RobustBNN RobustBNN Team Repository
MNIST using carlini code from git
- Accuracy without dropout = 99.15%
- Accuracy with dropout = 97.59%
- Absolute mean uncertainty on test data = 0.021
- Absolute mean uncertainty on adversarial = 0.224
- Accuracy of dropout on whitebox attack = 0.08
- Accuracy of dropout on graybox attack =
CIFAR-10 using carlini code from git
- Accuracy without dropout = 86.94%
- Accuracy with dropout = 86.54%
- Absolute mean uncertainty on test data = 0.003
- Absolute mean uncertainty on adversarial = 0.005
- Accuracy of dropout on whitebox attack = 0.08
- Accuracy of dropout on graybox attack =