Deep Neural Network

Overview

< source: galliot.us >

A simple Deep Neural Network that classifies MNIST digit images(28 x 28 pixels) in one-hot encoded form. The implementation is more focused on readability and maintainability for educational purpose rather than performance, following(trying to 😅) principles of object oriented programming.

Structure

Possible Improvements

• #2 Learning Rate Optimizer
• #3 Ridge Regularization
• #4 Parameter Initializer

Background

For theoritical background, please refer to those pdf docs directly exported from my personal Notion. They are largely based on the books Deep Learning: Foundations and Concepts by Hugh Bishop and Understanding Deep Learning By Simon J.D. Prince.

1. Classification
2. Deep Neural Networks
3. Gradient Descent
4. Backpropagation

License

MIT