An elementary Python neural network capable of recognizing handwritten digits using the MNIST dataset.
This project was built from scratch (no deep-learning frameworks) to better understand how neural networks and backpropagation work under the hood.
60,000 grayscale images of handwritten digits from 0 to 9, 28x28 pixels each. [1]
The default neural network structure is:
- Input layer:
- 784 neurons (one per pixel)
- Hidden layers:
- 128 neurons
- 30 neurons
- Output layer:
- 10 neurons (one per digit)
The network is trained using the backpropagation algorithm, which updates weights and biases via gradient descent.
python3 scr/app.py
Make sure to have all requirements beforehand.
pip install requiremets.txt