Artificial Neural Network created from scratch to classify hand-written numbers.
Instead of using a neural network library like TensorFlow I opted to create everything from scratch, including the Matrix class for linear algebra, to learn more about how neural networks work.
It's a three layer artificial neural network using gradient descent to train the network and forward propagation to classify data. It's trained using the MNIST database of hand written numbers. The dataset contains 60000 hand written numbers used for training and 10000 used for testing after training. I also wrote a reader to read the data, which is not saved in a standard way.
The network classifies 90% of the testing set correctly.
- Clone the repo
- Via terminal cd into Number-Recognition/Neural-Network
- Compile main.cpp:
clang++ -std=c++11 -Ofast main.cpp - Execute the compiled file:
./a.out - Follow the program instructions
