Recognition-for-Handwriting-Numerals

Implement DNN & Back-Propagation algorithm with only "Numpy" package.

Dataset

The numeral images used in this study is from 👉 https://github.com/bat67/TibetanMNIST

Execution & Overall Structure of system

1. Preprocessing : 2D Images --> 1D input data

2. Deep Neural Network for Classification : Back-Propagation algorithm with Numpy

   python3 DNN.py
   

Preprocessing

• Reshape the 2D images into 1D input data
  

Deep Neural Network for Classification

Introduction of DNN please refer to 👉 https://www.nature.com/articles/nature14539

• DNN Forward-propagation

  

• Computation for Forward-propagation

  

• DNN Backward-propagation
  Update the Weigt Matrix from back

  

• Computation for Backward-propagation

  

Overfitting & Regularization

In this project, Regularization term is introduced to avoid Overfitting problem.
Regularization term will impose the penalty when the complexity of DNN model increases.

Details of L1 & L2 Regularization please refer to 👉 https://iopscience.iop.org/article/10.1088/1742-6596/1168/2/022022

Experimental Results

• **Visualization of latent features with different training epochs **

• Distribution after sigmoid activation function
  

• Confusion Matrix