An Introduction to Reinforcement Learning

Dr. Stefano Signoriello (infoteam Software AG) will give an overview talk about basic principles and different techniques in reinforcement learning.
The title and abstract of his talk are:

An Introduction to Reinforcement Learning
Reinforcement Learning (RL) is an area of machine learning concerned with finding optimal strategies in a sequential decision process by introducing a transition based scalar reward signal and subsequently finding trajectories that maximize some notion of cumulative reward. For instance such decision processes can be navigation tasks, control tasks like balancing a simulated inverted pendulum or mastering a game like Chess or Go. In the talk we are going to have a look at some of the basic principles of RL starting from the case of small environments with known or, particularly, unknown dynamics to cases of large problems where approximation techniques like neural nets are inevitable. One major issue in model-free RL is the trade-off between exploration and exploitation: On one hand, in order to gather information about the environment and find transitions with high reward the software agent needs to perform randomly chosen, sub-optimal actions. On the other hand, however, the agent needs to perform specific, seemingly optimal actions to follow trajectories with high cumulative reward. Simple exploration methods are often inefficient but still most practical. The talk will not cover much theory like convergence results but instead gives an overview of used techniques which, for the matter of demonstration, are applied on relatively easy problems (“frozen lake grid world”, “cart pole”, “connect 4”).
Please note that this will be the last talk in the winter semester 2019/2020. I would be very happy to receive some feedback from you regarding our Machine Learning seminar in order to improve the overall quality of our regular meetings for the next semester. Any comment on content or the general format of the seminar is very welcome! I will also send around a new Doodle poll soon to find an optimal time slot in the summer semester for everyone.