The 18-week curriculum
Every week opens on its first day. Your place is remembered in this browser, so you can stop and come back.
Start here
Pick any week. Week 1 assumes no prior experience.
Foundations
Tensors, autograd, training loops and generalisationTensors and Automatic Differentiation
The machinery under every deep learning library: tensors, shapes, gradients and the five-step training loop.
7 daysStart day 1 → Week 02Not startedBuilding Models with nn.Module
Architectures that are not a straight line, losses that match the task, and a training script you can trust and come back to.
7 daysStart day 1 → Week 03Not startedOptimisation
Momentum, Adam, learning rate schedules, batch size, clipping, and the failure shapes worth recognising on sight.
7 daysStart day 1 → Week 04Not startedRegularisation and Generalisation
Closing the gap between training and validation, one measured technique at a time.
7 daysStart day 1 →Vision
Convolution, residual networks and transfer learningConvolutional Networks
The architecture that knows pixels have neighbours, built up from one kernel to a full CIFAR-10 model.
7 daysStart day 1 → Week 06Not startedResidual Networks and Transfer Learning
Why depth stopped working, the connection that fixed it, and how to start from somebody else's weights.
7 daysStart day 1 →Sequences
Recurrence, and the attention that replaced itRecurrent Networks
Carrying state across a sequence: RNN, LSTM and GRU, padding done properly, and next-element prediction.
7 daysStart day 1 → Week 08Not startedAttention
Scaled dot-product attention built from scratch, multiple heads, masks, positions, and the block every transformer is made of.
7 daysStart day 1 →Transformers
The architecture, language modelling and pretrained modelsThe Transformer
Stacking attention into encoders and decoders, masking properly, and training a small language model that generates.
7 daysStart day 1 → Week 10Not startedLanguage Modelling
Training a character-level transformer end to end, with the baselines, the tokenisation and the arithmetic that surround it.
7 daysStart day 1 → Week 11Not startedPretrained Transformers
Using models somebody else trained: frozen features, fine-tuning, LoRA, and how to decide between them.
7 daysStart day 1 →Generative and self-supervised
Autoencoders, GANs, diffusion and learning without labelsAutoencoders and Variational Autoencoders
Learning representations with no labels at all, and turning a compressor into something you can sample from.
7 daysStart day 1 → Week 13Not startedGenerative Adversarial Networks and Diffusion
Two ways to generate images without a pixel loss, and honest measurements of what each one gives you.
7 daysStart day 1 → Week 14Not startedSelf-Supervised Learning
Training an encoder on data nobody labelled, and measuring honestly whether it was worth it.
7 daysStart day 1 →Production
Keras, making a model cheap, and serving itKeras Alongside PyTorch
The same models written twice, measured against each other, and an honest account of when each framework is the right one.
7 daysStart day 1 → Week 16Not startedMaking a Model Cheap Enough to Ship
Quantisation, pruning, distillation and export, each measured rather than recommended.
7 daysStart day 1 → Week 17Not startedDeployment and Monitoring
Putting a model behind an interface, and finding out whether it has quietly stopped working.
7 daysStart day 1 →Capstone
Your own project, and where to go nextEvery week ends with a self-assessment quiz, and each day can be marked complete as you go. Nothing is sent anywhere; your progress lives in this browser only.
Programme detailsAsk a question