On effects of IP improvement of ESN reservoirs for reflecting of data structure
On effects of IP improvement of ESN reservoirs for reflecting of data structure
复制标题
DOI：
10.1109/ijcnn.2015.7280703
复制
发表时间：
2015-07
期刊：
影响因子：
--
通讯作者：
P. Koprinkova-Hristova
中科院分区：
文献类型：
--
作者：
P. Koprinkova-Hristova
本文对有关回波状态网络（ESN）储层固有塑性（IP）改善效果研究的一些结果进行了总结和评论。据观察，IP 训练以一种可用于聚类目的的方式将输入数据结构“捕获”到储存库稳态中。为了解释这些结果，引入了储层平衡状态记忆容量（MCRES）。使用三个基准人工数据集对不同规模的 ESN 储层的 IP 调谐 MCRES 进行了研究。
In the present paper some results concerning the investigation of effect of Intrinsic Plasticity (IP) improvement of Echo state networks (ESN) reservoirs were summarized and commented. It was observed that IP training “captures” input data structure into the reservoir steady state in a way that could be useful for clustering purposes. In search of explanation of these results the Memory Capacity of Reservoir Equilibrium State (MCRES) was introduced. Achieved due to IP tuning MCRES of ESN reservoirs with different sizes was investigated using three benchmark artificial data sets.