摘要： 现有基于梯度下降的脉冲神经元脉冲序列学习方法采用离线学习方式,即神经元运行结束后再根据全部输出脉冲调整突触权值。基于真实生物神经元突触强度的在线调整机制,提出一种在线方式的基于梯度下降的脉冲序列学习方法。在神经元运行过程中,当有脉冲激发时,即刻根据实际与期望输出脉冲调整突触权值。实验结果表明,该方法可有效实现脉冲序列学习,与离线的基于梯度下降的方法相比,具有更高的学习精度。

关键词: 脉冲神经元, 梯度下降, 在线学习, 脉冲序列学习, 脉冲反应神经元模型

Abstract: Now the existing spike sequence learning methods based on gradient descent are offline learning methods,i.e.the synaptic weights of neuron are adjusted according to the all output spikes after running of the neuron.Based on the online adjust mechanism of the real neuron synapse,an online spike sequence learning method based on gradient descent is proposed.The online method adjusts the synaptic weights according to the practical and corresponding desired output spike as soon as the neuron emits a spike.Experimental results show that the online learning method can realize the spike sequence learning effectively and has higher learning accuracy than the offline learning methods based on gradient descent.

Key words: spiking neuron, gradient descent, online learning, spike sequence learning, spike response neuron model

中图分类号: 

• TP391