Meta Analysis of Neural Network User Parameters

 

Meeting Times: TBD

Room: BS 404, Florida Atlantic University

Meet the Team:

Alex Clark – Team Leader (jamesclark2016@fau.edu)

About the Project:

This project is an analysis of the way in which Neural Network outcomes are affected by variations in network parameters.  The above network is AlexNet, and it can be tweaked with a wide variety of parameters.  To fully exhaust the combinations could take decades of network training.  This project begins with the manipulation of a few key parameters: image size, number of images, number of categories, data augmentation, and stride.  With a few possible positions for each of these variables, there are already 1,600 combinations of parameters to be tested.  We hope to run each combination at least 10 times, making this a very time consuming project that will entail 16,000 training runs.

We welcome new team members, as there is a lot of work to be shared, and this project is a good way to get practice with the basic steps of neural network setup and training.