A limit learning machine method based on maximum center cross-correlation entropy criterion

Abstract

The invention discloses a limit learning machine method based on maximum center cross correlation entropy criterion, At first, that input train data of the system is processed by the limit learn machine model based on the maximum center cross-correlation entropy criterion, with a larger core width, the output of the model is obtained, the actual output of the system and the output of the model arecalculated by difference, and the error of the model is obtained. The mean or median of the error is used as the center of the kernel, the maximum center cross-correlation entropy criterion is used,the weight vector of the fixed point is iterated, and the output of the model is obtained by the weight vector after the iteration is converged many times. The method of the invention has good universality, is suitable for non-Gaussian system or signal processing process, and has the performance of satisfying both high precision and fast parameter selection, solves the problems of non-robustness to non-Gaussian noise under the minimum mean square error criterion and non-ideal precision under the maximum cross-correlation entropy criterion under non-zero mean noise, and the time-consuming problem of choosing the core width by cross-validation.

Description

technical field [0001] The invention belongs to the field of robust machine learning and relates to an extreme learning machine method based on the maximum central cross-correlation entropy criterion. Background technique [0002] The machine learning model based on the maximum cross-correlation entropy criterion is a robust learning method developed in recent years. It has strong robustness and adaptability, so it is expected to be widely used in engineering practice, especially in information processing technology. [0003] Any actual information processing process has different degrees of uncertainty, and these uncertainties sometimes appear inside the process and sometimes appear outside the process. From the inside of the process, the structure and parameters of the mathematical model describing the research object, that is, the information dynamic process, are not necessarily known by the designer in advance. As the influence of the external environment on the inform...

AI Technical Summary

Problems solved by technology

However, when the initial error of the learning model iteration based on the maximum cross-correlation entropy criterion is relatively large, it will make it difficult for the accuracy of the method to achieve the desired effect.

Images