Handwritten-Digit_Recognization

Handwritten Digit Recognition (implemented with Feedforward Fully Connected)

Computational Intelligence Course 1st Project

In this project a simple neural network is implemented in 2 different ways (MNIST Dataset is used for training and testing):

1. Without vectorization
2. With vectorization

At the end an Adversarial attack has taken place on images (shifting the images). SGD with momentum is implemented as well.

Steps:

1. Reading from dataset and testing it

2. Calculate accuracy(feedforward)

3. Backpropagation

4. Vectorization and increase number of epoch

5. Calculate accuracy in train test and test set

Extra Point

6. Adversarial attack(shifting the images)

|

Calculate accuracy in train test and test set

7. Using another activation function(tanh)

Calculate accuracy in train test and test set

Check project description (in persian): here
Project report (in persian): here