Deep Learning
Your browser doesn't support speech here. You can still read everything, or try Chrome, Edge or Safari.

← All courses

Course 5 of 10 · Intermediate

Deep Learning

Build, train and debug neural networks. Tensors, autograd and networks from scratch; optimisers, regularisation and debugging; CNNs, RNNs and transformers; transfer learning, generative models and efficient deployment.

How this course works

Welcome

  • Level: Intermediate. You need: Machine Learning (at least the core and Deep learning sections).
  • Before you start: finish the neural network topics of the Machine Learning course, plus its matrices and gradients maths.
  • You build every piece from scratch in NumPy in the browser, then see the PyTorch version; Colab notebooks run the real thing on a free GPU.
  • Its game season is The Applied AI team.
  • Pick a topic from the sidebar; they go in order. Press ▶ Play this topic to hear a lesson read aloud, or 🔊 Explain on one section.
  • Topics climb from Beginner to Intermediate · Maths, Advanced and Code; use the Show chips to hide levels you don't need. Most have a Watch it animation, Try it demos and runnable code examples, and every topic ends with a quiz: answer it to mark the topic done ✓.
  • Stuck? Press 💬 Ask the AI tutor: a small free AI model that runs on your own device and answers from the course.
  • Progress is saved in this browser and shared across all the courses on this site.

Learning path: 🐍 Python → 🗄️ SQL & Databases → ∑ Maths for AI → 📈 Machine Learning → 🧠 Deep Learning → 👁️ Computer Vision → 💬 Natural Language Processing → ✨ Generative AI & LLM Apps → 🚀 Flask & Deployment → 🔬 Becoming an AI Researcher

Deep Learning

1. Deep Learning: Start HereWhat this track adds to the ML course, what deep-learning roles expect, and how to set up a GPU in Colab or on your own machine.2. Tensors, Shapes & BroadcastingThe data structure of deep learning: shapes and layout conventions, broadcasting rules and their silent bugs, reshaping, einsum and numerical precision.3. Backpropagation & Autograd from ScratchComputational graphs, the chain rule as message passing, gradients of the operations networks use, gradient checking, and how PyTorch's autograd works.4. Building Networks: Layers, Activations, Initialisation & LossesAssemble a multi-layer network piece by piece: dense layers, activation functions, why initialisation matters, output layers and stable losses, then train it from scratch.5. 🎮 Milestone D1: The reference implementationJoin Bluegum's new Applied AI team. Before anyone trains a big model, the team builds a small reference implementation from scratch: tensors, autograd, a network with a checked backward pass, and a training loop that learns. Includes 4 hands-on Python coding tasks (NumPy), checked by automated tests.6. Optimisers & Learning-Rate SchedulesMini-batch SGD, momentum, RMSProp, Adam and AdamW, why the learning rate matters most, warm-up and cosine schedules, and gradient clipping.7. Normalisation & RegularisationBatch, layer and RMS normalisation, dropout, weight decay, data augmentation, early stopping and label smoothing: what each does and when to use it.8. Debugging Training & Reading Loss CurvesA step-by-step recipe for training networks, diagnosing loss curves, vanishing and exploding gradients, residual connections, the classic bugs, and reproducible experiments.9. 🎮 Milestone D2: The run that wouldn't convergeThe first training run for the referral-form model fails: the loss spikes, turns into NaN, and the model gives different answers for the same form. Write the optimisers and normalisation layers yourself, read the curves, and fix the training script. Includes 4 hands-on Python coding tasks (NumPy), checked by automated tests.10. Convolutional Networks in DepthWhy convolution works for images, the shape arithmetic you need every day (padding, stride, channels, receptive field), pooling and 1×1 convolutions, and the architectures from LeNet to ResNet and ConvNeXt.11. Sequence Models: RNNs, LSTMs & GRUsNetworks with memory: recurrent networks, backpropagation through time, vanishing gradients, LSTM and GRU gates, sequence tasks, padding and masking, and when RNNs still make sense.12. 🎮 Milestone D3: Referral forms and vital signsTwo architectures for two problems: a convolutional reader for the handwritten boxes on scanned referral forms, and a recurrent early-warning model over hourly vital signs. Choose architectures against real constraints, build their core operations, and keep clinicians in the loop. Includes 3 hands-on Python coding tasks (NumPy, scikit-learn), checked by automated tests.13. Transformers from ScratchScaled dot-product attention, masks, multi-head attention, positional encodings, the transformer block, encoder vs decoder models, and the cost of attention.14. Transfer Learning, Fine-Tuning & EmbeddingsReuse what a pretrained model already knows: feature extraction vs fine-tuning, choosing a strategy, practical fine-tuning, LoRA and other parameter-efficient methods, and embeddings.15. Generative Models: Autoencoders, VAEs, GANs & DiffusionModels that learn to produce data: autoencoders for compression and anomaly detection, VAEs and latent spaces, GANs, diffusion models, and the risks of synthetic data.16. Efficient Training & DeploymentWhere GPU memory goes, mixed precision, gradient checkpointing, multi-GPU training, and making models small and fast: quantisation, pruning, distillation and export.17. 🎮 Milestone D4: Small, fast and safeRiverside wants the message-triage transformer on its own servers, with no GPU. Review the attention code, check how far pretrained features go with few labels, quantise the model, and decide how to ship it without missing an urgent message. Includes 4 hands-on Python coding tasks (NumPy, scikit-learn), checked by automated tests.18. ✅ Checkpoint: Deep Learning trackA mixed quiz on the whole Deep Learning track. Pass mark: 8 of 11 on the first try.
When you finish

Next course

Ready