Module COMP3547: DEEP LEARNING AND REINFORCEMENT LEARNING

Department: Computer Science

COMP3547: DEEP LEARNING AND REINFORCEMENT LEARNING

Prerequisites

• (COMP2261 Artificial Intelligence AND COMP2271 Data Science) OR COMP2231 Software Methodologies

Corequisites

• None

Excluded Combination of Modules

• None

Aims

• To enable students to be able to approach complex ill-defined problems that require deep layers of learning, and understand how this relates to learning in nature.
• To understand computational models of learning in dynamic environments, finding behaviours or paths that take actions to maximize cumulative rewards.
• To equip students with the ability to use modern deep learning libraries in order to effectively design, train, and test neural networks in different applications.

Content

• Theory of deep learning and learning in nature.
• Generative models and density estimation.
• Deep recurrent neural networks.
• Manifold learning and the relation between differential geometry and deep learning problems.
• Meta learning - learning how to learn.
• Markov decision processes and planning by dynamic programming.
• Model free prediction and control.
• Value-based and policy-based reinforcement learning.
• Integrating learning and planning, and exploration/exploitation.

Learning Outcomes

• On completion of the module, students will be able to demonstrate:
• an understanding of statistical learning theory with deep learning approaches.
• an understanding of state-of-the-art generative models and neural network architecture components.
• an understanding of the algorithms in manifold learning, meta learning, and reinforcement learning.

• On completion of the module, students will be able to demonstrate:
• an ability to use modern deep learning libraries to design, train, validate and test deep neural networks.
• an ability to design neural networks with respect to the task or dataset.
• an ability to identify inherent issues in dataset bias prior to training or architecture design.
• an ability to solve complex learning and planning problems in dynamic environments.

• On completion of the module, students will be able to demonstrate:
• the scientific approach to the design, training, validation, and testing of deep neural networks in a broad range of applications.
• an ability to learn, understand, and visualise the underlying structure of datasets.
• an ability to design and implement state-of-the-art generative models, bespoke deep neural network architectures, and reinforcement learning approaches.

Modes of Teaching, Learning and Assessment and how these contribute to the learning outcomes of the module

• Lectures enable the students to learn new material relevant to deep learning, manifold learning, meta learning and reinforcement learning, as well as their applications.
• Summative assessments assess the knowledge of deep learning libraries and application of methods and techniques, and examinations in addition assess an understanding of core theory and concepts.

Teaching Methods and Learning Hours

Summative Assessment

Formative Assessment:

Example formative exercises are given during the course. Additional revision lectures may be arranged in the module's lecture slots in the 3rd term.

■ Attendance at all activities marked with this symbol will be monitored. Students who fail to attend these activities, or to complete the summative or formative assessment specified above, will be subject to the procedures defined in the University's General Regulation V, and may be required to leave the University