Explain the architecture
Describe neurons, layers, activations and how information moves through a model.
Deep learning foundations
Neural networks from first principles
Understand layers, activations, loss and optimisation before building and evaluating a small neural model.
View curriculumz = weights · inputs + bias
activation = relu(z)
loss = compare(y, prediction)
weights -= rate * gradient
activation = relu(z)
loss = compare(y, prediction)
weights -= rate * gradient
Learning outcomes
See what happens inside the network.
Understand training
Connect loss functions, gradients and optimisation to the learning process.
Evaluate responsibly
Recognise overfitting, sensitivity, uncertainty and limits of model explanations.
Curriculum
Four deep-learning modules
Basic Python, algebra and machine-learning knowledge are recommended.
- Weights, biases and weighted sums
- Activation functions
- Forward propagation through layers
- Choose an appropriate loss function
- Gradients and backpropagation
- Learning rates and optimisers
- Feed-forward networks
- Convolutional networks
- Sequence models and transformers
- Validation curves and overfitting
- Regularisation and early stopping
- Interpretability claims and uncertainty