Deep learning with PyTorch: from tensors to real models

PyTorch is the most widely used framework for deep learning research and a large share of production work. It is flexible and feels like ordinary Python, which makes it a great tool for learning how neural networks really work.

In this course we do not start with a one line fit call. We start with tensors and gradients, write a training loop by hand and understand every step. Then we build real models: an image classifier, a fine tuned pretrained network and a small text classifier, using the habits that make deep learning projects reproducible and debuggable.

What you will learn

• Tensors, broadcasting and running code on a GPU or Apple Silicon
• Autograd and how gradients drive learning
• Building models with nn.Module and training them with your own loop
• Datasets, DataLoaders and data augmentation
• Convolutional networks for images
• Transfer learning with pretrained models from torchvision
• Embeddings and a simple text classifier
• Regularisation, learning rate schedules, mixed precision and torch.compile
• Experiment tracking, checkpoints and debugging training runs

Before you start

You should be comfortable with Python and NumPy and know basic machine learning ideas such as training and validation splits, overfitting and evaluation metrics. A GPU is helpful but not required: every exercise runs on a laptop CPU with smaller settings, and a free cloud notebook works for the larger ones.