Deep Learning with PyTorch

Learn why PyTorch is easy to learn, pythonic, fast, and flexible for prototyping and production. Benefit from its well-documented API, modular blocks, flexible data loading, and a large community.

Explore the building blocks of deep learning with tensors, autograd, and computational graphs. Learn tensor operations, automatic differentiation, and how graphs enable backpropagation through examples.

Explore automatic differentiation with autograd, learn how gradients power backpropagation, and control tensor derivatives with requires_grad to train neural networks.

Explore how artificial neural networks, from perceptrons to deep feedforward networks, use forward and backward propagation to adjust weights with gradients in PyTorch.

Train a neural network end-to-end on a handwritten digits dataset, using a simple multilayer perceptron, forward pass, backpropagation, cross-entropy loss, SGD, data loaders, and evaluation.

Explore convolutional neural networks and their convolutional, pooling, and fully connected layers that power computer vision tasks like image recognition, object detection, and semantic segmentation.

Review the evolution, core layers, and model designs of convolutional neural networks, highlighting popular ImageNet-driven architectures. Prepare for the upcoming focus on training these networks.

Plot training and validation losses and accuracies across epochs to gauge model health. Test on unseen data shows 45% accuracy, guiding hyperparameter optimization ideas like image augmentation and learning rates.

Learn how to optimize hyperparameters in deep learning, including architecture, batch size, epochs, and learning rate, and apply manual, random, and grid search to balance training and validation loss.

Learn to implement transfer learning in PyTorch by using pre-trained models, freezing convolutional layers, and retraining the last layer on CIFAR-10.