Nonlinear Spiking Neural Systems With Autapses for Predicting Chaotic Time Series.

Area of Science:

• Computational Neuroscience
• Artificial Intelligence
• Machine Learning

Background:

• Spiking neural P (SNP) systems are advanced neural network models inspired by biological neurons.
• Chaotic time series forecasting presents significant challenges for current machine learning approaches.
• Existing models often struggle with the inherent complexity and unpredictability of chaotic dynamics.

Purpose of the Study:

• To propose a novel nonlinear version of SNP systems, termed nonlinear SNP systems with autapses (NSNP-AU).
• To develop a new recurrent-type prediction model, the NSNP-AU model, for chaotic time series forecasting.
• To evaluate the efficacy of the NSNP-AU model against state-of-the-art and baseline prediction models.

Main Methods:

• Introduction of NSNP-AU systems featuring nonlinear spike consumption/generation and nonlinear gate functions.
• Development of the NSNP-AU model, a recurrent neural network variant, implemented in a deep learning framework.
• Comparative analysis using four chaotic time series datasets against five state-of-the-art and 28 baseline models.

Main Results:

• The NSNP-AU model significantly outperformed 28 baseline prediction models.
• The proposed model demonstrated advantages over five state-of-the-art models in chaotic time series forecasting.
• Experimental results validate the effectiveness of the NSNP-AU approach for complex time series prediction.

Conclusions:

• The NSNP-AU model represents a promising advancement in recurrent neural network architectures.
• This novel approach offers enhanced capabilities for tackling challenging chaotic time series forecasting problems.
• The developed model provides a robust and effective solution for predicting complex, nonlinear dynamics.