Scribd
Upload
18 views

Blue Line - Analytical Solution, Orange Line - Finite Differences, Green Line - Neural Network Solution

This document summarizes using neural networks to solve ordinary and second order differential equations. A simple neural network with one hidden layer was able to approximate the solution to an ordinary differential equation with an MSE of 0.149 after 1000 training iterations. When extended to a second order differential equation, the neural network achieved an MSE of 0.42 after 100 iterations. References for further information are also provided.

Uploaded by

Sayee Natarajan
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
18 views

Blue Line - Analytical Solution, Orange Line - Finite Differences, Green Line - Neural Network Solution

This document summarizes using neural networks to solve ordinary and second order differential equations. A simple neural network with one hidden layer was able to approximate the solution to an ordinary differential equation with an MSE of 0.149 after 1000 training iterations. When extended to a second order differential equation, the neural network achieved an MSE of 0.42 after 100 iterations. References for further information are also provided.

Uploaded by

Sayee Natarajan
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 2

Ordinary Differential Equation:

First, we approximated the simple ordinary differential equation using a simple


neural architecture that consists of a single input neuron, one hidden layer and one
output neuron to predict value of a solution in exact point on a grid.
For optimization purpose, we used “Autograd” which is a modern tool for automatic
differentiation. We found a numerical solution on a grid, approximating it with
neural network. Result of training neural network for 1000 iterations with final mean
squared error (MSE) of 0.149 you can see on the Fig 1.

Blue line — analytical solution, orange line — finite differences, green line — neural network solution

Fig.1: Optimization of Ordinary Differential Equation using Neural Network


Second Order Differential Equation:

After extending our solution to second order differential equation, we obtained an


MSE of 0.42 after 100 iterations (see Fig 2).

Blue line — analytical solution, orange line — neural network

Fig.2: Optimization of Second Order Differential Equation using Neural Network

References:
1. https://becominghuman.ai/neural-networks-for-solving-differential-
equations-fa230ac5e04c
2. http://cs229.stanford.edu/proj2013/ChiaramonteKiener-
SolvingDifferentialEquationsUsingNeuralNetworks.pdf

You might also like