ABSTRACT
Policy transfer occurs when a system transfers a policy learnt for one task to another task with little or no retraining, and allows a system to perform robustly and learn efficiently, especially when the new task is more complex than the original task. In this paper we report on work in progress into policy transfer using a relational learning classifier system. The system, FOX-cs, uses a high level relational language (a subset first order logic) in combination with a P-learning technique adapted for Xcs and its derivatives. FOX-CS achieved successful policy transfer in two blocks world tasks, stacking and onab, by learning a policy that was independent of the number of blocks, thus avoiding the prohibitive training times that would normally arise due to the exponential explosion in the number of states as the number of blocks increases.
- Joshua Cole, John Lloyd, and Kee Siong Ng. Symbolic learning for adaptive agents. In Proceedings of the Annual Partner Conference, Smart Internet Technology Cooperative Research Centre, 2003. www.smartinternet.com.au/SITWEB/publication/publications.jsp.Google Scholar
- Sašo Džeroski, Luc De Raedt, and Kurt Driessens. Relational reinforcement learning. Machine Learning, 43(1-2):7--52, 2001. Google ScholarDigital Library
- Pier Luca Lanzi. Mining interesting knowledge from data with the XCS classifier system. In Proceedings of the Genetic and Evolutionary Computation Conference (GECCO-2001), pages 958--965, San Francisco, California, USA, 2001. Morgan Kaufmann.Google Scholar
- Drew Mellor. A first order logic classifier system. In Proceedings of the Genetic and Evolutionary Computation Conference (GECCO-2005), To Appear, 2005. Google ScholarDigital Library
- John Slaney and Sylvie Thiébaux. Blocks World revisited. Artificial Intelligence, 125:119--153, 2001. Google ScholarDigital Library
- Stewart W. Wilson. Generalization in the XCS classifier system. In Genetic Programming 1998: Proceedings of the Third Annual Conference, pages 665--674, University of Wisconsin, Madison, Wisconsin, USA, 1998. Morgan Kaufmann.Google Scholar
Index Terms
- Policy transfer with a relational learning classifier system
Recommendations
Efficient Deep Reinforcement Learning through Policy Transfer
AAMAS '20: Proceedings of the 19th International Conference on Autonomous Agents and MultiAgent SystemsTransfer Learning (TL) has shown great potential to accelerate Reinforcement Learning (RL) by leveraging prior knowledge from past learned policies of relevant tasks. Existing TL approaches either explicitly computes the similarity between tasks or ...
Transferable XCS
GECCO '16: Proceedings of the Genetic and Evolutionary Computation Conference 2016Traditional accuracy-based XCS classifier system generally learns and evolves classifiers from scratch when facing each particular problem. Inspired by humans with the ability to learn new skills by inducing knowledge from related problems, transfer ...
Learning classifier system with average reward reinforcement learning
In the family of Learning Classifier Systems, the classifier system XCS is most widely used and investigated. However, the standard XCS has difficulties solving large multi-step problems, where long action chains are needed to get delayed rewards. Up to ...
Comments