多核学习方法

On Multiple Kernel Learning Methods

多核学习方法是当前核机器学习领域的一个新的热点.核方法是解决非线性模式分析问题的一种有效方法,但在一些复杂情形下,由单个核函数构成的核机器并不能满足诸如数据异构或不规则、样本规模巨大、样本不平坦分布等实际的应用需求,因此将多个核函数进行组合,以获得更好的结果是一种必然选择.本文根据多核的构成,从合成核、多尺度核、无限核三个角度,系统综述了多核方法的构造理论,分析了多核学习典型方法的特点及不足,总结了各自的应用领域,并凝炼了其进一步的研究方向.

Multiple kernel learning is a new research focus in the current kernel machine learning field.The kernel method is an effective approach for non-linear pattern analysis problems.But in some complicated cases,researchers find that the kernel machines with a single kernel function can not meet some practical requirements such as heterogeneous information or unnormalised data,large scale problems,non-flat distribution of samples,etc.Therefore,it is an inevitable choice to consider the combination of kernel functions for better results.According to the composition of multiple kernels,the construction theories of multiple kernel methods are systematically reviewed,the learning methods of multiple kernel with the corresponding characteristics and disadvantages are also analyzed,and the respective applications are summarized from three aspects,which are the composite kernels,the multi-scale kernels,and the infinite kernels.In addition,the paper generalizes the conclusions and some new directions for future work.