• Designed a three-layer convolutional neural network architecture with PyTorch that classified 10 dog breeds within a 12,000+. images dataset. Implemented filter size, stride length, weight initializations, and forward propagation with max pooling layers.
• Explored the uses of transfer learning, data augmentation (grayscaling, rotating, and flipping), freezing layers, and early stopping.
• Utilized matplotlib to visualize validation and training performances on accuracy, loss, and AUROC score measures and tracked the number of epochs trained. Final model achieved a 85%+ AUROC score on the testing set.