Reinforcement Learning for Engineering (ELEN90098)

The key focus of this subject is the design and implementation of decision-making policies for enabling a dynamical system to behave autonomously and achieve a desired objective. This subject covers both model-based and model-free learning methods, with a focus on evaluating, contrasting, and combining methods. The influence of noisy sensor data on performance, and the trade-offs between exploration and exploitation during a learning phase, will also be covered. The examples used in this subject range across existing and emerging decision-making methods, and their application to consumer and industrial engineering systems.

INDICATIVE CONTENT
Topics covered may include:

• Reinforcement learning fundamentals such as principle of optimality, Bellman equation, value and policy iteration.
• Temporal-difference learning, Q-learning, Deep Q-learning, Actor critic methods and hybrid approaches in engineering context.
• Model based vs model free approaches, multi-agent RL and their engineering applications. 
• RL methods for Cyber-physical resilience and security such as fuzzing methods.