Optimise, Regularise, and Accelerate Model Training using GPU-Enabled Workstations

Learning Objectives

At the end of this theory note, trainees must be able to:

1. Apply dropout, batch normalisation, and L2 regularisation to reduce overfitting.
2. Apply learning rate scheduling and early stopping to improve training convergence.
3. Configure and run model training on GPU-enabled workstations and NVIDIA Jetson Orin Nano hardware.
4. Monitor training performance and verify GPU utilisation using standard tools.
5. Enable mixed-precision training and tune batch size within hardware constraints.
6. Benchmark optimised models and document results in accordance with reporting requirements.