This module provides a foundational understanding of deep learning from both theoretical and practical perspectives. It introduces the main neural network architectures and training algorithms, with particular emphasis on understanding how and why these methods work. It also covers basic diagnostic tools for assessing whether a model is learning meaningful structure in the data or relying on spurious correlations and artifacts. Lectures are complemented by hands-on practical sessions, in which participants experiment with different architectures, training strategies, and learning problems.
Key topics:
- Main neural network architectures and their inductive biases
- Training algorithms and optimization strategies
- Diagnostic tools for distinguishing meaningful structure from spurious correlations and artifacts
- Hands-on experimentation with architectures, training strategies, and learning problems
Learning Outcomes:
- Acquire a conceptual framework for understanding how deep learning methods work
- Develop, train, and evaluate neural network models on different learning problems
- Diagnose model behaviour, recognising when a model relies on artifacts rather than meaningful structure
