基于回路神经网络的特征子空间估值算法(英文)

Algorithms of Eigensubspace Estimation Based on Discrete Recurrent Neural Networks

基于运用回复式离散神经网络进行特征子空间估值的理论,提出了解决正定对称矩阵的特征子空间估值问题的算法。其神经网络被描述为离散时间系统,它们在整个连续时间神经网络模型的数字化处理即计算机仿真方面具有优势,可以很容易地应用于数字化硬件。仿真结果的给出进一步阐明了网络良好的性能。

Based on the eigensubspace estimation using discrete recurrent neural networks, we propose algorithms to solve the problem of eigensubspace estimation for positive definite symmetric matrix. Neural networks are formulated as discrete time systems, they have advantages for computer simulations over digital simulations of continuous time neural network models. Thus they can be easily implemented in digital hardware. Simulation results are given to show the performance of networks.