SVM: Support vector machine , At present, it is widely used in classification

Feature fusion : It is mainly used to describe different feature fusion methods , The common way is early integration , That's what we've described before, stitching together features , Later integration will be mentioned later in this article

kernel function :SVM When the problem of linear indivisibility is encountered , The vector can be mapped to high dimensional space by kernel function , Linearly separable in high dimensional space

Multi core learning : In use SVM When training , It involves the selection of kernel function , For example, linear kernel ,rbf Nuclear and so on , Multi core is the fusion of several different cores to train . This method is a kind of post fusion , By taking different nuclear methods for different characteristics , Multiple cores are formed for different parameters , Then the weight of each core is trained , The best combination of kernel functions is selected for classification .

Give some base kernels, such as linear,Polynomial,RBF,Sigmoid, For each one , You can specify multiple sets of parameters , That is, there are a total of M individual base
kernels, We want to use their linear combination as the final kernel function . adopt training, We get each of the components in this linear combination kernel The weight of d(weight). The most classic is simpleMKL,GMKL,G Namely Generalized, The optimization method uses PGD(Projected
Gradient Descend). In order to improve the convergence effect ,Vishwanathan Again SPG-GMKL(Spectral Projected
Gradient), At the same time, a multi-core algorithm is proposed product combination .SPG-GMKL It is also regarded by the latecomers as state-of-art. 

MKL The classic implementation of SimpleMKL,Shogun,SPG-GMKL,SMO-MKL.

Technology