ABSTRACT
In this paper we argue that the performance of evolutionary computation on sequential decision problems strongly depends on the characteristics of the task at hand. On "error-avoidance" tasks, in which the decision process is interrupted every time a bad decision is made, evolutionary methods usually perform well. However, the same is not true for "goal-seeking" tasks, in which the objective is to find one or more target locations. In this case, it is not clear how to evaluate the unsuccessful candidate solutions, and the performance of evolutionary computation may depend on prior knowledge about the problem. Even though the hypothesis of this paper is essentially a conceptual one, we support our ideas with a computational experiment.
- D. E. Moriarty, A. C. Schultz, and J. J. Grefenstette. Evolutionary algorithms for reinforcement learning. Journal of Artificial Intelligence Research, 11:241--276, 1999.Google ScholarCross Ref
- R. S. Sutton and A. G. Barto. Reinforcement Learning: An Introduction. MIT Press, 1998. Google ScholarDigital Library
- C. Watkins. Learning from Delayed Rewards. PhD thesis, University of Cambridge, England, 1989.Google Scholar
- D. Whitley, S. Dominic, R. Das, and C. W. Anderson. Genetic reinforcement learning for neurocontrol problems. Machine Learning, 13(2--3):259--284, 1993. Google ScholarDigital Library
Index Terms
- On the characteristics of sequential decision problems and their impact on evolutionary computation
Recommendations
On the characteristics of sequential decision problems and their impact on evolutionary computation and reinforcement learning
EA'09: Proceedings of the 9th international conference on Artificial evolutionThis work provides a systematic review of the criteria most commonly used to classify sequential decision problems and discusses their impact on the performance of reinforcement learning and evolutionary computation. The paper also proposes a further ...
Evolutionary Computation for Dynamic Optimization Problems
GECCO Companion '15: Proceedings of the Companion Publication of the 2015 Annual Conference on Genetic and Evolutionary ComputationMany real-world optimization problems are subject to dynamic environments, where changes may occur over time regarding optimization objectives, decision variables, and/or constraint conditions. Such dynamic optimization problems (DOPs) are challenging ...
Analysis of evolutionary multi-tasking as an island model
GECCO '18: Proceedings of the Genetic and Evolutionary Computation Conference CompanionRecently, an idea of evolutionary multi-tasking has been proposed and applied to various types of optimization problems. The basic idea of evolutionary multi-tasking is to simultaneously solve multiple optimization problems (i.e., tasks) in a ...
Comments