Design and Application of Spiking Neural Networks Based on LIF Neurons

In recent years, brain-like computing has gained significant attention in artificial intelligence, driven by advances in neuroscience and technology. Spiking neural networks (SNNs), inspired by the principles of neuroscience, have emerged as a leading framework in this domain. Unlike traditional artificial neural networks, SNNs simulate temporal dynamics through spiking neurons, enhancing biological interpretability. However, existing research often focuses on single-layer or shallow-layer structures, limiting their applicability to simple classification tasks due to inefficient learning algorithms. To address these challenges, this paper introduces two novel SNN structures based on leaky integrate and fire (LIF) neurons. The first structure combines spike timing-dependent plasticity (STDP) with competitive learning in a two-layer network, improving classification accuracy to 83.179% on the MNIST dataset. The second innovation involves embedding spiking mechanisms into deep models, resulting in two structures, DSNN and convolutional DSNN. These deep SNNs demonstrate enhanced performance in image classification, achieving accuracy rates of 97.67% and 97.09%, respectively, on MNIST and Fashion MNIST datasets. Additionally, the proposed models exhibit shorter training times and increased efficiency, addressing challenges associated with spiking neural networks in deeper applications.