基于脉冲序列合成核的脉冲神经元在线监督学习算法

Online Supervised Learning Algorithm for Spiking Neuron Based on Spiking Sequence Composite Kernel

脉冲神经网络使用时间编码的方式进行数据处理,是进行复杂时空信息处理的有效工具。为此,将多脉冲序列合成核引入脉冲序列处理过程,提出一种在线监督学习算法,采用累加和累积合成核机制进行实验学习,并与基于单一核函数的在线PSD算法进行比较。实验结果表明,该算法具有较好的学习性能,特别在数据样本较大时优势更为突出。同时结果也表明,通过多个核函数的组合可以获得更稳定高效的脉冲序列合成核表示。

Spiking neural network uses temporal coding for data processing,which is an effective tool for complex spatial and temporal information processing. In view of this, this paper applies multiple sequence composite kernel into the spiking sequence processing and proposes an online supervised learning algorithm. It uses accumulation and cumulative mechanisms for supervised learning and makes an experiment compared with online PSD algorithm based on single kernel. Experimental results show the better performance of the proposed algorithm. Especially in the performance of larger data sample, it can maintain this excellent performance more significantly. The results also show that the combination of multiple kernel functions can get more stable and efficient spiking sequence composite kernel representation.