DL Architectures & Training

Days 1–2

The first two days of Week 3 introduce the landscape of deep learning architectures, then get hands-on building and training a model.

Common architectures

• The basic classes of deep learning architectures and what each is good for.
• Matching an architecture to a scientific problem and data type.
• Where deep learning helps in Earth systems data science — and where simpler approaches still win.

Building and training a model

• Building a basic model in PyTorch.
• Training it on scientific data, on AWS GPUs.
• The training loop in practice: data, loss, optimization, and iteration.

Working at scale

• Using GPU resources effectively.
• Practical habits for reproducible training runs.

Outcomes

By the end of these two days you should understand the main architecture families and have built and trained your own PyTorch model on scientific data using AWS GPUs.