Introduction

International Journal of Neural SystemsVol. 35, No. 11, 2502003 (2025) Free AccessIntroductionMarian Gheorghe, Alberto Leporati, Ferrante Neri, David Orellana Martín, and Mario J. Pérez-JiménezMarian GheorgheUniversity of Bradford, Bradford, UK, Alberto LeporatiUniversity of Milan-Bicocca, Milan, Italy, Ferrante NeriUniversity of Surrey, Guildford, UK, David Orellana MartínUniversity of Seville, Seville, Spain, and Mario J. Pérez-JiménezUniversity of Seville, Seville, Spainhttps://doi.org/10.1142/S0129065725020034Cited by:0 (Source: Crossref) Next This article is part of the issue: Special Issue: Spiking Neural NetworksGuest Editors: Marian Gheorghe, Alberto Leporati, Ferrante Neri, David Orellana-Martín and Mario J. Pérez-Jiménez AboutSectionsPDF/EPUB ToolsAdd to favoritesDownload CitationsTrack CitationsRecommend to Librarian ShareShare onFacebookTwitterLinked InRedditEmail Artificial neural networks (ANNs) are computational models inspired by the structure and function of biological neural networks in the human brain. Several architectures and various approaches to the communication between neurons have been proposed, many of which are based on natural (biological, chemical, physical, …) laws. Moreover, the performance of several learning algorithms has been studied when solving a plethora of computational problems. While some models incorporate new ingredients to solve computationally hard problems, others try to be as similar as possible to real-life neural networks, thus being suitable for in vivo implementation. In this context, spiking neural networks (SNNs) constitute the third generation of ANNs, after perceptrons and deep neural networks. SNNs have been widely investigated due to their unique characteristics. First, information is stored within neurons as sets of objects called spikes and is transmitted in the form of electrical pulses (spike trains) to adjacent neurons. Furthermore, from an implementation standpoint, SNNs promise to consume much less energy than traditional neural networks. When studying this type of neural network, two interesting points of view can be adopted. First, new models of spiking neurons can be defined depending on how the stored and/or transmitted information is encoded, and how the neuron behaves, possibly simulating the operation of biological neurons. We thus obtain models such as the Hodgkin–Huxley model, the Leaky Integrate and Fire model, and the Izhikevich model. Second, the learning methods for these models are different from usual algorithms for feed-forward and convolutional neural networks, like backpropagation, since by their nature they need not involve the same concepts as previous models of ANNs. Besides that, inspired by the plasticity of the connections between biological neurons, the creation or elimination of synapses can be included as part of the training mechanisms of these systems. Among different algorithms and extensions, a backpropagation algorithm adapted to SNNs, Hebbian learning, and anti-Hebbian learning have been part of several research works. From computer vision and image segmentation to robot control and complex systems modeling, including the solution of some important theoretical problems, SNNs have been used in real-life applications. In the framework of membrane computing, different variants of SNNs have been developed, and several studies have been conducted to search for discrete models that are closer to real-life neural networks, while keeping them feasible to simulate in in silico platforms. The studies in this area extend from theoretical works such as proofs of universality and presumed efficiency to many real-life applications, such as fault diagnosis and image classification or segmentation, paying attention to different approaches while simulating them. Indeed, these studies offer many open research lines that can be an interesting starting point for any researcher. The present special issue includes six studies about spiking neural systems, concerning both theoretical aspects, like the use of SNNs with colored spikes to solve computationally difficult combinatorial problems, and approaches from other models of computation like virus machines (used to solve the discrete logarithm problem), to the definition of a SNN-based associative memory, to applications in experimental areas such as object detection and thermal image segmentation. Each paper included in this issue went through at least two cycles of the journal's rigorous review process, receiving between five and seven reviews. FiguresReferencesRelatedDetails Recommended Vol. 35, No. 11 Metrics History Published: 15 September 2025 PDF download