ALSA: Associative Learning Based Supervised Learning Algorithm for SNN.

Area of Science:

• Computational Neuroscience
• Artificial Intelligence

Background:

• Spiking Neural Networks (SNNs) mimic biological brains but face training challenges due to non-differentiable spikes.
• Existing supervised learning methods for SNNs are limited, hindering their practical application.

Purpose of the Study:

• To propose a biologically plausible supervised learning method for SNNs.
• To address the training limitations of SNNs by leveraging associative learning principles.

Main Methods:

• Developed the Associative Learning for Supervised Artificial Neural Networks (ALSA) method.
• Implemented improved Spike-Timing-Dependent Plasticity (STDP) rules combined with a teacher layer.
• Simulated associative learning akin to animal conditioned reflexes for synaptic plasticity.

Main Results:

• ALSA achieved 95.7% accuracy on the IRIS dataset.
• ALSA demonstrated 91.58% accuracy on the MNIST dataset.
• The results validate ALSA as a feasible supervised learning approach for SNNs.

Conclusions:

• ALSA provides a biologically plausible and effective supervised learning framework for SNNs.
• The method successfully integrates STDP rules with associative learning mechanisms.
• This work advances SNN training methodologies, enhancing their potential for complex tasks.