Entropy and mutual information can improve fitness evaluation in coevolution of neural networks

Authors:
Boye Annfelt Høverstad;Haaken A. Moe;Min Shi
Affiliations:
Complex Adaptive Organically-inspired Systems Group, Department of Computer and Information Science, Norwegian University of Science and Technology, Trondheim, Norway;Complex Adaptive Organically-inspired Systems Group, Department of Computer and Information Science, Norwegian University of Science and Technology, Trondheim, Norway;Complex Adaptive Organically-inspired Systems Group, Department of Computer and Information Science, Norwegian University of Science and Technology, Trondheim, Norway
Venue:
CEC'09 Proceedings of the Eleventh conference on Congress on Evolutionary Computation
Year:
2009

Citing 11
Cited 0

On the approximate realization of continuous mappings by neural networks

Neural Networks
Spikes: exploring the neural code

Spikes: exploring the neural code
Connectivity and complexity: the relationship between neuroanatomy and brain dynamics

Neural Networks - Special issue on the global brain: imaging and modelling
Machine Learning

Machine Learning
Evolving neural networks through augmenting topologies

Evolutionary Computation
Robust non-linear control through neuroevolution

Robust non-linear control through neuroevolution
Learning basic navigation for personal satellite assistant using neuroevolution

GECCO '05 Proceedings of the 7th annual conference on Genetic and evolutionary computation
Evolving Neural Network Ensembles by Minimization of Mutual Information

International Journal of Hybrid Intelligent Systems
Revisiting the personal satellite assistant: neuroevolution with a modified enforced sub-populations algorithm

Proceedings of the 9th annual conference on Genetic and evolutionary computation
Information and Complexity in Statistical Modeling

Information and Complexity in Statistical Modeling
Active guidance for a finless rocket using neuroevolution

GECCO'03 Proceedings of the 2003 international conference on Genetic and evolutionary computation: PartII

Quantified Score

Hi-index	0.00

Visualization

Abstract

Accurate fitness estimates are notoriously difficult to attain in cooperative coevolution, as it is often unclear how to reward the individual parts given an evaluation of the evolved system as a whole. This is particularly true for cooperative approaches to neuroevolution, where neurons or neuronal groups are highly interdependent. In this paper we investigate this problem in the context of evolving neural networks for unstable control problems. We use measures from information theory and neuroscience to reward neurons in a neural network based on their degree of participation in the behavior of the network as a whole. In particular, we actively seek networks with high complexity and little redundancy, and argue that this can lead to efficient evolution of robust controllers. Preliminary results support this claim, and indicate that measures from information theory may provide meaningful information about the role of each neuron in a network.