Synaptic facilitation and learning of multiplexed neural signals

Area of Science:

• Computational Neuroscience
• Artificial Intelligence
• Information Theory

Background:

• Spiking neural networks (SNNs) traditionally encode information using spike frequency.
• Temporal coding in SNNs, where spike timing is crucial, presents challenges in information encoding and learning.
• Existing temporal coding schemes often rely on population statistics rather than individual synaptic properties.

Purpose of the Study:

• To develop a novel synaptic plasticity mechanism for temporal coding in SNNs.
• To enable learning of information timing at the single-synapse level.
• To enhance information capacity and efficiency in SNNs.

Main Methods:

• Utilized information theory to analyze phase-coded spike trains.
• Developed a new synaptic plasticity rule for temporal coding.
• Demonstrated the mechanism using a simple demonstration network.

Main Results:

• Showcased multiplexing of multiple signals onto a single spike train.
• Demonstrated synaptic adaptation to specialize in different interspike intervals (phase relationships).
• Achieved denser encoding and improved energy efficiency in neural networks.

Conclusions:

• The novel approach allows distinct temporal codings to be distinguished through synaptic learning in SNNs.
• This work addresses a fundamental problem in SNNs, demonstrating single-synapse temporal learning.
• Results offer insights for functional neuroscience and potential links to biological temporal coding mechanisms.