加权自适应模式识别系统的数学描述

Mathematical Description of Weighted Adaptive Pattern Recognition System

基于n元原理,采用加权训练思想,为克服经典自适应模式识别系统——WISARD的简单训练策略之不足而提出的加权自适应模式识别系统,具有自动消除伪特征、突出训练模式的固有特征、改进系统的分类性能筹优点。本文首次用数学手段描述了加权自适应模式识别系统,把系统的训练与分类过程抽象为对模式进行矩阵变换的过程。系统的分类行为由训练模式所建立的逻辑函数矩阵来表征。文中还对系统的性质作了初步探讨,指出只要合理地选取加权训练阀值,加权系统的分类性能不会弱于非加权系统。最后还给出了加权训练阀值的上界和下界。

Based on then -tuple method and the weighted training principle,the weighted adaptive pattern recognition system is proposed in order to overcome the disadvantages of the simple training strategy used in the typical adaptive patternrecognction system——WISARD. Tt can eliminate the pseudo-feature automatically,emphasize the inherent feature oftraining patterns, improve the recognition performance. In this paper is presented for the first time the mathematical description of the weighted adaptive pattern recognition system. The training and classifying procedures are expressed abstractly as the matrix transformation for patterns. The classification behaviour of the system is determined by the logical function matrix, which are built by the training patterns. Some preliminary properties of the system are also discussed in the paper. It is indicated that the recognition performance of the weighted system is much better than that of the non-weighted system if the weighted training thresholds are selected reasonably. Both uppper and lower bounds of the weighted training threshold are also given at the end of the paper.