Handwritten Digit Recognition with Pure Numpy Coded Neural Network and Residual Block on MNIST Dataset

The Handwritten Digit Recognition.ipynb maintains the structure of the project.

Introduction

Handwritten digit recognition is a good first-lesson on image recognition for beginners.

The MNIST dataset consists 60000 training examples and 10000 test examples, of digits 0-9, can be downloaded from Yann Lecun's website.

In this project we will:

• Implement a 2-layer neural network.
• Compute the forward pass and backward pass with svm loss.
• Use the normlizaiton to help network converge.
• Use a stochastic gradient descent to optimize the loss function.
• Try a new type of neural network: Resnet, and apply the methodology to a three layer neural network.

Examples

Here are some digit examples from training set:

Here is the performance of our model on test set:

The test set accuracy is about 98.2%.

LICENSE:

MIT