Input-Anticipating Critical Reservoirs Show Power Law Forgetting of Unexpected Input Events
Input-Anticipating Critical Reservoirs Show Power Law Forgetting of Unexpected Input Events
复制标题
DOI：
10.1162/neco_a_00730
复制
发表时间：
2014-04
影响因子：
2.9
通讯作者：
N. Mayer
中科院分区：
文献类型：
--
作者：
N. Mayer
通常，储备池计算呈现指数型记忆衰减。这封信研究了回声状态网络在何种情况下会呈现幂律遗忘。这意味着在很长的时间跨度内，早期事件的痕迹都能在储备池中被发现。这样的设置要求临界连接性恰好处于根据回声状态条件所允许的极限。然而，对于一般的矩阵，从理论上无法精确确定这个极限。此外，网络的行为受到输入流的强烈影响。文中给出了使用某些类型的受限递归连接性以及针对输入的预期学习的结果，在这些情况下确实可以实现幂律遗忘。
Usually reservoir computing shows an exponential memory decay. This letter investigates under which circumstances echo state networks can show a power law forgetting. That means traces of earlier events can be found in the reservoir for very long time spans. Such a setting requires critical connectivity exactly at the limit of what is permissible according to the echo state condition. However, for general matrices, the limit cannot be determined exactly from theory. In addition, the behavior of the network is strongly influenced by the input flow. Results are presented that use certain types of restricted recurrent connectivity and anticipation learning with regard to the input, where power law forgetting can indeed be achieved.