Quantum-Inspired Feature and Parameter Optimisation of Evolving Spiking Neural Networks with a Case Study from Ecological Modeling Stefan Schliebs, Micha¨ el Defoin Platel, Sue Worner and Nikola Kasabov Abstract— The paper introduces a framework and implemen- tation of an integrated connectionist system, where the features and the parameters of an evolving spiking neural network are optimised together using a quantum representation of the features and a quantum inspired evolutionary algorithm for optimisation. The proposed model is applied on ecological data modeling problem demonstrating a significantly better classi- fication accuracy than traditional neural network approaches and a more appropriate feature subset selected from a larger initial number of features. Results are compared to a Na¨ ıve Bayesian Classifier. I. I NTRODUCTION R ECENTLY spiking neural networks (SNN) [1], [2] have been developed as biologically plausible connectionist models, which use trains of spikes for internal information representation. Today many applications using SNN receive a lot of research attention, some of them demonstrating very promising results on solving important real world problems. Based on [3] an evolving spiking neural network was pro- posed and applied to audio-visual pattern recognition [4], [5]. A similar type of network was later used in the context of a taste recognition task [6]. Other applications include e.g. neural based word recognition using liquid states [7], neural associative memory [8] and function approximation [9], just to name a few. With encouraging results spiking neural networks were presented in the context of a feature selection prob- lem [10]. In this work a state-of-art optimisation algorithm, namely the Versatile Quantum-inspired Evolutionary Algo- rithm (vQEA) [11], was combined with an Evolving Spiking Neural Network (eSNN) [4]. Implementing quantum princi- ples vQEA evolves in parallel a number of independent prob- ability vectors, which may interact at certain intervals with each other, forming a multi-model Estimation of Distribution Algorithm (EDAs) [12]. Following the wrapper approach, vQEA was used to identify relevant feature subsets and simultaneously evolve an optimal eSNN parameter setting. We will refer to this extended architecture as the Quantum- inspired SNN (QiSNN) framework during the course of Nikola Kasabov and Stefan Schliebs are with the Knowledge Engineer- ing and Discovery Research Institute (KEDRI), Auckland University of Technology, New Zealand (e-mail: nkasabov@aut.ac.nz, sschlieb@aut.ac.nz; http://www.kedri.info). Sue Worner is with the Lincoln University and Centre for Bioprotection, New Zealand (e-mail: worner@lincoln.ac.nz). Micha¨ el Defoin Platel is with the Biomathematics and Bioinformatics at Rothamsted Research, United Kingdom (e-mail: michael.defoinplatel@gmail.com) this paper. Applied to carefully designed benchmark data, containing irrelevant and redundant features of varying in- formation quality, the QiSNN-based feature selection led to excellent classification results and an accurate detection of relevant information in the dataset. This study intends to apply QiSNN on an ecological modeling problem. Meteorological data, such as monthly and seasonal temperature, rain fall and soil moisture recordings for different geographical sites, were compiled from pub- lished results. Furthermore for each global site the presence or absence of the Mediterranean fruit-fly (a serious fruit pest) was determined. Motivated by only inadequate results [13]– [15] using a different method, namely the Multi-layer Percep- tron (MLP), this study aims towards the identification of im- portant features relevant for predicting the presence/absence of this insect species. The obtained results may be also of importance to evaluate the risk of invasion of certain species into specific geographical regions. In the following sections we will first present the QiSNN framework, explain the experimental setup along with a description of the data used, followed by an analysis and discussion of the obtained results. II. FRAMEWORK AND I MPLEMENTATION OF QI SNN Based on our previous results on eSNN and quantum inspired evolutionary algorithms [3], [5], [12], [16], here we propose and explore an integrative quantum inspired feature selection using the eSNN architecture, tightly coupled with the learning environment (the data). A. eSNN Architecture The eSNN architecture uses a computationally very simple and efficient spiking neural model, in which early spikes, received by a neuron, are stronger weighted than later ones. The model was inspired by the neural processing of the human eye, which performs a very fast image processing. Experiments have shown that a primate only needs several hundreds of milliseconds to make reliable decisions about images that were presented in a test scenario [17]. Since it is known that neural image recognition involves several suc- ceeding layers of neurons, these experiments suggested that only very few spikes could be involved in the neural chain of image processing. In [18] a mathematical definition of these neurons was attempted and tested on some face recognition tasks, reporting encouraging experimental results. The same