循环神经网络的记忆：我们计算得正确吗？（英文版）

英文标题：Memory of recurrent networks: Do we compute it right?中文摘要：探讨了循环神经网络的记忆容量问题，对于线性回声状态网络，提出了通过利用记忆容量对输入掩码矩阵具有中性的结果，利用鲁棒的数值方法解决了记忆的不准确性。英文摘要：Numerical evaluations of the memory capacity (MC) of recurrent neuralnetworks reported in the literature often contradict well-establishedtheoretical bounds. In this paper, we study the case of linear echo statenetworks, for which the total memory capacity has been proven to be equal tothe rank of the corresponding Kalman controllability matrix. We shed li