Assignments

There are totally 3 assignments, which account for 55% of the final grade. Students have 1-3 weeks for each assignment and are responsible for submitting the report on Canvas by the due date. The due dates for the assignments can be found in Schedule and Syllabus.

Late Policy: 2 late days in total are allocated across all 3 assignments, after that 25% penalty per day

Assignment #0

Due: Wednesday September 11, 11 am

This assignment is to help you get ready for future assignments. It is not for credit, but you are required to submit your responses to tasks at the end of each section in the assignment to Canvas. Note that your submission must be in PDF.

You can download Assignment #0 here.

Assignment #1

Due: Wednesday, October 9, 11 am

In this assignment, you have two main tasks:

  1. Implement backpropagation algorithm to train a neural network in Python.
  2. Implement and train a deep convolutional neural network in Tensorflow. We will also introduce you to different techniques to monitor your training and visualize your trained network.

You can download Assignment #1 here. We also provide you with starter codes three_layer_neural_network.py and dcn_mnist.py on Canvas, under ‘Files’ Section. Submit your report to Canvas.

Assignment #2

Due: Wednesday, November 6, 11 am

In this assignment, you have three main tasks:

  1. Train a convnet on CIFAR10 and visualize the trained model.
  2. Read and summarize the paper Visualizing and Understanding Convolutional Networks by Matthew D Zeiler and Rob Fergus.
  3. Train recurrent models on MNIST.

You can download Assignment #2 here. The starter codes are also provided, and their download links are included in the assignment.

Submit your report to Canvas.