A Reinforced, Event-Driven, and Attention-Based Convolution Spiking Neural Network for Multivariate Time Series...

Area of Science:

• Artificial Intelligence
• Computational Neuroscience
• Machine Learning

Background:

• Spiking Neural Networks (SNNs) offer efficient event-driven processing for time series data.
• Extracting complex correlations in multivariate time series (MTS) remains a challenge for current SNN models.
• Existing methods often struggle with preserving spatio-temporal features during analysis.

Purpose of the Study:

• To propose a reinforced, event-driven, and attention-based convolution SNN model (REAT-CSNN) for enhanced MTS analysis.
• To develop novel methods for converting MTS into spike images and processing them effectively.
• To improve the feature extraction capabilities of SNNs for complex temporal patterns.

Main Methods:

• A joint Gramian Angular Field and Rate (GAFR) coding scheme converts MTS into spike images.
• An advanced Leaky Integrate-and-Fire (LIF) pooling strategy preserves critical features from spike images.
• A redesigned Convolutional Block Attention Mechanism (CBAM) is adapted for spike-based input, enhancing event-driven weighting.

Main Results:

• The REAT-CSNN model demonstrated superior performance on stock and PM2.5 MTS datasets.
• The proposed model achieved up to 3% better performance compared to state-of-the-art CNN and RNN techniques.
• The model exhibited significantly lower energy consumption than traditional deep learning approaches.

Conclusions:

• The REAT-CSNN model effectively addresses the challenge of complex correlation extraction in MTS.
• The novel GAFR coding, LIF-pooling, and adapted CBAM contribute to improved SNN performance.
• REAT-CSNN offers a promising, energy-efficient alternative for multivariate time series analysis.