Efficient Shallow Neural Network to Rapidly Solve Tough PDEs
Efficient Shallow Neural Network to Rapidly Solve Tough PDEs
Efficient Shallow Neural Network to Solve Surface PDEs and PDEs with Singularities
Tags: National Yang Ming Chiao Tung University, Taiwan, Computing Technology, Science & Exploration
Researchers at NYCU have developed a shallow neural network (DCSNN) to efficiently solve partial differential equations (PDEs), including those with discontinuities and singularities. This network captures jump discontinuities sharply, uses a single hidden layer, and operates without a mesh, making it suitable for complex and irregular domains. Applications include fluid-structure interaction problems in computational fluid mechanics, especially in microfluidic systems and soft matter physics. The DCSNN model achieves comparable accuracy to traditional methods while being easier to train. The study highlights the potential of machine learning in addressing scientific computing challenges.
IP Type or Form Factor: Design; Software & Algorithm
TRL: Not specified
Industry or Tech Area: Computing Architecture; Nuclear Energy & Physics
