Information recall using relative spike timing in a spiking neural network

Area of Science:

• Computational Neuroscience
• Artificial Intelligence

Background:

• Spiking neural networks (SNNs) are biologically inspired computational models.
• Representing and recalling information in SNNs, especially with noisy or incomplete data, remains a challenge.

Purpose of the Study:

• To develop and analyze a continuous-time neural network for completing and correcting spiking patterns.
• To investigate the network's capacity for information recall using multiple patterns simultaneously.

Main Methods:

• A novel neural network architecture operating in continuous time was designed.
• Information representation was based on the relative timing of individual spikes.
• Network performance was evaluated using two distinct measures of information recall capacity.

Main Results:

• The network successfully corrected and recalled multiple spiking patterns from partial, noisy inputs.
• Information recall capacity was found to scale linearly with the number of neurons and pattern period.
• A flexible encoding scheme was demonstrated, transitioning from precise spike timing to broader scene encoding.

Conclusions:

• The developed neural network effectively handles noisy and incomplete spiking patterns.
• Linear scaling of information capacity suggests natural measures for SNN information processing.
• The network's adaptable encoding supports the potential for learning diverse computational tasks.