bonsai

21fall system security final project

Baseline framework

• Pytorch

Target model

• SqueezeNet
• MobileNet

Dataset

• Cifar10 28x28, 60000 training, 10000 test
• Digit-Five
  • MNIST(pytorch) 28x28, 60000 training, 10000 test, greyscale
  • USPS(pytorch) 16x16, total size: 9298, greyscale
  • SVHN(pytorch) format 1: random shape, format 2: 32x32, total size: 600000, colored
  • MNIST-M 28x28, 59001 training, 90001 test, colored
  • Synthetic Digits random image shape, 12000 training, 2000 test, colored

TODO

• add validation
• refine test