SpiFoG: an efficient supervised learning algorithm for the network of spiking neurons.

Area of Science:

• Computational Neuroscience
• Machine Learning

Background:

• Spiking neural networks (SNNs) show promise for computational efficiency.
• Supervised learning in SNNs, particularly using evolutionary algorithms, remains underdeveloped.

Purpose of the Study:

• To introduce an efficient algorithm, SpiFoG, for supervised training of multilayer feedforward SNNs.
• To incorporate biologically plausible features like random synaptic delays and both excitatory/inhibitory neurons.

Main Methods:

• Utilized an elitist floating-point genetic algorithm with hybrid crossover for training.
• Employed leaky-integrate-and-fire spiking neurons with random synaptic delays.
• Trained both synaptic weights and random synaptic delays efficiently.
• Optimized computational efficiency by reducing simulation time and increasing time steps.

Main Results:

• SpiFoG demonstrated effective training of SNNs with both positive and negative synaptic weights.
• The algorithm successfully trained random synaptic delays alongside synaptic weights.
• Benchmarking on Iris and WBC datasets showed superior performance compared to state-of-the-art methods.

Conclusions:

• SpiFoG offers an efficient and effective approach for supervised learning in SNNs.
• The integration of random synaptic delays and diverse neuron types enhances biological plausibility and performance.
• This evolutionary algorithm-based method represents a significant advancement in SNN training.