Enhancing SNN Performance by Using Dynamic Membrane Threshold and Inhibitory Block Division

Spiking Neural Networks (SNNs) are artificial neural networks inspired by the functioning of biological neurons. Unlike traditional neural networks, SNNs communicate through discrete, time-based signals called spikes, enabling parallel and energy-efficient information processing. While SNNs can capture biological properties that are not feasible with traditional Artificial Neural Networks (ANNs), existing learning methods for SNNs suffer from either poor performance or unreliable biological plausibility. In this paper, we propose an improvement approach that enhances performance by increasing the biological plausibility. Specifically, we divide the inhibitory layer into multiple blocks, allowing excitatory neurons to only inhibit neighboring neurons. In the training of SNNs, there is a reduction in spike activity with each layer, leading to suboptimal performance in the final classification. To address this, we incorporate the dynamic threshold method, ensuring complete spike emission in SNNs. We evaluate the performance of our method on the MNIST dataset, demonstrating its superiority in terms of accuracy compared to state-of-the-art methods. Our proposed approach provides a promising solution to the problem of incomplete spike emission in SNNs and offers potential for improved performance and efficiency in various SNN applications.