Master of Science in Machine Learning and Artificial Intelligence

The purpose of the qualification is manifold

• To serve as a flagship Machine Learning and Artificial Intelligence qualification on the African continent, comparable to the specialised structured Master's Degrees in Machine Learning (ML) and Artificial Intelligence (AI) offered at world-leading, research-intensive institutions (including UCL, Cambridge, Edinburgh, Carnegie Mellon, and Berkeley).
• To give learners as strong and broad a base as possible to succeed in subsequent Doctoral Degrees. An intense one-year structured Master of Science (MSc) qualification covering material at the forefront of ML and AL research could also incentivise more learners to continue with PhD studies in South Africa.
• To draw the very best learners in South Africa, Africa and beyond. The advantage for them is economical: a degree on par with the best in the world (potentially tailored to Africa's unique challenges), at a fraction of the cost of a European or US-based equivalent.
• To advance high-tech innovation around a research-intensive institution, and to develop talent for the South African and African high-tech business sector.

Qualifying learners will be able to

• Follow and reproduce state-of-the-art ML and AI research, and mathematically and practically grasp the key concepts in ML and AI;
• Design and implement ML and AI solutions to large-scale real-world business problems;
• Understand key techniques in probabilistic learning and inference;
• Design, create and apply ML and AI solutions to problems about:

> Visual image and video processing (Computer Vision).

> Time-series and complex, temporally varying data (Sequence Modelling).

> Text corpora and language understanding (Natural Language Processing).

> Planning and control (Reinforcement Learning).

• Understand the intersection of neuroscience and the frontiers of AI research.

Rationale

Of all universities on the African continent, the institution is uniquely placed to introduce a one-year structured (taught) Masters qualification in advanced ML and AI. The University has strong research groups within the faculties of Science and Engineering, as well as visibly increasing interest in ML within almost all other faculties. A cross-disciplinary interest group for postgraduate learners and academics has attracted close to 200 subscribers and active participants (mostly Postgraduate learners and faculty) within its first year of existence. Although the institution has a long-standing history of research excellence in ML and closely related fields like Computer Vision, Signal Processing, and Probabilistic Modelling, the centre of gravity for ML and AI on the African continent might soon shift toward central Africa. The African Masters in Machine Intelligence (AMMI) was launched in Rwanda in September 2018, with support from Google and Facebook. The second AMMI will be launched in Ghana in September 2019. Presently, Google Brain is also opening its first African research lab in Ghana. With such dedicated investment elsewhere, it is critical that South Africa continually positions itself as a centre for advanced ML and AI teaching and research.

Due to the rapid and pervasive growth of these fields, leading universities like Carnegie Mellon are already setting up undergraduate qualifications that specialise in AI. We foresee that with renewed investment in ML and AI in Africa, and given the popularity and successes of Pan-African qualifications like the Deep Learning Indaba, there will be a larger demand from within Africa (and specifically South Africa) for education and advanced training in ML and AI. Against this backdrop, it would seem that institution has a window of opportunity to introduce the proposed Masters of Science Degree in advanced ML and AI. The qualification is aimed at learners with a strong mathematical background, and in particular:

• Learners who wish to pursue a doctoral research Degree in ML or AI, and need a world-class foundation of the relevant fundamentals;
• Learners who wish to apply sophisticated ML and AI techniques in other fields (like Physics, Astronomy, Bioinformatics, Geoinformatics, Genetics, etc.);
• Learners who wish to start or be involved in innovative, high-tech start-ups;
• Practitioners who are working as data scientists or ML/AI developers in the industry, and need to be equipped with a technically deeper base to build advanced technologies;
• International learners seeking an MSc degree equivalent to top-tier universities in the US or Europe, at a more affordable cost (and those seeking an African experience).

With the last target group listed, the qualification also intends to increase the footfall of talented researchers and tech entrepreneurs through South Africa.

A common theme in the local high-tech ecosystem is that demand far exceeds supply and that the size of the talent pool of learners versed in advanced ML and AI techniques is becoming a bottleneck for growth. The qualification aims to fill this pressing need in the South African high-tech ecosystem.

Recognition of Prior Learning (RPL)

A learner who do not meet the normal admission requirements, but have demonstrated through prior learning that they have achieved a similar level of expertise, may be considered for admission by the RPL route. The qualification committee will consider all such applications, and weigh the formal (CAT) and non-formal or informal learning (RPL) against the knowledge required to complete the requirements of the structured qualification. This selection will be performed following the RPL rules of the institution.

Entry Requirements

The minimum entry requirement for this qualification is

• Honours Degree in Applied Mathematics, Computer Science, Mathematics or Statistics, NQF Level 8.

Or

• A 4-year Bachelor's Degree in Electrical Engineering, NQF Level 8.

Or

• An equivalent NQF Level 8 qualification in a field closely linked to ML and AI (including their application domains).

The learner will also be expected to

• Have existing and demonstrable proficiency in Python or an equivalent programming language,
• And be comfortable with numerical linear algebra and multivariable calculus,
• And possess basic knowledge of probability theory and statistics.

This qualification consists of the following compulsory and elective modules at NQF Level 9 totalling 180 Credits.

Compulsory Modules, Level 9, 120 Credits

• Mathematics for Machine Learning, 15 Credits.
• Probabilistic Modelling and Reasoning, 15 Credits.
• Foundations of Deep Learning, 15 Credits.
• Applied Machine Learning at Scale, 15 Credits.
• Research Project, 60 Credits.

Elective modules, 60 Credits (Select any 6 modules)

• Computer Vision, 10 Credits.
• Natural Language Processing, 10 Credits.
• Reinforcement Learning and Planning, 10 Credits.
• Sequence Modelling, 10 Credits.
• Advanced Probabilistic Modelling, 10 Credits.
• Optimisation for Machine Learning, 10 Credits.
• Monte Carlo Methods, 10 Credits.
• Artificial Intelligence and the Brain, 10 Credits.
• Advanced Topics in ML and AI I, 10 Credits.
• Advanced Topics in ML and AI II, 10 Credits.

1. Demonstrate the ability to follow and reproduce state-of-the-art ML and AI research, and mathematically and practically grasp the key concepts in ML and AI.
2. Demonstrate the ability to design and implement ML and AI solutions to large-scale real-world business problems.
3. Demonstrate the ability to understand key techniques in probabilistic learning and inference.
4. Demonstrate the ability to design, create, and apply ML and AI solutions about the visual image and video processing (Computer Vision), time series and complex, temporally varying data (Sequence Modelling), text corpora and language understanding (Natural Language Processing); and planning and control (Reinforcement Learning).
5. Demonstrate the ability to understand the intersection of neuroscience and the frontiers of AI research.

Associated Assessment Criteria for Exit Level Outcome 1

• Formulate and interpret fundamental mathematical concepts from linear algebra, calculus and statistics required to study ML and AI at MSc Level;
• Summarise, explain, and critically review key assumptions and mathematical methods of peer-reviewed ML papers and textbooks;
• Apply various mathematical tools to model currently relevant ML and AI problems, elegantly and effectively;
• Formulate, interpret and combine various probabilistic and information-theoretic concepts common in ML and AI approaches;
• Design and implement relevant probabilistic solutions to common data-centric problems, where modelling, reasoning and performing inference under uncertainty are key requirements;
• Judge probabilistic approaches to ML problem solving, in terms of their strengths and limitations;
• Apply probabilistic modelling and reasoning methods to own research work;
• Describe, summarise and compare standard deep neural network architectures, as a precursor for more advanced modules in the MSc curriculum;
• Build and practically implement those standard networks to perform a variety of supervised and unsupervised learning tasks;
• Analyse, critically assess, and explain the observed performance of those networks under various settings;
• Formulate and interpret the theory behind how ML is applied and deployed on internet-scale problems;
• Summarise, implement and critically assess basic and state-of-the-art deep learning approaches to solve a variety of Computer Vision problems;
• Formulate and interpret the mathematical theory underlying common Computer Vision approaches;
• Apply several devices necessary to build practical solutions to Computer Vision problems;
• Discuss problems and approaches at the cutting edge of Computer Vision research, from an ML and AI perspective.

Associated Assessment Criteria for Exit Level Outcome 2

• Give a broad account of the history of deep learning, and how various advances in the field are interconnected;
• Build mathematical and algorithmic models for internet users and entities that they engage with online;
• Formulate and interpret the theory behind how ML is applied and deployed on internet-scale problems;
• Summarise, implement and critically assess basic and state-of-the-art deep learning approaches to solve a variety of Computer Vision problems;
• Formulate and interpret the mathematical theory underlying common Computer Vision approaches;
• Apply several devices necessary to build practical solutions to Computer Vision problems;
• Discuss problems and approaches at the cutting edge of Computer Vision research, from an ML and AI perspective;
• Design and build recommender systems, formulate key performance indicators (KPIs), and test for causality in systems that affect these KPIs;
• Build mathematical and algorithmic models for internet users and entities that they engage with online.

Associated Assessment Criteria for Exit Level Outcome 3

• Formulate and interpret fundamental mathematical concepts from linear algebra, calculus and statistics required to study ML and AI at MSc level;
• Summarise, explain, and critically review key assumptions and mathematical methods of peer-reviewed ML papers and textbooks;
• Describe, summarise and compare standard deep neural network architectures, as a precursor for more advanced modules in the MSc curriculum;
• Build and practically implement those standard networks to perform a variety of supervised and unsupervised learning tasks;
• Analyse, critically assess, and explain the observed performance of those networks under various settings;
• Formulate and interpret the theory behind how ML is applied and deployed on internet-scale problems;
• Summarise, implement and critically assess basic and state-of-the-art deep learning approaches to solve a variety of Computer Vision problems;
• Formulate and interpret the mathematical theory underlying common Computer Vision approaches;
• Effectively use multiple sources of information (Accessing, processing and managing information, producing and communicating information);
• Review data critically (knowledge literacy, problem-solving)
• Apply learning strategies and take responsibility for work (Ethics and professional practice, Management of Learning, Accountability, Context and systems);
• Discuss problems and approaches at the cutting edge of Computer Vision research, from an ML and AI perspective,
• Summarise, implement and critically assess basic as well as state-of-the-art techniques to solve a variety of learning problems in Natural Language Processing;
• Formulate and interpret the mathematical theory underlying common Natural Language Processing approaches;
• Apply several devices necessary to build practical solutions to Natural Language Processing problems;
• Discuss problems and approaches at the cutting edge of Natural Language Processing research, from an ML and AI perspective;
• Summarise, implement and critically assess basic as well as state-of-the-art techniques to solve common Reinforcement Learning and Planning problems;
• Formulate and interpret the mathematical theory underlying common Reinforcement Learning and Planning approaches;
• Apply several devices necessary to build a practical solution to Reinforcement Learning and Planning problems;
• Discuss problems and approaches at the cutting edge of Reinforcement Learning and Planning research, from an ML and AI perspective;

Associated Assessment Criteria for Exit Level Outcome 4

• Design, create and apply ML and AI solutions about the visual image and video processing (Computer Vision), time series and complex, temporally varying data (Sequence Modelling), text corpora and language understanding (Natural Language Processing) and planning and control (Reinforcement Learning);
• Summarise, implement and critically assess basic as well as state-of-the-art techniques to solve a variety of learning problems in Natural Language Processing, Computer Vision, Sequence Modelling and Reinforcement Learning;
• Formulate and interpret the mathematical theory underlying common approaches;
• Apply several devices necessary to build practical solutions;
• Discuss problems and approaches at the cutting edge of research, from a Machine Learning and Artificial Intelligence perspective.

Associated Assessment Criteria for Exit Level Outcome 5

• Summarise and discuss basic elements of Neuro-and Cognitive Science, particularly brain structure and organisation in AI, form the point of view of mathematical modelling and algorithms;
• Formulate and interpret the theory of visual pathway and memory from an AI perspective;
• Discuss and appraise open problems in the brain-inspired AI;
• Formulate and interpret the theory behind a cutting-edge topic in ML/AI;
• Distil and practically implement the knowledge gained to model and solve new or familiar problems in ML/AI;
• Define a problem as well as an approach to solve that problem unambiguously and very clearly;
• Identify relevant publications and critically assess their importance and relevance to a specific problem;
• Design, select, tailor and apply technically advanced methods, techniques and theories from multiple sources, to solve complex practical and theoretical ML and AI problems;
• Synthesise coherent conclusions from experimental results, and identify key limitations and scope for future work;
• Write up the outcome of a research project in the format of a high-quality, high impact conference or journal style paper, and communicate those outcomes effectively to academic peers;
• Conduct independent research at the forefront of current ML and AI research.

Integrated Assessment

The modules will equip the learner with specialist knowledge and skills to the level where they will be able to critically evaluate the suitability of existing theories and techniques for a specific application.

The modules (with their associated assignments) and the research project will also develop learners' abilities to design, select and apply technically advanced methods, techniques and theories to complex practical and theoretical ML and AI problems.