Provably Efficient Algorithms for Reward Learning
Reinforcement Learning (RL) is a powerful tool for solving sequential decision-making problems. However, in practice, the design of the reward function is a rather difficult problem. In this work, we analyze Reward Learning (Rel), the discipline that aims to learn the reward function from a variety of human feedbacks, from a theoretical viewpoint. The ultimate goal of the thesis is to deepen our understanding of the subject and to describe novel provably efficient algorithms to solve the problem.
Back to Current Students