Optimal Adaptive Exploration in Reinforcement Learning

Reinforcement Learning (RL) is concerned with learning efficient control policies for systems with unknown dynamics and reward function.

RL plays an increasing important role in a large spectrum of application domains including online platforms (recommender systems and search engines), robotics, and self-driving vehicles.

Over the last decade, RL algorithms, combined with modern function approximators such as deep neural networks, have shown unprecedented performance and have been able to solve highly complex sequential decision tasks better than humans.

The success of RL has been merely empirical so far, and in spite of interesting recent developments, we are still critically lacking theoretical tools to understand and guide the design of computationally and statistically efficient RL algorithms.Our research project aims at contributing to the theoretical foundations for the design of RL algorithms.

We will focus on problems arising in real-world systems characterized by a large number of states and actions.

We will develop novel RL algorithms able to learn and leverage any existing structure present in the system as well as combining statistical and computational efficiency.