Abstract:Within the family of explainable machine-learning, we present Fredholm neural networks (Fredholm NNs), deep neural networks (DNNs) which replicate fixed point iterations for the solution of linear and nonlinear Fredholm Integral Equations (FIE) of the second kind. Applications of FIEs include the solution of ordinary, as well as partial differential equations (ODEs, PDEs) and many more. We first prove that Fredholm NNs provide accurate solutions. We then provide insight into the values of the hyperparameters and trainable/explainable weights and biases of the DNN, by directly connecting their values to the underlying mathematical theory. For our illustrations, we use Fredholm NNs to solve both linear and nonlinear problems, including elliptic PDEs and boundary value problems. We show that the proposed scheme achieves significant numerical approximation accuracy across both the domain and boundary. The proposed methodology provides insight into the connection between neural networks and classical numerical methods, and we posit that it can have applications in fields such as Uncertainty Quantification (UQ) and explainable artificial intelligence (XAI). Thus, we believe that it will trigger further advances in the intersection between scientific machine learning and numerical analysis.

What problem does this paper attempt to address?

The core problem that this paper attempts to solve is to explore and establish the connection between Fredholm integral equations (FIE) and deep neural networks (DNN), thereby developing a new and interpretable machine - learning method, namely Fredholm neural networks (Fredholm NNs). Specifically: 1. **Problem Background**: - Fredholm integral equations (FIE) are an important class of mathematical tools, widely used in solving ordinary differential equations (ODE), partial differential equations (PDE), and other fields. - Traditional numerical methods such as the Nyström method can be used to solve FIE, but these methods may lack interpretability or have low computational efficiency in some cases. 2. **Research Motivation**: - Although existing deep neural networks (DNN) perform well in many fields, when dealing with integral equations, they usually need to optimize weights and biases through standard training steps, which may cause the model to become a "black box" and be difficult to interpret. - The author hopes to use the fixed - point iteration theory to transform the solution process of FIE into the structure of DNN, thereby improving the interpretability and accuracy of the model. 3. **Research Objectives**: - Develop a DNN architecture based on fixed - point iteration (i.e., Fredholm neural networks) that can directly solve linear and nonlinear FIE without performing standard training steps. - By connecting the mathematical theory of FIE and the internal calculations of DNN, provide an intuitive explanation for the weights and biases in DNN, making it more interpretable. - Demonstrate the advantages of this method in numerical approximation accuracy and explore its potential applications in fields such as uncertainty quantification (UQ) and explainable artificial intelligence (XAI). 4. **Main Contributions**: - Propose the concept of Fredholm neural networks (Fredholm NNs), which is a DNN specifically designed to solve FIE. - Prove that Fredholm NNs can provide accurate solutions to FIE, and give explicit mathematical meanings to the weights and biases in the network through the fixed - point iteration theory. - Demonstrate the effectiveness and efficiency of this method in solving linear and nonlinear FIE through specific examples. In summary, this paper aims to develop a new and interpretable machine - learning method by combining the mathematical theory of FIE and the computational power of DNN to solve the challenges faced by traditional numerical methods and standard DNN in solving FIE.

Fredholm Neural Networks

Deep Neural Network Solutions for Oscillatory Fredholm Integral Equations

Fredholm integral equations for function approximation and the training of neural networks

ODEN: A Framework to Solve Ordinary Differential Equations using Artificial Neural Networks

Solving Fredholm Integral Equations Using Deep Learning

Neural Network As a Function Approximator and Its Application in Solving Differential Equations

A feedforward neural network framework for approximating the solutions to nonlinear ordinary differential equations

Fredholm Integral Equations Neural Operator (FIE-NO) for Data-Driven Boundary Value Problems

On the feed-forward neural network for analyzing pantograph equations

A Deep Learning Neural Network Framework for Solving Singular Nonlinear Ordinary Differential Equations

dNNsolve: an efficient NN-based PDE solver

An Artificial Neural Network Approach for Solving Space Fractional Differential Equations

Finite Difference Neural Networks: Fast Prediction of Partial Differential Equations

Some elliptic second order problems and neural network solutions: Existence and error estimates

PFNN: A Penalty-Free Neural Network Method for Solving a Class of Second-Order Boundary-Value Problems on Complex Geometries

Generalized Lagrangian Neural Networks

Fourier neural networks as function approximators and differential equation solvers

Numerical solution for initial and boundary value problems of high-order ordinary differential equations using Hermite neural network algorithm with improved extreme learning machine

Physics-informed Neural Networks for Functional Differential Equations: Cylindrical Approximation and Its Convergence Guarantees

Solving Partial Differential Equations Using Artificial Neural Networks

Application of Intelligent Paradigm through Neural Networks for Numerical Solution of Multiorder Fractional Differential Equations