Module manager: David Hogg
Email: scsdch@leeds.ac.uk
Taught: Semester 2 (Jan to Jun) View Timetable
Year running 2021/22
| COMP5450M | Knowledge Repres. & Reasoning |
| COMP5712M | Programming for Data Science |
This module is not approved as an Elective
The module introduces the field of Deep Learning, taking a strongly integrative and state of the art approach. In line with the use of AI in key sectors (e.g. finance, health, law), there is an emphasis on the combination of multiple input modalities – specifically, combining images, text and structured data. Students gain hands-on experience in developing systems to address real-world problems, providing the knowledge and skills necessary to develop an AI system as part of an MSc project.
To prepare students for a research degree involving innovation in the construction of AI systems.
To prepare students for a role in industry or the public-sector, where they may help to build, specify, recommend and critique AI systems.
To provide sufficient knowledge for students to be able to think creatively about possible AI solutions to real-world problems.
To gives students an understanding of the limitations of the state of the art and future research directions, including the need for causal explanations.
On completion of this module, students will be able to:
1. Design and implement AI systems in a language such as Python, based on machine learning;
2. Apply deep learning in standard AI tasks (e.g. image classification, sentiment analysis);
3. Represent data from various modalities (e.g. text, images, records) for multi-modal deep learning (e.g. translating images into textual captions, translating text into cognitive categories such as human sentiment);
4. Critically evaluate systems, using standard performance metrics;
5. Demonstrate an understanding of the current limitations of deep learning, the dependence on data and computational resources, and the challenge of causal explanation;
6. Apply their knowledge to address challenges from a specific sector; for example in detecting financial fraud from transaction data, and in informing medical diagnosis through combining medical images and structured data from electronic patient records.
Indicative content for this module includes:
• Use of Python for implementing neural networks.
• Representations for images, audio, text and structured data.
• Learning from data, performance evaluation and network visualisation.
• Deep feed-forward neural networks, convolutional neural networks, autoencoder networks.
• Handling sequential and relational data.
• Hands-on application of deep learning for the analysis of images, text and structured data, and combinations of modalities.
• Hands on application of deep learning with data and tasks from a specific domain (e.g. health, finance).
| Delivery type | Number | Length hours | Student hours |
|---|---|---|---|
| Lecture | 22 | 1 | 22 |
| Practical | 10 | 1 | 10 |
| Private study hours | 118 | ||
| Total Contact hours | 32 | ||
| Total hours (100hr per 10 credits) | 150 | ||
The student will design, build, train and evaluate several varieties of deep neural network models, some relating to their specialist domain (e.g. finance, health, law). Written reports from a subset of these exercises will form part of the formal assessment of the module.
Formative assessment will be coupled to the practical exercises in order to monitor student progress and provide feedback.
| Assessment type | Notes | % of formal assessment |
|---|---|---|
| Report | Practical Exercise - Program development with DFN/ CNN | 25 |
| In-course Assessment | Practical Assignment 2 | 25 |
| Total percentage (Assessment Coursework) | 50 | |
This module will be reassessed by an online time-constrained assessment.
| Exam type | Exam duration | % of formal assessment |
|---|---|---|
| Standard exam (closed essays, MCQs etc) (S1) | 3.0 Hrs Mins | 50 |
| Total percentage (Assessment Exams) | 50 | |
This module will be reassessed by an online time-constrained assessment.
The reading list is available from the Library website
Last updated: 15/03/2022
Errors, omissions, failed links etc should be notified to the Catalogue Team