Some machine learning and deep learning practice with PyTorch.

• FullyConnectedNet_MNIST
• AlexNet_CIFAR10
• Problems to solve
• Solved problems

Use a simple fully connected network to classify MNIST, which can achieve 97% test accuracy. But MNIST is too easy, even linear regression can achieve 90% test accuracy.

Use the original AlexNet and my modified AlexNet with global average pooling (GAP) to classify CIFAR10. Original AlexNet can achieve 76% test accuracy and AlexNet with GAP can achieve 80% test accuracy. CIFAR10 is much more difficult than MNIST.

1. Problem: Don't know why dataLoader parameter: num_worker>0 in .py file will lead to runtime error. While it can be set and works fine in .ipynb file.
   • Solution: it's just a Windows10 problem, don't sweat on that.