Memory-Dependent Computation and Learning in Spiking Neural Networks Through Hebbian Plasticity.

Area of Science:

• Neuromorphic Engineering
• Computational Neuroscience
• Artificial Intelligence

Background:

• Spiking neural networks (SNNs) are crucial for energy-efficient neuromorphic hardware but lack biological memory capabilities.
• Biological memory, essential for long-term information retention, is linked to Hebbian plasticity but its role in SNNs is underexplored.
• Current artificial SNNs often struggle with generalization and complex cognitive tasks compared to biological systems.

Purpose of the Study:

• To propose Hebbian plasticity as fundamental for computations in both biological and artificial spiking neural systems.
• To introduce a novel memory-augmented SNN architecture incorporating Hebbian synaptic plasticity.
• To demonstrate the enhanced computational and learning capabilities of SNNs enriched with Hebbian plasticity.

Main Methods:

• Developed a novel SNN architecture integrating memory components.
• Enriched the SNN architecture with Hebbian synaptic plasticity mechanisms.
• Evaluated the performance of the memory-augmented SNNs on various cognitive tasks.

Main Results:

• The Hebbian-enriched SNNs demonstrated significant improvements in computational and learning abilities.
• Enhanced performance was observed in out-of-distribution generalization, one-shot learning, and cross-modal association.
• The architecture showed improved capabilities in language processing and reward-based learning tasks.

Conclusions:

• Hebbian synaptic plasticity is a fundamental principle for advancing SNN capabilities.
• Memory-augmented SNNs with Hebbian plasticity offer enhanced versatility and cognitive functions.
• This approach provides a pathway for building powerful cognitive neuromorphic systems.