Practical Feature Selection for Online Reinforcement Learning and Planning in Relational MDPs

Jun Hao Alvin Ng, Ronald P. A. Petrick

Research output: Contribution to conferencePaperpeer-review

Abstract

Planning problems with large state spaces and unknown models can be solved by value-based reinforcement learning methods, using function approximation which abstracts states with feature vectors. Sample complexity increases when unnecessary features are selected. Conversely, omitting necessary features results in poor performances. A feature selection algorithm should add all necessary features while avoiding unnecessary ones. We introduce an online feature selection that considers the structure of relational MDPs (RMDPs) to reduce unnecessary features. Generalisation over different planning problems can be achieved by representing the approximated value-function with lifted representations of state variables and actions. Empirical results show that our method is practical for solving large RMDPs with reduced sample complexities and can generalise over different problems.
Original languageEnglish
Publication statusPublished - 16 Dec 2020
Event35th Workshop of the UK Planning and Scheduling Special Interest Group 2020 - Online
Duration: 16 Dec 202016 Dec 2020

Workshop

Workshop35th Workshop of the UK Planning and Scheduling Special Interest Group 2020
Abbreviated titleUK PlanSIG 2020
Period16/12/2016/12/20

Fingerprint

Dive into the research topics of 'Practical Feature Selection for Online Reinforcement Learning and Planning in Relational MDPs'. Together they form a unique fingerprint.

Cite this