This assignment is to evaluate the performance of a deep learning
machine learning program on the MNIST digits dataset. You can either:
Track I: Write the code yourself and report on your implementation's performance. A modestly sized network would satisfy the requirements. You should explain your code and focus on what you did.
Track II: Use web sources to implement a significant sized network and report on different architectures or different hyperparameters.
Here is the grading guide:
Write a report containing your method, performance and detail what you did and how well it worked. It should contain:
- Introduction: Specify the track you chose, and outline the method you chose and why you chose this method (1 points)
- Method: a) Track I: Detail the implementation of your method. b) Track II: List the package that you used, and detail the network architectures and hyperparameters you chose. (4 points)
- Results: a) Track I: Show the losses and accuracies of both training and test set. b) Track II: Compare the results of at least two different network architectures or two sets of hyperparameters. (3 points)
- Discussion: Explain the results that you got. Why is it good or not? (2 points)