ABSTRACT
Rule-based machine learning (RBML) algorithms such as learning classifier systems (LCS) are well suited to classification problems with complex interactions and heterogeneous associations. Alternatively, genetic programming (GP) has a complementary set of strengths and weaknesses best suited to regression problems and homogeneous associations. Both approaches yield largely interpretable solutions. An ideal ML algorithm would have the capacity to adapt and blend representation to best suit the problem at hand. In order to combine the strengths of these respective algorithm representations, a framework allowing coexistence and co-evolution of trees and rules is needed. In this work, we lay the empirical groundwork for such a framework by demonstrating the capability of GP trees to be evolved within an LCS-algorithm framework with comparable performance to a set of standard GP frameworks. We discuss how these results support the feasibility of a GP-LCS framework and next-step challenges to be addressed.
- L Bull. 2004. Learning Classifier Systems: A Brief Introduction. Studies in Fuzziness and Soft Computing, Vol. 150. Springer, Berlin, Heidelberg, Chapter 1, 1--12.Google Scholar
- Félix-Antoine Fortin, François-Michel De Rainville, Marc-André Gardner, Marc Parizeau, and Christian Gagné. 2012. DEAP: Evolutionary algorithms made easy. Journal of Machine Learning Research 13, Jul (2012), 2171--2175.Google ScholarDigital Library
- John R Koza and John R Koza. 1992. Genetic programming: on the programming of computers by means of natural selection. Vol. 1. MIT press.Google ScholarDigital Library
- T Stephens. 2015. Gplearn Model, Genetic Programming. (2015).Google Scholar
- Ryan Urbanowicz, Randal Olson, and Jason Moore. 2016. Pareto inspired multi-objective rule fitness for noise-adaptive rule-based machine learning. In International Conference on Parallel Problem Solving from Nature. Springer, 514--524.Google ScholarCross Ref
- Ryan Urbanowicz, Niranjan Ramanand, and Jason Moore. 2015. Continuous endpoint data mining with exstracs: A supervised learning classifier system. In Proceedings of the Companion Publication of the 2015 Annual Conference on Genetic and Evolutionary Computation. 1029--1036.Google ScholarDigital Library
- Ryan J Urbanowicz and Jason H Moore. 2015. ExSTraCS 2.0: description and evaluation of a scalable learning classifier system. Evolutionary intelligence 8, 2-3 (2015), 89--116.Google Scholar
- Ryan J Urbanowicz, Ben Yang, and Jason H Moore. 2018. Problem Driven Machine Learning by Co-evolving Genetic Programming Trees and Rules in a Learning Classifier System. In Genetic Programming Theory and Practice XV. Springer, 55--71.Google Scholar
Index Terms
- Evolving genetic programming trees in a rule-based learning framework
Recommendations
A Comparison of three evolutionary strategies for multiobjective genetic programming
We report what we believe to be the first comparative study of multi-objective genetic programming (GP) algorithms on benchmark symbolic regression and machine learning problems. We compare the Strength Pareto Evolutionary Algorithm (SPEA2), the Non-...
Generating interpretable reinforcement learning policies using genetic programming
GECCO '19: Proceedings of the Genetic and Evolutionary Computation Conference CompanionThe search for interpretable reinforcement learning policies is of high academic and industrial interest. Especially for industrial systems, domain experts are more likely to deploy autonomously learned controllers if they are understandable and ...
Genetic programming methods for reinforcement learning
GECCO '19: Proceedings of the Genetic and Evolutionary Computation ConferenceReinforcement Learning (RL) algorithms can be used to optimally solve dynamic decision-making and control problems. With continuous-valued state and input variables, RL algorithms must rely on function approximators to represent the value function and ...
Comments