C03 Stage B · Core AI 48 h (16 T / 32 P)
Deep Learning
Trains, diagnoses and adapts neural networks.
Prerequisite
C01
Modules
5
NOS
—
Statutory mapping
—
Tech stack
PyTorchtorchvisionWeights & Biases
Modules
5 modules · 48 hours
M1 · Network fundamentals
10 h TOPICS
- Neurons and activation functions
- Feed-forward networks
- Loss functions
- Backpropagation derivation
- Computational graphs
- Autograd
Lab: Implement a two-layer network and backprop manually, then in PyTorch
Course material for this module is in production.
M2 · Training dynamics
10 h TOPICS
- Initialisation
- Optimisers: SGD, momentum, RMSProp, Adam, AdamW
- Learning-rate schedules
- Batch size
- Gradient clipping
- Vanishing and exploding gradients
Lab: Diagnose and repair a non-converging training run
Course material for this module is in production.
M3 · Regularisation and generalisation
10 h TOPICS
- Dropout
- Batch and layer normalisation
- Weight decay
- Early stopping
- Data augmentation
- Transfer learning and fine-tuning
Lab: Transfer-learn a pretrained backbone to a small dataset
Course material for this module is in production.
M4 · Convolutional networks
10 h TOPICS
- Convolution and pooling
- Stride and padding
- Receptive field
- Classic architectures
- Image pipelines
Lab: Train and evaluate an image classifier with augmentation
Course material for this module is in production.
M5 · Sequence models
8 h TOPICS
- RNN, LSTM, GRU
- Sequence-to-sequence
- Teacher forcing
- Limitations that motivate attention
Lab: Sequence model on time-series or text; compare to a transformer baseline
Course material for this module is in production.
Real-world work scenario
A training run plateaus at chance accuracy. Systematically eliminate causes — data pipeline, initialisation, learning rate, loss definition — and document the diagnosis path.
Assessment
Training diagnostics practical · transfer-learning project