Due on April 25

• In this HW, you will try to implement the backpropagation step for linear layer with numpy. Please download the jupyter notebook here. Btw, if you don't want to set up the environment. You may open the Github gist notebook from colab directly also. Select Github when you open new file (you need to click the magnifying glass on the left).

  • It is a simple two-layer neural network to try to classify “3” vs “7”

  • Note that you only need to write a couple of lines of code if you know what you are doing

  • Please submit your jupyter notebook and a screenshot of your code

  • Don't expect you can overfit the data. Fully-connected networks are no good for this problem. That's why we need CNN. But you should be able to get training error rate to be below 0.1 most of the time.

• Extra credit (50%). Add code to compute test error rate

• Extra credit (100%). Reimplement the network with one of the following three packages (Pytorch/Tensorflow/Keras)