MNIST Digit recognition Neural Network [Perceptron]

---

• Original:

  It is the neural network that I first developed, my first attempt. 784-neuron INPUT layer, arbitrary number of neurons in 2 HIDDEN layers, and 10-neuron in the OUTPUT layer.

• 1d:

  It is my second attempt to optimize the network by transforming 2d matrices into 1d arrays. Spoiler: it didn't increase the training speed much (by less than 3%).

• Enhanced:

  It is the neural network that I worked on with ClaudeAI. It has momentum, L2 regulation as improvements.

---

RUN (train):

Clone the repository:

git clone https://github.com/YerdosNar/digitNN.git

You can find run script in every directory:

./run

You will be prompted to input the size of each layer. And training starts.

---

Test:

Just run ./test_nn for testing with test data or ./draw_test for testing by drawing your own digits.

---

Files:

Train data, test data, font files are inside files directory.