Overview of Spiking Neural Network Learning Approaches and Their Computational Complexities

Area of Science:

• Neuroscience
• Computer Science
• Artificial Intelligence

Background:

• Spiking neural networks (SNNs) more closely mimic biological neural processes than artificial neural networks (ANNs).
• SNNs hold potential for significant energy efficiency on event-driven neuromorphic hardware, reducing operational costs.
• Current limitations include the limited availability of neuromorphic hardware and performance gaps compared to ANNs on conventional architectures.

Purpose of the Study:

• To review and categorize existing learning algorithms for SNNs.
• To assess the computational complexity of various SNN learning methods.
• To provide a comprehensive overview of the SNN learning landscape.

Main Methods:

• Literature review of SNN learning algorithms.
• Categorization of algorithms based on their methodologies.
• Analysis of computational complexity for each category.

Main Results:

• SNNs show promise for energy efficiency but face challenges in speed and performance on standard hardware.
• ANNs currently outperform SNNs in execution speed and learning algorithm performance on typical machine learning tasks.
• The review categorizes SNN learning algorithms, offering insights into their computational demands.

Conclusions:

• Further research and development in SNN learning algorithms are crucial for realizing their full potential.
• The development of accessible neuromorphic hardware is essential for widespread SNN adoption.
• Understanding the computational complexity of SNN learning algorithms is key to optimizing their implementation.