You have to train a simple single layer Perceptron Model for recognizing the hand-written digits in the attached data set. Following is a set of guidelines for completing the assignment.

• Implement perceptron (785 inputs, 10 outputs) and perceptron learning algorithm in Python language
• Download MNIST data from class Google Drive folder.
• Preprocess MNIST data: Scale each feature to a fraction between 0 and 1 as discussed in the class.

Experiments:

You have to produce results for the following three experiments. In each experiment, train the 10 perceptrons with the following given learning rates:

Experiment 1: = 0.01

Experiment 2: =0.1

Experiment 3: = 1.0

For each learning rate:

1. Initialize perceptrons with small random weights wi[-0.05, 0.05] (chosen independently for each connection in each perceptron)

2. Run perceptron learning for 50 epochs. At each epoch (including epoch 0), compute accuracy on training and test data. (Dont change the weights while computing accuracy.)

For each experiment, plot training and test accuracy over epochs:

For each experiment, give confusion matrix for results on test data after 50 epochs of training.