Multi-View Support Vector Machines With Sub-View Learning

In recent years,with the continuous improvement of the level of data collection,storage and processing,the form of data is no longer single,and the same object often has multiple angles of feature description.How to mine information quickly and effectively has become a key problem to be solved.By simply splicing the feature data of each perspective,it is easy to produce the problem of overfitting when the feature dimension is too large.In order to make better use of multi-view feature data,a large number of multi-view learning methods have been proposed.These methods follow the principles of consistency and complementarity of multi-view learning,but do not deeply consider the hidden features inside the view.Based on this,this paper first proposes a sub-view learning strategy,expands the parallel structure of multi-view learning to a hierarchical structure,and extends the model to a generalized multi-view solution,introducing a multi-dimensional fruit fly optimization algorithm to solve its parameter optimization problem.Classification experiments are carried out in multiple data sets,and the effectiveness of the proposed method is proved from the perspective of multiple evaluation indicators.The innovation of this paper mainly includes :(1)A sub-view learning method based on easy-to-implement and effective use of hidden features and data structure features in multi-view data is proposed.Two new classification models SL-PSVM-2V and SL-MCPK with sub-view structure are constructed respectively,and the solution strategy based on quadratic programming is given.Compared with several classical multi-view learning models,the conformity with the consistency principle and the complementarity principle is analyzed,and the transformation methods between the models are summarized.The effectiveness of the new method and the classification performance of each data set are verified by multiple sets of experiments,and the performance of each classification method in the noise data set is compared.At the same time,a non-parametric test was performed to verify the significant differences between the models.(2)Based on the two-view model,a generalized multi-view support vector machine model based on sub-view learning is constructed.An improved multi-dimensional fruit fly optimization algorithm is proposed,and a multi-dimensional search strategy and an improved step size scheme are formulated to reduce the probability of the results falling into local optimum.In multiple experiments,the feasibility and effectiveness of the proposed model are proved by comparing with the benchmark model,and the influence of the population size and the number of iterations in the multi-dimensional fruit fly optimization algorithm on the performance of the model is analyzed.