脉冲神经网络:模型、学习算法与应用

脉冲神经网络是目前最具有生物解释性的人工神经网络,是类脑智能领域的核心组成部分.首先介绍各类常用的脉冲神经元模型以及前馈和循环型脉冲神经网络结构;然后介绍脉冲神经网络的时间编码方式,在此基础上,系统地介绍脉冲神经网络的学习算法,包括无监督学习和监督学习算法,其中监督学习算法按照梯度下降算法、结合STDP规则的算法和基于脉冲序列卷积核的算法3大类别分别展开详细介绍和总结;接着列举脉冲神经网络在控制领域、模式识别领域和类脑智能研究领域的应用,并在此基础上介绍各国脑计划中,脉冲神经网络与神经形态处理器相结合的案例;最后分析脉冲神经网络目前所存在的困难和挑战.

Spiking neural networks: Model, learning algorithms and applications

Spiking neural network is the most biologically plausible neural network model, which is the core component of brain-inspired intelligence. Firstly, the most commonly used spiking neuron models and the structures of feedforward and recurrent spiking neural networks are introduced, and the temporal coding method of spiking neural networks is introduced. Then, the learning algorithms of spiking neural networks systematically, including the unsupervised learning and supervised learning algorithms are introduced. The supervised learning algorithm follows the gradient descent algorithm, the algorithm based on STDP rules and the algorithm based on spike trains convolution rules are introduced and summarized in details. Subsequently, this survey gives application examples of spiking neural networks in the fields of control, pattern recognition and brain-inspired intelligence. Some case studies of spiking neural networks and neuromorphic processor in various national brain initatives are presented. Finally, this survey discusses the current difficulties and challenges of spiking neural networks.