Countering Poisonous Inputs with Memetic Neuroevolution
Applied to certain problems, neuroevolution frequently gets stuck in local optima with very low fitness; in particular, this is true for some reinforcement learning problems where the input to the controller is a high-dimensional and/or ill-chosen state description. Evidently, some controller inputs are “poisonous”, and their inclusion induce such local optima. Previously, we proposed the memetic climber, which evolves neural network topology and weights at different timescales, as a solution to this problem. In this paper, we further explore the memetic climber, and introduce its population-based counterpart: the memetic ES. We also explore which types of inputs are poisonous for two different reinforcement learning problems.
KeywordsRandom Search Memetic Algorithm Standard Input Neural Network Weight Reinforcement Learning Problem
Unable to display preview. Download preview PDF.
- 1.Lucas, S.M., Togelius, J.: Point-to-point car racing: an initial study of evolution versus temporal difference learning. In: Proceedings of the IEEE Symposium on Computational Intelligence and Games (2007)Google Scholar
- 2.Igel, C.: Neuroevolution for reinforcement learning using evolution strategies. In: Proceedings of the Congress on Evolutionary Computation (CEC) (2003)Google Scholar
- 3.De Nardi, R., Togelius, J., Holland, O., Lucas, S.M.: Evolution of neural networks for helicopter control: Why modularity matters. In: Proceedings of the IEEE Congress on Evolutionary Computation (2006)Google Scholar
- 4.Togelius, J., Gomez, F., Schmidhuber, J.: Learning what to ignore: memetic climbing in weight and topology space. In: Congress on Evolutionary Computation (CEC) (to be presented, 2008)Google Scholar
- 5.Yao, X.: Evolving artificial neural networks. Proceedings 1447, 87(9) (1999)Google Scholar
- 7.Krasnogor, N., Pacheco, A.A.: Memetic algorithms. In: Metaheuristics in Neural Networks Learning, pp. 225–247. Springer, Heidelberg (2006)Google Scholar
- 10.Togelius, J.: Optimization, Imitation and Innovation: Computational Intelligence and Games. PhD thesis, Department of Computing and Electronic Systems, University of Essex, Colchester, UK (2007)Google Scholar