Abstract: Spiking network is a kind of neural network model based on biological neuronal behavior. Its input and output are firing times of spikes and its working mechanism is more like the biological neural network. Information transfer between spiking neurons is encoded in the timing of individual spike times and due to this temporal coding its learning performance is better. The time characteristic of spiking neuron leads to its complex working mechanism, while the sensitivity of spiking neuron which reflects the perturbation of output caused by input perturbation can be used as a tool of researching on inner working mechanism of spiking neuron. Unlike the traditional neural network, the definition of sensitivity of spiking neuron is the ratio of the number of changed output spike times to the running time of the neuron and it can reflect directly the influence of input perturbation. The analysis of the sensitivity caused by different input perturbation shows the sensitivity of spiking neuron is complicated. The sensitivity is a constant when all the synapses are perturbed equally and different perturbed synapses will cause different sensitivity.

Key words: spiking neuron, sensitivity, synapse, perturbation

摘要： 脉冲神经网络是一种基于生物的网络模型，它的输入输出为具有时间特性的脉冲序列，其运行机制相比其他传统人工神经网络更加接近于生物神经网络。神经元之间通过脉冲序列传递信息，这些信息通过脉冲的激发时间编码能够更有效地发挥网络的学习性能。脉冲神经元的时间特性导致了其工作机制较为复杂，而spiking神经元的敏感性反映了当神经元输入发生扰动时输出的spike的变化情况，可以作为研究神经元内部工作机制的工具。不同于传统的神经网络，spiking神经元敏感性定义为输出脉冲的变化时刻个数与运行时间长度的比值，能直接反映出输入扰动对输出的影响程度。通过对不同形式的输入扰动敏感性的分析，可以看出spiking神经元的敏感性较为复杂，当全体突触发生扰动时，神经元为定值，而当部分突触发生扰动时，不同突触的扰动会导致不同大小的神经元敏感性。

关键词: 脉冲神经元, 敏感性, 突触, 扰动