SCALER

About: Reinforcement learning (RL) is a subfield of machine learning concerned with how intelligent agents interact with unknown environments to maximise their rewards. The potential application of RL techniques on challenging real-world problems, such as autonomous vehicle control or smart energy grids, has brought significant attention to the field. However, state-of-the-art RL algorithms are not applicable in the most promising domains, largely due to the lack of formal performance guarantees. The EU-funded SCALER project aims to address this challenge by taking a principled approach to developing a new generation of provably efficient and scalable reinforcement learning algorithms. The methodology will be based on identifying novel structural properties of large-scale Markov decision processes that enable computationally and statistically efficient learning.

Duration: 1 October 2021 - 30 September 2026