Recommended prerequisite for participation in the module

This module builds upon the basic knowledge of machine learning, including supervised, unsupervised, and reinforcement learning. Furthermore, students are expected to have a solid background in linear algebra, probability theory, basics of neural networks, and good programming skills.

Content, progress and pedagogy of the module

The course is composed of two parts. The first part of the course covers both single-agent deep reinforcement learning (SADRL) and multi-agent deep reinforcement learning (MADRL) algorithms (e.g., Deep Q-Networks (DQN), Double Deep Q-Networks (DDQN), Deep Deterministic Policy Gradient (DDPQ), and Actor-Critic methods). Then, the second part provides an overview of key concepts of generative artificial intelligence (Gen AI) and practical applications across industries

Learning objectives

Knowledge

• Must know the fundamentals of deep reinforcement learning 

• Must be able to understand the Bellman Equation and the function approximation in deep reinforcement learning 

• Must know the strategies to balance exploration and exploitation in deep reinforcement learning 

• Must know the challenges such as stability and convergence challenges in deep reinforcement learning 

• Must be able to understand advanced deep reinforcement learning algorithms  

• Must be able to discuss the real-world applications of deep reinforcement learning 

• Must know the key concepts of Gen AI 

• Must know the model architectures of Gen AI 

• Must know the real-world applications of Gen AI 

Skills

• Must be able to implement the taught DRL algorithms/methods and evaluate their performance 

• Must be able to apply the taught DRL algorithms/methods to solve concrete engineering and scientific problems 

• Must be able to develop and train generative models using frameworks like TensorFlow or PyTorch  

• Must have hands-on experience on Gen AI Models

Competences

• Must have competences in analyzing a given problem and identifying which deep reinforcement learning algorithms should be used for the problem 

• Must have competences in understanding the limitations of different deep reinforcement learning algorithms and how to overcome those limitations 

• Must be proficient in deep learning techniques, model training and optimization, data augmentation and preprocessing, model deployment, and evaluation of generated data 

Type of instruction

The instruction will combine lectures, invited talks, assignments, and exercises 

Exam

Exams

Name of exam	AI and Advanced Machine Learning
Type of exam	Written or oral exam
ECTS	5
Permitted aids	With certain aids: See exam specification
Assessment	7-point grading scale
Type of grading	Internal examination
Criteria of assessment	The criteria of assessment are stated in the Examination Policies and Procedures