Description
You are required to complete the following tasks:
1. Extend your code from hw3 to implement a 20-layer resnet model (you do not need to use batch norm layers)
Optional
You are not required to complete the following tasks,
however they are good exercises to get you familiar with pytorch.
1. Use the batch normalization layer
1. Reproduce the main result from the resnet paper by:
1. implement the 20 layer plain network, 56 layer resnet, and 56 layer plain network
1. verify that training error for the 56 layer plain model is worse than for the 20 layer plain model
1. verify that training error for the 56 layer resnet is better than the 20 layer resnet (and the 20/56 layer plain models)
1. Implement the dense blocks from the “Densely Connected Convolutional Networks” paper
Submission
Upload your python file to sakai
