fairfolio
Saved Projects

Testing a Variety of Methods for Training a Neural Network to Determine Which One Achieves the Lowest Error Rate Fastest?

ISEF · 2020 Robotics and Intelligent Machines

Sanders, Elias · Idaho Falls High School

Overview

In my project, I wanted to find out what method, out of the ones that I tested, would best train a neural network. Testing these methods is important because neural networks are used on a wide range of things that can affect daily life. Consequently, quickly and efficiently training a neural network can also impact daily life. Some uses of neural networks are in games, advertisement predictions, therapy, and some can even provide dietary advice. The way that I defined the best method was which one achieved the lowest error rate in the least amount of training iterations. I predicted that backwards propagation would best train a neural network. The methods that I tested are as follows: Backwards propagation using a large training data set, backwards propagation with a smaller training data-set, random splicing using a large training data-set, and random splicing using a smaller training data-set. In order to start experimentation, I wrote a library for making neural networks and populations of neural networks. After that, I wrote several scripts to run the different methods along with one script that runs all of them in order. Each network trains for 1,000 iterations --this is because at this point the error slopes should be pretty apparent, if this isn't enough, I would use 10,000, however, it did end up working-- and every 100 iterations, it saves its error on a new data-set to a file for graphing later. Once all of them were done, I graphed the errors with the error as the y-axis and the iterations as the x-axis. I came to the conclusion that backwards propagation using a large training data-set performed the best by achieving a low error rate before any other method tested.

Competition history

Resources

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

Source: Regeneron International Science and Engineering Fair

Save projects to your library

Sign in with Google to keep track of projects you find interesting, organized into folders. An account also raises your daily allowance for “Has this been done?”, and lets you create a key for the MCP server with a much higher limit than anonymous use. Browsing stays public.

Continue with Google