System Design & Optimization Lab @ Inha University

Reinforcement learning (RL) is one of the key components of artificial intelligence-based decision-making. RL optimizes dynamic decisions via a feedback loop including a target system and learning agents. We study RL algorithms to tackle real-world applications and aim to enhance the practicality of RL algorithms in real-world systems.

Figure from Sutton and Barto, (2018), Reinforcement learning: An introduction. MIT press.

Sequential decision-making problems represent dynamic decision problems that can adapt decisions according to the current system state. We use various sequential decision-making models, including  Markov Decision Processes (MDP), decentralized-Partially Observable MDP (dec-POMDP), and Stochastic Game (SG), corresponding to the characteristics of a system.

Figure from Puterman , Martin L. (2009).  Markov decision processes: discrete stochastic dynamic programming, John Wiley & Sons.

Engineering systems are characterized by a large number of interrelated elements. Our main focus is to control individual components to achieve the system-level goal. We are interested in designing better systems before making operational decisions as well as making better decisions in the current system.