AI502_Midterm_Project

python code, notebooks and Images used for AI502 Midterm Project. I made almost all of code and all images.

Images

Images from original papers are didn't uploaded here, only images from me are uploaded.

Figure 2: Comparison of test loss and train accuracy for Batch size increasing and Learning rate decaying.

Figure 4: Comparison of Adam, RAdam and SGD for MNIST classifier

Figure 5: Comparison for No Batch Norm and 4 kinds of Batch Norm

Additional figure 1: Comparison of Resnet50 & MobileNetV2

Dataset = CIFAR100
Batch size = 4096
lr = 5e-2
num_epochs = 1000
Optimizer = Adam
HW = RTX 2080Ti \

References

Codes without specification on this References tab are coded by me.

2. On the Variance of the Adaptive Learning Rate and Beyond
   RAdam : https://github.com/LiyuanLucasLiu/RAdam

3. MobileNetV2: Inverted residuals and linear bottlenecks
   ResNet : https://pytorch.org/docs/stable/torchvision/models.html
   MobileNetV2 : https://github.com/pytorch/vision/tree/master/torchvision/models