Deep reinforcement learning
Contact person: Matthia Sabatelli
Reinforcement learning algorithms enable a software agent to learn from its interaction with an environment. The goal is to optimize a policy that obtains the largest sum of rewards in its future. For this, the agent is situated in an environment and observes the environmental state. This state is used to select an action, after which the agent transits to a next state and may receive a reward or punishment. In our research, we have put the main focus on learning to play games. This allows for a controlled environment, simple rules, and huge state spaces. The agent is usually equipped with a (deep) neural network to learn to generalize over the continuous or high-dimensional state space.
|Last modified:||13 September 2022 3.00 p.m.|