Active Reinforcement Learning -- A Roadmap Towards Curious Classifier Systems for Self-Adaptation
This is an incremental proposal aimed at improving reinforcement learning for self-adaptive systems, with no concrete results or numbers provided.
The paper identifies drawbacks in traditional reinforcement learning approaches, such as trial-and-error and isolated problem handling, and proposes a research agenda for 'active reinforcement learning' to address these issues in intelligent systems.
Intelligent systems have the ability to improve their behaviour over time taking observations, experiences or explicit feedback into account. Traditional approaches separate the learning problem and make isolated use of techniques from different field of machine learning such as reinforcement learning, active learning, anomaly detection or transfer learning, for instance. In this context, the fundamental reinforcement learning approaches come with several drawbacks that hinder their application to real-world systems: trial-and-error, purely reactive behaviour or isolated problem handling. The idea of this article is to present a concept for alleviating these drawbacks by setting up a research agenda towards what we call "active reinforcement learning" in intelligent systems.