Back to ListDeep Learning
Level
Intermediate, Broad, Algorithmic, Methodological.
This topic covers the design, implementation, and optimisation of deep learning architectures.
Learning outcomes
Content /Knowledge
Students should be able to:
- Understand and design different deep learning architectures including convolution and residual networks, transformers and diffusion models. Develop architectures for high dimensional data and also dynamic/time varying data.
- Design and implement different optimisation procedures, using differentiable programming (automatic differentiation/ back-propagation), as well as optimisation by stochastic gradient methods combined with acceleration and variance reduction techniques (mini-batching/normalisation).
Methodologicalskills
Students should be able to:
- Design and develop deep learning solutions taking advantage of available libraries and compute infrastructure.
- Correctly set the value of hyper-parameters for model evaluation and model selection in deep networks
- Evaluate the accuracy of the derived solutions in a systematic way, using available benchmarks and considering different performance metrics.
Transferrable/Application
Students should be able to:
- Work effectively with others in an interdisciplinary and/or international team.
- Design and manage individual projects.
- Clearly and succinctly communicate their ideas to technical audiences.