面向匹配决策问题的漏整合神经元稀疏ESN网络

Sparse ESN with a leaky integrator for matching decision-making problems

为了对匹配决策问题进行建模与预测,提出了一种具有更多神经生理学特征的稀疏回声状态网络(ESN),并基于在线监督学习方法对网络进行训练.为了评估网络的匹配决策性能,设计了三组测试数据集对网络性能进行测试,并提出了一种基于网络期望输出与实际输出序列最大相关系数的评价方法.仿真结果表明,新模型只需要较少的训练时间即可获得较好的决策性能,且对发放时间间隔、平移和网络噪声具有较好的鲁棒性.

A new sparse echo state network （ESN） with a leaky integrator, which is expected to has more neurophysiology characteristics, was proposed and trained using the online supervised learning method so as to make the modeling and prediction of the matching decision-making problem. To evaluate the matching decision-making performance of the network, three kinds of test datasets were set up and an estimation method based on the maximum correlation coefficient for the actual output and the desired one was present. Simulation experimental resuhs show that the proposed model can achieve a better decision-making performance with a less training time. Meanwhile the model has a better robustness on spiking interval change, shifting, and network noise.