On the probabilistic optimization of spiking neural networks

Area of Science:

• Computational Neuroscience
• Artificial Intelligence
• Machine Learning

Background:

• Optimizing Spiking Neural Networks (SNNs) involves complex challenges in selecting network topology and configuring internal parameters.
• Evolutionary Algorithms (EAs) provide effective solutions, with heterogeneous optimization methods showing particular promise for exploring complex search spaces.

Purpose of the Study:

• To present a literature review on heterogeneous optimization algorithms.
• To introduce and experimentally analyze a novel Heterogeneous Multi-Model Estimation of Distribution Algorithm (hMM-EDA).
• To derive practical configuration guidelines for hMM-EDA and compare its performance against state-of-the-art methods.

Main Methods:

• Literature review of heterogeneous optimization algorithms.
• Detailed discussion of probabilistic optimization for SNNs.
• Experimental analysis of the proposed hMM-EDA on a synthetic heterogeneous benchmark problem.

Main Results:

• The hMM-EDA is demonstrated to be a light-weight, fast, and reliable optimization method.
• The algorithm requires configuration of very few parameters.
• hMM-EDA shows highly competitive performance on a synthetic heterogeneous benchmark problem.

Conclusions:

• The novel hMM-EDA is a promising and efficient approach for optimizing Spiking Neural Networks.
• Its ease of configuration and strong performance suggest suitability for various SNN applications.
• The findings contribute to the advancement of optimization techniques in computational neuroscience and AI.