jyaacoub / cnn_mnist Goto Github PK

This is a project I started in June 2019 (coming out of my first year in university) and completed in July to introduce myself to deep learning and how neural networks work. I learned all I needed to start this project from 3Blue1Brown's playlist on the topic (https://www.3blue1brown.com/neural-networks), and by reading through Michael Nielsen's book on Neural Networks and Deep Learning (http://neuralnetworksanddeeplearning.com/).

This is a Convolutional Neural Network I built to detect numbers from a webcam. It is all programmed in Java from scratch using only a linear algebra library to help with matrix multiplication and OpenCV to take in pixel values from my webcam. From knowing nothing about artificial intelligence, I taught myself how to do this in a month through online research and taking extensive notes. I even built a fully functional GUI for it that displayed the camera view and the number it read in real-time.

NETWORK: CNN_mnist/src/main/java/ConnectedNetwork/

• Network: This is the network class where a CNN is stored for training and testing.
• networkMain: This is the main class for operations to be done from the networkOperations file.
• networkOperations: This file contains all the methods that can be performed on a network in a clean manner so that it can be called in the networkMain file (methods such as training, testing, getting serialized network, etc..).

DATA: CNN_mnist/src/main/java/MNISTReading/

• MnistDataReader: class to read and parse through the Mnist data to provide meaningful inputs and labels for our network.
• MnistMatrix: class that holds the actual data for each image.

GUI: CNN_mnist/src/main/java/GUI/

• Display: This is the file to run if you want to see the network work in real-time. I have already included some serialized networks that have been trained to be used for this purpose.