Training Spiking Neural Networks with Metaheuristic Algorithms

Taking inspiration from the brain, spiking neural networks (SNNs) have been proposed to understand and diminish gap between machine learning neuromorphic computing. Supervised is most commonly used algorithm in traditional ANNs. However, directly training SNNs with backpropagation-based supervised methods challenging due discontinuous non-differentiable nature of neuron. To overcome these problems, this paper proposes a novel metaheuristic-based method for by adapting temporal error function. We investigated seven well-known metaheuristic algorithms called Harmony Search (HS), Cuckoo (CS), Differential Evolution (DE), Particle Swarm Optimization (PSO), Genetic Algorithm (GA), Artificial Bee Colony (ABC), Grammatical (GE) as search carrying out network training. Relative target firing times were instead fixed predetermined ones, making computation function simpler. The performance our approach was evaluated using five benchmark databases collected UCI Machine Learning Repository. experimental results showed that had competitive advantage solving four classification datasets compared other algorithms, accuracy levels 0.9858, 0.9768, 0.7752, 0.6871 iris, cancer, diabetes, liver datasets, respectively. Among CS reported best performance.