Part 4: Deep learning
Neural networks and how to train them, then the model families used for images, text, audio, generation, decision-making and connected data.
Key terms in this part (24)
- Batch normalisation
- Gradient clipping
- Learning-rate schedule
- Epsilon-greedy
- Attention
- Backpropagation
- Batch
- CNN (convolutional neural network)
- Data augmentation
- Deep learning
- Diffusion model
- Dropout
- Early stopping
- Epoch
- GAN (generative adversarial network)
- Graph neural network (GNN)
- Learning rate
- Neural network
- Reinforcement learning (RL)
- RNN and LSTM
- Segmentation
- Transfer learning
- Transformer
- Weight
Neural networks
- Neural networks in plain words
Explain neurons, layers, activations and backpropagation in plain words, and train a small network in Python and C#.
- Training deep networks well
Use batches, epochs, learning rates, optimisers, dropout, normalisation, early stopping, checkpoints and GPUs to train networks reliably.
Model families
- Computer vision: classification, detection and segmentation
Explain convolutional networks, transfer learning, object detection, segmentation and vision transformers, and run a trained vision model from Python and C#.
- Text and sequences: tokens, embeddings and RNNs
Explain how text becomes numbers (tokens and embeddings), how word vectors capture meaning, and how recurrent networks process sequences and why transformers replaced them.
- Transformers and attention
Explain attention with queries, keys and values, the parts of a transformer block, and the three transformer families (encoder, decoder, encoder-decoder).
- Generative models: autoencoders, GANs and diffusion
Explain how autoencoders, variational autoencoders, GANs and diffusion models create new data, and when industry uses each.
- Speech and audio: recognition and text-to-speech
Explain how audio becomes features, how speech recognition and text-to-speech work, and measure transcription quality with word error rate.
- Reinforcement learning
Explain agents, actions, rewards and policies, implement Q-learning and a policy-gradient agent, and connect reinforcement learning to how chat models are tuned.
- Graph neural networks: learning from connections
Represent data as nodes and edges, explain message passing, and classify nodes from very few labels using their neighbours.