基于非对称度量的分类器优化问题

Quasi-pseudo-metrics of measurable classifiers

研究在模式识别中可测分类器之间的非对称度量及优化问题。以分类器之间相对平均分类错误率建立分类器之间的非对称度量,进一步研究了子空间上的最优分类器及其估计。最后对模式识别中两个重要的例子在非对称度量空间上进行了讨论。

This paper studies the quasi-pseudo-metrics of measurable classifiers in pattern recognition. A quasi-pseudo-metrics of measurable classifiers in probability spaces is proposed, which is an average expense between classifiers. The optimal classifiers in subspace and their estimations are investigated, and two important examples are discussed in the space.