Spiking neural networks application to regression problems
Abstract
Spiking neural networks application to regression problems
Incoming article date: 24.10.2025The paper presents an analysis of third-generation spiking neural networks application to solving regression tasks. It considers the main models of spiking neurons (LIF, Izhikevich, Hodgkin-Huxley) from the perspective of their computational complexity and suitability for regression problems. Methods for encoding real-valued data into spike sequences are analyzed: rate coding, temporal coding, and population coding. Special attention is given to methods of decoding output spikes into continuous values, including rate decoding, first spike timing decoding, membrane potential decoding, and population voting. An assessment of the energy efficiency of various approaches is conducted, demonstrating a 100-200 fold reduction in energy consumption compared to traditional neural networks while maintaining acceptable accuracy. The research results confirm the promising application of spiking networks in embedded systems and Internet of Things devices.
Keywords: spiking neural networks, spike neuron model, spike coding, regression, energy efficiency