Temporal single spike coding for effective transfer learning in spiking neural networks.

Area of Science:

• Artificial Intelligence
• Computational Neuroscience
• Machine Learning

Background:

• Spiking Neural Networks (SNNs) offer energy-efficient computation but face training challenges.
• Transfer Learning (TL) effectively utilizes pre-trained models but requires efficient integration with SNNs.
• Temporal coding in SNNs, particularly single-spike coding, presents opportunities for efficient learning rules.

Purpose of the Study:

• To introduce a novel supervised learning rule, Temporal Single Spike Coding for Effective Transfer Learning (TS4TL), for training multilayer fully connected SNNs.
• To propose an "Absolute Target" assignment method for single-spike temporal coding to simplify and accelerate SNN training.
• To demonstrate the efficacy of TS4TL within a Transfer Learning framework for classification tasks, especially with limited data.

Main Methods:

• Developed TS4TL, a supervised learning rule integrating an "Absolute Target" method for single-spike temporal coding.
• Implemented TS4TL for training SNNs as classifier blocks within a Transfer Learning pipeline.
• Evaluated TS4TL on benchmark datasets including Eth80, Fashion-MNIST, MNIST, and Caltech101-Face/Bike.

Main Results:

• Achieved state-of-the-art accuracies: 98.91% on Eth80, 91.89% on Fashion-MNIST, 98.45% on MNIST, and 97.75% on Caltech101-Face/Bike.
• Demonstrated reduced computational complexity, training time, and energy consumption compared to existing methods.
• Successfully reduced neuron misfires, ensuring accurate first-spike coding and stable training.

Conclusions:

• TS4TL provides a scalable, efficient, and high-performance solution for temporal learning in SNNs.
• The "Absolute Target" method simplifies training while maintaining accuracy and reducing resource demands.
• TS4TL effectively leverages Transfer Learning for SNN classification, even with limited or varied data distributions.