Learning Outcomes:
On successful completion of this module, students will be able to:
- Demonstrate an understanding of core deep learning concepts and their relationship to broader machine learning approaches.
- Apply fundamental training principles in the context of modern deep learning models.
- Analyse and compare deep learning architectures and model topologies appropriate for different data modalities and application domains.
- Design and implement deep learning solutions for applied problems using contemporary neural network architectures.
- Use established deep learning frameworks to develop, train, and evaluate models in practical settings.
- Assess model performance and behaviour using appropriate evaluation strategies.
- Critically evaluate the suitability, strengths, and limitations of deep learning methods for specific application contexts.
Indicative Module Content:
Neural Network Review
- Overview of deep learning and neural computation in modern machine learning
- Review of core neural network concepts, training objectives, optimisation principles, and model evaluation
- Generalisation, overfitting, and strategies for improving model robustness
Deep Learning Architectures for Vision and Sequential Data
- Convolutional Neural Networks (CNNs): architectural design and applications in visual understanding
- Deep and residual architectures, including the use of pretrained models
- Recurrent and sequence-based models: motivation and application contexts
- Transfer learning and domain adaptation
Sequence Models, Attention, and Transformers
- Sequence modelling paradigms and encoder–decoder architectures
- Attention mechanisms and their role in modelling long-range dependencies
- Transformer architectures and their impact on language and sequence modelling
- Representative applications in natural language processing and other domains
Unsupervised and Generative Models
- Representation learning and autoencoder-based models
- Variational Autoencoders (VAEs) and probabilistic generative modelling
- Generative Adversarial Networks (GANs): principles and applications
- Diffusion-based generative models and recent advances
Practical Implementation: hands-on learning using Jupyter notebooks and deep learning project implemented with contemporary deep learning frameworks.