv Introduces Data-Free Physics-Informed Neural Operator for Level-Set Interface Advection

Avatar of Mr.Xu

By Mr.Xu Compiled & Reviewed by Editorial

Published: · 9 views

中文阅读 (Chinese) English Version
Co-reading Select any sentence in the article to highlight, annotate, or copy a quote card. Join discussion ↓

Summary:arXiv introduces a novel data-free physics-informed neural operator designed for level-set interface advection problems. Unlike traditional supervised learning methods that rely on reference solutions, this operator is trained using only the transport residual and geometric constraints, eliminating the need for reference data. Experimental results demonstrate that while it slightly underperforms the supervised baseline in some benchmarks, it excels in specific constrained conditions, such as ach


Background and Motivation

Traditional neural operators for interfacial problems rely on reference solutions generated by the target solver for training. However, this approach is limited in scenarios where data is scarce or obtaining reference solutions is difficult. To address this, researchers have developed a data-free physics-informed neural operator specifically for level-set interface advection problems.

Technical Highlights

  1. Data-Free Training Mechanism: The method uses only the transport residual and geometric constraints for training, completely eliminating the need for reference solutions.
  2. Spatiotemporal Fourier Backbone: The spatiotemporal Fourier backbone allows the operator to emit the entire trajectory in one pass, avoiding the competition between the anchoring term and the residual that arises in traditional methods.
  3. Performance Evaluation: In the single vortex test, the data-free operator achieves a relative L2 error of 1.614% compared to the supervised baseline's 0.369%; in the solid-body rotation test, the corresponding figures are 3.804% and 2.576%.
  4. Advantage Under Constraints: In regions where the eikonal constraint is valid, the data-free operator conserves enclosed area 2.7 times better than the supervised baseline, demonstrating higher physical consistency.

Applications and Benefits

This method is particularly suitable for scenarios where data is scarce or obtaining reference solutions is difficult, such as modeling and simulating complex physical systems. Its ability to perform without reference solutions allows it to surpass traditional supervised learning methods in certain specific conditions.

Future Directions

In the future, researchers can further optimize this method and explore its potential in more complex physical problems. Additionally, combining it with hybrid methods or training with a small number of reference solutions might further enhance its performance.


Source: ArXiv Machine Learning (cs.LG) (2026-10-01)

— END —

Tags: #Data-Free Training #Physics-Informed Neural Networks #Level-Set Interface Advection #arXiv #Data-Scarce

Editorial & Fact-Checking Note: This article is compiled from primary research, official release documentation, and source papers by the ZICQ Newsroom pipeline with automated entity verification and human editorial review. If you notice any technical inaccuracy, please submit a correction via our corrections policy or email our editorial desk directly.

What's your take on this story?
🚀 Explore & Submit Works

Community Comments

Loading live comments and annotations…