The problem with training spiking neural networks (SNNs) is relevant due to the ultra-low power consumption these networks could exhibit when implemented in neuromorphic hardware. The ongoing progress in the fabrication of memristors, a prospective basis for analogue synapses, gives relevance to studying the possibility of SNN learning on the base of synaptic plasticity models, obtained by fitting the experimental measurements of the memristor conductance change. The dynamics of memristor conductances is (necessarily) nonlinear, because conductance changes depend on the spike timings, which neurons emit in an all-or-none fashion. The ability to solve classification tasks was previously shown for spiking network models based on the bio-inspired local learning mechanism of spike-timing-dependent plasticity (STDP), as well as with the plasticity that models the conductance change of nanocomposite (NC) memristors. Input data were presented to the network encoded into the intensities of Poisson input spike sequences. This work considers another approach for encoding input data into input spike sequences presented to the network: temporal encoding, in which an input vector is transformed into relative timing of individual input spikes. Since temporal encoding uses fewer input spikes, the processing of each input vector by the network can be faster and more energy-efficient. The aim of the current work is to show the applicability of temporal encoding to training spiking networks with three synaptic plasticity models: STDP, NC memristor approximation, and PPX memristor approximation. We assess the accuracy of the proposed approach on several benchmark classification tasks: Fisher’s Iris, Wisconsin breast cancer, and the pole balancing task (CartPole). The accuracies achieved by SNN with memristor plasticity and conventional STDP are comparable and are on par with classic machine learning approaches.

A method to solve the classification task using a spiking neural network with encoding the input by patterns of spike times along with Spike-Timing-Dependent Plasticity learning is proposed. Input data is encoded using Gaussian receptive fields. The method is tested on Fisher’s Iris dataset. As the result, after learning a neuron responds with less latency to patterns encoding samples of the class on which it was trained, in comparison to the classes it was not trained on.

A study of possibility to model the learning process on base of different forms of timing-dependent plasticity (STDP) was performed. It is shown that the learning ability depends on the choice of spike pairing scheme and the type of input signal used for learning. The comparison of performance of several STDP rules along with several neuron models (leaky integrate-and-fire, static, Izhikevich and Hodgkin-Huxley) was carried out using the NEST simulator. The combinations of input signal and STDP spike pairing scheme, which demonstrate the best learning abilities, were extracted.

Spiking neural networks are increasingly popular for machine learning applications, thanks to ongoing progress in the hardware implementation of spiking networks in low-energy-consuming neuromorphic hardware. Still, obtaining a spiking neural network model that solves a classification task with the same level of accuracy as a artificial neural network remains a challenge. Of especial relevance is the development of spiking neural network models trained on base of local synaptic plasticity rules that can be implemented either in digital neuromorphic chips or in memristive devices. However, existing spiking networks with local learning all have, to our knowledge, one-layer topology, and no multi-layer ones have been proposed so far. As an initial step towards resolving this problem, we study the possibility of using a non-trainable layer of spiking neurons as an encoder layer within a prospective multi-layer spiking neural network, implying that the prospective subsequent layers could be trained on base of local plasticity. We study a spiking neural network model with non-trainable synaptic weights preset on base of logistic maps, similarly to what was proposed recently in the literature for formal neural networks. We show that one layer of spiking neurons with such weights can transform input vectors preserving the information about the classes of the input vectors, so that this information can be extracted from the neuron’s output spiking rates by a subsequent classifier, such as Gradient Boosting. The accuracy obtained on the Fisher’s Iris classification task is 95%, with the deviation range of 5% over the five cross-validation folds. This is on par with other existing methods for Fisher’s Iris classification with spiking neural networks, which shows the prospective possibility of using the proposed layer as an encoder within a multi-layer network.