Supervised Learning Algorithm for Multilayer Spiking Neural Networks with Long-Term Memory Spike Response Model.

Area of Science:

• Artificial Intelligence
• Computational Neuroscience
• Machine Learning

Background:

• Spiking neural networks (SNNs) are brain-inspired computational models processing information through timed spike trains.
• Developing efficient learning methods for SNNs is a critical research area.
• Existing methods often struggle with the complexity of multi-spike neurons and adaptation effects.

Purpose of the Study:

• To present a novel supervised learning algorithm for multilayer feedforward spiking neural networks.
• To enable neurons in all layers to fire multiple spikes.
• To incorporate biologically plausible adaptation effects into the learning process.

Main Methods:

• A supervised learning algorithm for multilayer feedforward SNNs was developed.
• Neurons utilize a biologically plausible long-term memory spike response model, accounting for spike history and adaptation.
• The gradient descent method was applied to derive synaptic weight updating rules for learning spike trains.

Main Results:

• The algorithm was tested on spatiotemporal pattern learning tasks, including spike train learning and nonlinear classification on UCI datasets.
• Simulation results demonstrated improved learning accuracy compared to existing supervised learning algorithms.
• The model effectively handles multi-spike neurons and adaptation effects.

Conclusions:

• The proposed supervised learning algorithm offers improved performance for spiking neural networks.
• This method advances the field of SNNs by effectively learning complex spatiotemporal patterns.
• The algorithm provides a robust framework for SNNs with biologically plausible neuron models.