%0 Journal Article
%T A Radial Basis Function Spike Model for Indirect Learning via Integrate-and-Fire Sampling and Reconstruction Techniques
%A X. Zhang
%A G. Foderaro
%A C. Henriquez
%A A. M. J. VanDongen
%A S. Ferrari
%J Advances in Artificial Neural Systems
%D 2012
%I Hindawi Publishing Corporation
%R 10.1155/2012/713581
%X This paper presents a deterministic and adaptive spike model derived from radial basis functions and a leaky integrate-and-fire sampler developed for training spiking neural networks without direct weight manipulation. Several algorithms have been proposed for training spiking neural networks through biologically-plausible learning mechanisms, such as spike-timing-dependent synaptic plasticity and Hebbian plasticity. These algorithms typically rely on the ability to update the synaptic strengths, or weights, directly, through a weight update rule in which the weight increment can be decided and implemented based on the training equations. However, in several potential applications of adaptive spiking neural networks, including neuroprosthetic devices and CMOS/memristor nanoscale neuromorphic chips, the weights cannot be manipulated directly and, instead, tend to change over time by virtue of the pre- and postsynaptic neural activity. This paper presents an indirect learning method that induces changes in the synaptic weights by modulating spike-timing-dependent plasticity by means of controlled input spike trains. In place of the weights, the algorithm manipulates the input spike trains used to stimulate the input neurons by determining a sequence of spike timings that minimize a desired objective function and, indirectly, induce the desired synaptic plasticity in the network. 1. Introduction This paper presents a deterministic and adaptive spike model obtained from radial basis functions (RBFs) and a leaky integrate-and-fire (LIF) sampler for the purpose of training spiking neural networks (SNNs), without directly manipulating the synaptic weights. Spiking neural networks are computational models of biological neurons comprised of systems of differential equations that can reproduce some of the spike patterns and dynamics observed in real neuronal networks [1, 2]. Recently, SNNs have also been shown capable of simulating sigmoidal artificial neural networks (ANNs) and of solving small-dimensional nonlinear function approximation problems through reinforcement learning [3¨C5]. Like all ANN learning techniques, existing SNN training algorithms rely on the direct manipulation of the synaptic weights [4¨C9]. In other words, the learning algorithms typically include a weight-update rule by which the synaptic weights are updated over several iterations, based on the reinforcement signal or network performance. In many potential SNN applications, including neuroprosthetic devices, light-sensitive neuronal networks grown in vitro, and CMOS/memristor nanoscale
%U http://www.hindawi.com/journals/aans/2012/713581/