-
Notifications
You must be signed in to change notification settings - Fork 2
/
verification.txt
15 lines (15 loc) · 1.51 KB
/
verification.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
OK R 1. `ComparisonIncreasingDropoutMNIST.ipynb` compares networks trained with different dropout on MNIST using bounds
OK R 2. `Regularization.ipynb` regularizes networks with b3 variance bound to achieve fault tolerance
OK 3. `ConvNetTest-MNIST.ipynb` trains a small convolutional network and verifies bound on it
OK 4. `ConvNetTest-VGG16.ipynb` loads the pre-trained VGG model and verifies the bound on it
OK 5. `ConvNetTest-ft.ipynb` compares fault tolerance of pre-trained CNN models
OK 6. `FaultTolerance-Continuity-FC-MNIST.ipynb` shows a decay of VarDelta when n increases when using our regularizer (Eq. 1 main paper)
OK 7. `TheAlgorithm.ipynb` tests Algorithm 1 on a small convnet for MNIST
OK 1. `FilterPlayground.ipynb` allows to tune the smooth() coefficients online
OK R 2. `WeightDecay-FC-MNIST.ipynb` shows that weights do not decay as we expect without regularization
OK R 3. `DerivativeDecay-FC-MNIST.ipynb` shows that derivatives do not decay without regularization as we expect
OK R 4. `WeightDecay-Continuity-FC-MNIST.ipynb` shows that with regularization, continuity holds (derivatives decay, weights stabilize)
OK FIXED 5. `ConvNetTest-VGG16-ManyImages.ipynb` investigates into filter size and how well b3 works in CNNs and tries to apply pooling on input for VGG
OK FIXED 6. `ErrorAdditivityRandom.ipynb` is the test of error additivity on Boston dataset
OK 7. `ErrorComparisonBoston.ipynb` compares Boston-trained networks (see main paper)
OK 9. `ErrorOnTraining.ipynb` tests the prediction of AP9 in the supplementary