Problem (Build LeNet for colorful image classification). In this problem, you are asked to train and test a neural network for entire CIFAR‐10 colorful image dataset. Some information of the network is as follows:
Its structure is modified LeNet. You can check the 4th slide in Lecture 10 for details.
An incomplete code has been given. You can fill it or re‐write all the codes by yourself.
Performance Requirement and Submission:
The test accuracy should achieve above 50%
You need to submit three results: 1) network without dropout/batch normalization, 2) network with one additional dropout layer and 3) network with one additional batch normalization. Compare the results in your submission.
Submission should include your source codes and screen snapshot of your train and test accuracy, plus the training time
Suggestion for hyperparameter setting (not necessary to follow): Check the default setting in the code. You are allowed to change them
About dataset loading: Check the default setting in the code. You are allowed to change them
Reminding: You can check PyTorch torch.nn to find the packed Batch Normalization and Dropout layer if you would like to use.