Problem Solution

$30.00

Description

Train a deep convolution network on a GPU with PyTorch for the CIFAR10 dataset. The convolution network should use (A) dropout, (B) trained with RMSprop or ADAM, and (C) data augmentation. For full credit, the model should achieve 80-90% Test Accuracy.

Solution:

Test Accuracy = 85%

Train Accuracy = 92.2%

Number of epochs = 100

Learning Rate = 0.0001

Optimizer Used: ADAM

Dropout Probability = 0.2

Data Augmentation Techniques: Random Crop and Random Horizontal Flip

Convolution Network Model Architecture:

Convolution layer 1: 64 channels, k = 4, s = 1, P = 2. Batch normalization

Convolution layer 2: 64 channels, k = 4, s = 1, P = 2.

Max Pooling: s = 2, k = 2. Dropout

Convolution layer 3: 64 channels, k = 4, s = 1, P = 2. Batch normalization

Convolution layer 4: 64 channels, k = 4, s = 1, P = 2.

Max Pooling → Dropout

Convolution layer 5: 64 channels, k = 4, s = 1, P = 2. Batch normalization

Convolution layer 6: 64 channels, k = 3, s = 1, P = 0.

Dropout

Convolution layer 7: 64 channels, k = 3, s = 1, P = 0. Batch normalization

Convolution layer 8: 64 channels, k = 3, s = 1, P = 0. Batch normalization, Dropout

Fully connected layer 1: 500 units.

Fully connected layer 2: 500 units.


error: Content is protected !!