Improving Multispike Learning With Plastic Synaptic Delays.

Area of Science:

• Computational Neuroscience
• Artificial Intelligence
• Machine Learning

Background:

• Spiking Neural Networks (SNNs) mimic brain processing for efficient cognition.
• Designing effective learning algorithms for SNNs is challenging due to complex dynamics.
• Existing methods primarily adjust synaptic weights, neglecting other adaptive components like delays.

Purpose of the Study:

• To propose a novel joint weight-delay plasticity rule (TDP-DL) for SNNs.
• To investigate the cooperative learning of synaptic weights and delays.
• To improve the performance and efficiency of SNNs through enhanced learning.

Main Methods:

• Developed the TDP-DL rule, integrating plastic delays into a multispike learning framework.
• Evaluated TDP-DL against baseline methods using simulations.
• Analyzed the cooperation between synaptic weights and delays via interval selectivity tasks.

Main Results:

• TDP-DL significantly improved multispike learning performance compared to baseline rules.
• Plastic delays were shown to enhance neuronal selectivity and flexibility by shifting information across time.
• The rule effectively integrated time-distributed information, boosting accuracy in image, speech, and event-based recognition tasks.

Conclusions:

• The TDP-DL rule offers an effective and efficient method for SNN learning.
• Cooperative plasticity in synaptic weights and delays is crucial for advancing SNN capabilities.
• This work contributes to improving the performance of spike-based neuromorphic computing.