非线性弱分类器的存在性

Existence of nonlinear weak classifier

S.Mannor和R.Meir将分形几何中的几何差异性引入了弱分类器存在性的研究,最终将线性弱分类器的存在性归结为样本集上的某些几何性质,从而仅凭样本集的信息,就可以判断弱分类器是否存在.在S.Mannor和R.Meir对线性弱分类器研究结果的基础上,利用核方法将输入空间上的非线性分类器与特征空间上的线性分类器联系在一起,得到非线性弱分类器的判定条件.

Many studies have shown that there is always a weak classifier under appropriate conditions.It is particularly striking that the geometric differences was introduced into the research of the existence of weak classifiers by S.Mannor and R.Meir and,finally,the existence of linear weak classifier was subsumed into some geometric properties on sample sets.Therefore,with the information of the sample sets,only,it could be determined whether the weak classifier exists.On the basis of investigation results of S.Mannor and R.Meir,the nonlinear weak classifier in input space was united together with the linear classifier in characteristic space,by using the kernel method so that a judgment condition of the existence of the nonlinear weak classifier was obtained.