On the comparison of tests for optimization of adaptive pattern recognition systems

Neural network technologies for information systems security

“…Later it was found out that the method of the back-propagation in the West was developed by P. Werbos as early as in 1974 [8] and applied in a variety of applications, but for broad scientific community this method remained not unknown until the publication [7]. The first learning algorithms of the MNN have been proposed in the USSR in the papers [9][10][11][12][13][14] in the late 60s -early 70s. These results were systematized in the monograph [14] (1974), "Synthesis of multilayer systems for pattern recognition."…”

“…Note that in the Russian methodology learning algorithms of the MNN are essentially parallel, while the backpropagation algorithm is consecutive. In [9][10][11][12][13][14] not only learning algorithms with a quadratic optimization criterion have been synthesized and extensively studied, but also with other criteria, such as the modular one. The selection of the criterion which is different from the quadratic criterion is important because there are optimal optimization criteria for each density function of noise measurements [16].…”

“…The selection of the criterion which is different from the quadratic criterion is important because there are optimal optimization criteria for each density function of noise measurements [16]. Learning algorithms of MNN which were presented in [9][10][11][12][13][14] were based on the method of their synthesis. The structure of this method is shown in Fig.…”