Convolutional Neural Networks

May 2020

Introduction

In this project, we built a Convolutional Neural Network (CNN) for solving the problem of handwritten digit recognition. The constraints of the project were to choose an advanced machine learning technique, and make an interactive demonstration that could introduce our peers to the technique. For demonstrative purposes, our CNN consists of only 1 convolutional layer, an ReLu activation layer, a max pooling layer, and a fully connected neural network with a single hidden layer. Though a true CNN could achieve much better results, the simple structure we chose to illustrate these concepts resulted in an 8% error rate when classifying the testing dataset.

Read the full project report here.

Demonstration (Activities)

To run the interactive activities, you must have Jupyter notbook installed on your machine.