Objective: Implementation of handwritten number recognition by CNN.
This is our 3rd exercise in a series that deal with the MNIST Database (http://yann.lecun.com/exdb/mnist/). The MNIST Database is a collection of samples of handwritten digits from many people, originally collected by the National Institute of Standards and Technology (NIST), and modified to be more easily analyzed computationally.
Read: “How to develop a CNN (convolutional neural nets) for handwritten digit classification”.
(https://machinelearningmastery.com/how-to-develop-a-convolutional-neural-network-from-scratch-for-mnist-handwritten-digit-classification/).
Use the MNIST data samples (http://yann.lecun.com/exdb/mnist/) for training and testing.
Develop a code in Python to design a CNN to perform 10 digit classification.
Experiment with 3 different combination of layers (conv and pooling).
You may use any libraries from Keras, Tensorflow, or Pytorch for developing a Python code.
Summarize the results and report them. Include the code you ran with your report as follows:
Describe what you have done for the homework assignment.
Elucidate and justify your network design and hyperparameters.
(e.g., filter size,filter numbers, pooling, of layers, of nodes on each layer, choice of activation functions on each layer, cost function, learning rate, optimizer, and so on)
MUST include a Learning curve (from an experiment)
MUST include five accuracy and their average from 5-fold cross validation (CV= use 80% for training. 20% training for 5 different partitions of dataset. See https://en.wikipedia.org/wiki/Cross-validation_(statistics) )
Compare the averaged accuracy of CNN with of ANN (HW 2).
Source code file(s)
Must be well organized (comments, indentation, …)
You need to upload the “original python file (*.py)” after changing to “*.py.txt”. For example, “*.py” to “*.py.txt”
