We introduce a method to train a bio-inspired neural network model, having the characteristics of spiking-timing-dependent interaction and learning, in a manner of supervised learning. We assume the spiking neural network model has the tendency to obey the charge conservation principle or the junction rule on a long (or the learning dynamics) time scale. The tendency makes the distribution of connectivities is determined depending on not only the incoming stimuli to input neurons but also the outgoing stimuli from output neurons as if a solution of the finite elementary method in a fluid system. We apply the learning method to several cases in simulations and find the adoption of the conservation principle exerts desired effects on the neural network learning. Finally, we discuss the significance and the drawbacks of the introduced method and compare it with the supervised learning method implemented by the artificial neural network model.

我们介绍了一种以监督学习的方式训练仿生神经网络模型的方法，该模型具有脉冲时间依赖的交互和学习的特征。我们假设尖峰神经网络模型在长（或学习动态）时间尺度上倾向于遵守电荷守恒原理或结点规则。这种趋势使得连通性的分布不仅取决于对输入神经元的传入刺激，还取决于来自输出神经元的传出刺激，就好像流体系统中有限基本方法的解决方案一样。我们将学习方法应用于模拟中的几种情况，并发现采用守恒原理对神经网络学习产生了预期的影响。最后，