Mohsen Ghaffari will give a talk about his work on Symbolic Reinforcement Learning. Details below.
Mohsen Ghaffari, PhD fellow, ITU.
Symbolic Reinforcement Learning
Reinforcement learning is a type of active learning in which the autonomous agent interacts with its environment, observes the results of its actions, and adapts its behavior appropriately. This type of learning has been widely studied as a learning method for determining optimal actions. However, in many cases (e.g. safety-critical systems), reinforcement learning faces significant challenges such as sample efficiency, safety assurance, explainability, reproducibility, sparse rewards, interpretability, and analyzability. This PhD thesis is structured to overcome some of the aforementioned challenges by hiring formal methods, specifically symbolic execution. I evaluate the proposed approaches’ correctness, efficiency, and reliability for several examples. I intend to use a mixture of standard textbook case studies and realistic industrial problems.