Description
The general interest in the Internet of Things allows control devices and/or machines to connect through cloud-based architectures in order to share information about their status and environment. In many settings, as for example wind farms, similar machines are instantiated to perform the same task, which is called a fleet. Exploiting such a formation is especially useful in control settings. Specifically, seamless data sharing between fleet members could greatly improve the sample-efficiency of reinforcement learning techniques. However, in practice these devices, while similar, have small discrepancies due to production errors or degradation, preventing devices to simply aggregate and employ all fleet data. We propose a novel reinforcement learning method that learns to transfer knowledge between similar fleet members and creates member-specific dynamic models for control. To this end, our algorithm uses Gaussian processes to establish cross-member covariances. We demonstrate our approach on the continuous mountain car setting as a preliminary experiment. Our method significantly outperforms two baseline approaches, namely individual learning and joint learning where all fleet samples are used.Periode | 14 jul 2018 → 15 jul 2018 |
---|---|
Evenementstitel | The Adaptive Learning Agents Workshop at AAMAS 2018 |
Evenementstype | Conference |
Locatie | Stockholm, Sweden |
Mate van erkenning | International |