Training multi-layer spiking neural networks with plastic synaptic weights and delays

Area of Science:

• Computational Neuroscience
• Artificial Intelligence
• Machine Learning

Background:

• Spiking neural networks (SNNs) represent the third generation of neural networks, offering potential for ultra-low power consumption.
• SNNs are highly suitable for processing temporal information, but efficient training remains a challenge.
• Existing learning methods often focus solely on synaptic weight plasticity.

Purpose of the Study:

• To propose a novel supervised learning algorithm for multi-layer spiking neural networks.
• To enhance the biological plausibility and learning performance of SNNs.
• To leverage temporal information processing capabilities inherent in SNNs.

Main Methods:

• Developed a new supervised learning algorithm building upon the SpikeProp method.
• Incorporated adjustable synaptic weights and delays as trainable parameters.
• Utilized temporal information from spikes for learning, similar to SpikeProp.

Main Results:

• Experimental results demonstrate competitive learning performance for the proposed method.
• The algorithm shows improved efficacy compared to existing related works.
• The approach effectively utilizes temporal information for SNN training.

Conclusions:

• The proposed method offers an effective approach to training multi-layer spiking neural networks.
• Adjusting both synaptic weights and delays enhances SNN learning and biological relevance.
• This work contributes to the ongoing development of efficient SNN training algorithms.