Memory-inspired spiking hyperdimensional network for robust online learning.

Area of Science:

• Artificial Intelligence
• Computational Neuroscience
• Cognitive Computing

Background:

• Brain-inspired computing models like Spiking Neural Networks (SNNs) and HyperDimensional Computing (HDC) offer advantages in robustness and energy efficiency over traditional deep learning.
• SNNs mimic the brain's physical properties, while HDC models its abstract, functional aspects, suggesting complementary strengths for integration.

Purpose of the Study:

• To propose SpikeHD, the first framework integrating SNNs and HDC, inspired by psychological memory models.
• To create a scalable cognitive learning system that enhances brain functionality mimicry.

Main Methods:

• Spiking Neural Networks (SNNs) are used for low-level feature extraction, preserving spatial and temporal correlations in event-based spike data.
• HyperDimensional Computing (HDC) processes SNN outputs by mapping signals to high-dimensional space for abstract information learning and data classification.
• A two-stage information processing approach is employed for enhanced cognitive learning.

Main Results:

• SpikeHD significantly enhances learning capability through its two-stage processing.
• The framework demonstrates substantial robustness against noise and system failures.
• SpikeHD reduces network size and the number of parameters required for complex information learning compared to SNNs.

Conclusions:

• SpikeHD represents a novel and effective integration of SNNs and HDC for advanced cognitive learning.
• The proposed framework offers superior performance in learning capability, robustness, and efficiency.
• SpikeHD paves the way for more brain-like and powerful artificial intelligence systems.