FIT2112 · Deep learning

Overview

Deep learning sits behind many of the most visible advances in modern artificial intelligence, from computer vision and natural language processing to generative systems and decision support. This unit develops the conceptual and practical capability needed to understand how deep learning systems work, when they are effective, and where their limitations emerge. You will study neural networks and deep learning within the broader context of machine learning, including backpropagation, optimisation, convolutional networks, recurrent architectures, attention mechanisms, transformers, embeddings and adversarial robustness. Through practical activities, you will use contemporary deep learning methods and tools to construct, train, evaluate and refine models for tasks such as image recognition, object recognition and text classification. The unit emphasises judgement, interpretation and communication as well as implementation. You will analyse why deep learning models succeed or fail, select appropriate architectures and training strategies, evaluate performance and robustness, and communicate technical findings, model behaviour and limitations for appropriate audiences and decision-making contexts.

Offerings

Assessment

The Handbook publishes no assessment items for this unit yet.

Assessment details may change. Please refer to the assessment information in Moodle closer to the start of the teaching period.

Requisites

Learning outcomes

1. Explain deep learning architectures, training processes and model behaviours within the broader context of modern artificial intelligence systems;
2. Construct, train, evaluate and refine deep learning models, including neural networks, convolutional networks, recurrent architectures and transformers, for artificial intelligence tasks;
3. Use contemporary deep learning methods and tools to manage model-development workflows, including data preparation, optimisation, experimentation, performance evaluation and model improvement;
4. Analyse complex artificial intelligence problems to justify suitable deep learning approaches, model architectures and training strategies, including when deep learning is unlikely to be appropriate;
5. Communicate deep learning concepts, model choices, experimental results, performance limitations and implications using appropriate technical language, evidence and visualisations.

Workload

Minimum total expected workload to achieve the learning outcomes for this unit is 144 hours per semester typically comprising a mixture of scheduled online and face to face learning activities and independent study. Independent study may include associated reading and preparation for scheduled teaching activities

Student experience, not official rules. Nothing here changes what the Handbook says.