UNCORRECTED PROOF NN: 2585 Model 5G pp. 1–10 (col. fig: nil) ARTICLE IN PRESS Neural Networks xx (xxxx) xxx–xxx Contents lists available at ScienceDirect Neural Networks journal homepage: www.elsevier.com/locate/neunet 2009 Special Issue Integrated feature and parameter optimization for an evolving spiking neural network: Exploring heterogeneous probabilistic models Stefan Schliebs a,∗ , Michaël Defoin-Platel b , Sue Worner c , Nikola Kasabov a a Knowledge Engineering and Discovery Research Institute (KEDRI), Auckland University of Technology, New Zealand b Biomathematics and Bioinformatics at Rothamsted Research, United Kingdom c Lincoln University, Centre for Bioprotection, New Zealand article info Article history: Received 6 May 2009 Received in revised form 3 June 2009 Accepted 25 June 2009 Keywords: Evolving spiking neural network Quantum-inspired evolutionary algorithm Multiple probabilistic model Estimation of distribution algorithm abstract This study introduces a quantum-inspired spiking neural network (QiSNN) as an integrated connectionist system, in which the features and parameters of an evolving spiking neural network are ∧ optimized together with the use of a quantum-inspired evolutionary algorithm. We propose here a novel ∧ optimization method that uses different representations to explore the two search spaces: A binary representation for ∧ optimizing feature subsets and a continuous representation for evolving appropriate real-valued configurations of the spiking network. The properties and characteristics of the improved framework are studied on two different synthetic benchmark datasets. Results are compared to traditional methods, namely a multi-layer-perceptron and a naïve Bayesian classifier (NBC). A previously used real world ecological dataset on invasive species establishment prediction is revisited and new results are obtained and ∧ analyzed by an ecological expert. The proposed method results in a much faster convergence to an optimal solution (or a close to it), in a better accuracy, and in a more informative set of features selected. © 2009 Elsevier Ltd. All rights reserved. 1. Introduction 1 Recently spiking neural networks (SNN) (Gerstner & Kistler, 2 2002; Izhikevich, 2003) have been developed as biologically plau- 3 sible connectionist models, which use trains of spikes for in- 4 ternal information representation. It was argued that SNN have 5 at least similar computational power than the traditional Multi- 6 Layer-Perceptron derivates (Maass, 1999). Nowadays many studies 7 attempt to use Spiking Neural Networks (SNN) for practical ap- 8 plications, some of them demonstrating very promising results 9 on solving complex real world problems. Substantial progress 10 has been made in areas like speech recognition (Verstraeten, 11 Schrauwen, & Stroobandt, 2005), learning rules (Bohte, Kok, & 12 Poutré, 2002), associative memory (Knoblauch, 2005), and func- 13 tion approximation (Iannella & Kindermann, 2005), just to name 14 a few. Based on Kasabov (2007) an evolving spiking neural net- 15 work was proposed and applied to audio-visual pattern recogni- 16 tion (Wysoski, Benuskova, & Kasabov, 2006, 2008). A similar type 17 of network was later used in the context of a taste recognition 18 task (Soltic, Wysoski, & Kasabov, 2008). 19 ∗ Corresponding author. E-mail addresses: sschlieb@aut.ac.nz, sschliebs@gmail.com (S. Schliebs), michael.defoinplatel@gmail.com (M. Defoin-Platel), worner@lincoln.ac.nz (S. Worner), nkasabov@aut.ac.nz (N. Kasabov). With encouraging results, spiking neural networks were ∧ pre- 20 sented in the context of a feature selection problem (Schliebs, 21 Defoin-Platel, & Kasabov, 2009). In this work a binary state- 22 of-art ∧ optimization algorithm, namely the Versatile Quantum- 23 inspired Evolutionary Algorithm (vQEA) (Defoin-Platel, Schliebs, & 24 Kasabov, 2007), was combined with an Evolving Spiking Neural 25 Networks(eSNN) (Wysoski et al., 2006). Through implementing 26 quantum principles, vQEA evolves in parallel a number of 27 independent probability vectors, that may interact at certain 28 intervals with each other, forming a multi-model Estimation of 29 Distribution Algorithm (EDA) (Defoin-Platel, Schliebs, & Kasabov, 30 in press). 31 Following the wrapper approach, vQEA was used to identify 32 relevant feature subsets and simultaneously evolve an optimal 33 eSNN parameter setting. This extended architecture was referred 34 to as the Quantum-inspired SNN (QiSNN) framework. Applied 35 to carefully designed benchmark data, containing irrelevant and 36 redundant features of varying information quality, the QiSNN- 37 based feature selection led to excellent classification results and 38 an accurate detection of relevant information in the dataset. 39 The QiSNN framework was used on a case study of ecological 40 modeling (Schliebs, Defoin-Platel, Worner, & Kasabov, in press). 41 Meteorological data, such as monthly and seasonal temperature, 42 rain fall and soil moisture recordings for different geographical 43 sites, were compiled from published results, and each global 44 site was ∧ labeled according to the presence or absence of the 45 0893-6080/$ – see front matter © 2009 Elsevier Ltd. All rights reserved. doi:10.1016/j.neunet.2009.06.038 Please cite this article in press as: Schliebs, S., et al. Integrated feature and parameter optimization for an evolving spiking neural network: Exploring heterogeneous probabilistic models. Neural Networks (2009), doi:10.1016/j.neunet.2009.06.038