Training Spiking Neural Networks with an Adaptive Leaky Integrate-and-Fire Neuron

Spiking neural network (SNN) is more biologically plausible than traditional artificial neural networks. Since the spiking network uses binary values of spike to process, it can offer an excellent power and energy efficiency when implementing it in hardware. Therefore, it is widely utilized in various machine learning applications, such as pattern recognition. In this paper, we introduce an adaptive leaky integrate-and-fire (LIF) neuron model that improves the accuracy of the spiking network. The proposed method is employed in a spiking network that includes more than 1,500 neurons to classify the MNIST handwritten digits. The unsupervised spike timing-dependent plasticity (STDP) learning rule is used to train the network. The experimental results are shown that the accuracy performance of the network with the proposed method outperforms the baseline spiking network.