Implement the NN described on the following URL to recognize handwritten digits. http://neuralnetworksanddeeplearning.com/chap1.html#a_simple_network_to_classify_handwritten_digits Create a three-layer neural network as shown below, the input layer contains 784=28×28 neurons, the hidden layer contains 15 neurons, and the output layer contains 10 neurons represent the 10 digits’ classes. Use the MNIST data set, which contains tens of thousands of scanned images of handwritten digits, to train the network using stochastic gradient descent algorithm. Try different values of the hyper-parameters, learning rate η (0.01, 0.1, and 1), batch size (online incremental learning, mini-batch, and full batch of all instances), and epochs. Report the results achieved, classification accuracy and training time (convergence speed), and make your conclusion regarding the hyper-parameters effect on the learning process. Hint: You may use the same code on the URL above. Step-by-step description of the code is listed.
