Reinforcement learning (RL) is a subfield of machine learning that allows agents to learn optimal decision-making processes through trial and error procedures that receives rewards for good actions and penalties for bad ones. RL approaches are effective when a mathematical description of the process is not available. In particular, the use of RL methodologies showed its potential in the field of control system design, robotics, resource management and game development. This course provides a foundation in the core concepts of RL, such as Markov Decision Processes (MDPs), prominent RL algorithms like Q-learning and policy gradients, along with their theoretical aspects and practical applications. The course includes the presentations of real-world applications of RL to show how such algorithms are employed to solve relevant problems in simulated environments.

Basics of probabililty, statics and optimization. Elementary concepts of feedback control.

- Theoretical foundations of Reinforcement Learning (RL), including Markov Decision Processes (MDP) and exploration vs exploitation tradeoff. - Basic RL algorithms: Q-learning and policy gradients. - Design and implementation of basic RL agents to tackle problems in simulated environments. - Performance evaluation of RL agents. - Apply RL approaches for the design of feedback control systems.

In presenza

Prova di laboratorio di natura pratica sperimentale o informatico

P.D.2-2 - Maggio