Research On Multi-view Machine Learning Classification And Clustering Algorithms

In the real world,data usually has multiple feature representations or data sources,each representation or source is referred to as a particular view of data.There are comple-mentary and compatible information between different data views,and how to integrate these view information and make reasonable decisions is called multi-view learning.In machine learning,if you use only the information of one view to learn,it will be the same as”seeing a leopard in sight and seeing only one thing.”You can only get one-sided information and cannot make correct decisions.Compared to learning using only a single view,multi-view learning can synthesize information from multiple views,thus making the information learned more comprehensive.At present,multi-view learning has become a research hotspot in academia and has been widely used in medical image analysis,natural language processing,face recognition and other fields.This paper main-ly studies several key technologies such as multi-view data construction,multi-view data fusion,multi-view data classification and clustering algorithms.The main contribution and innovation of this article are summarized as follows:?1?A multi-view clustering method based on extreme learning machine is pro-posed.In this paper,the Extreme Learning Machine?ELM?is introduced into multi-view clustering task,and a multi-view clustering framework based on ELM is proposed.Based on this framework,we implement three algorithms.In this framework,the single-view normalized features are mapped to high-dimensional feature spaces through ELM random mapping to obtain better data feature representations.Later,we conducted unsupervised multi-view clustering in this feature space.As far as we know,this is the first work that adopts ELM to solve multi-view clustering problem.A large number of experiments show that the proposed algorithm achieves a significant performance improvement over the multi-view clustering method proposed in the literature in recent years.?2?A multi-view clustering method based on local kernel alignment is proposed.In order to solve the problem of data lack of view,this paper proposes a multi-view feature representation using the ELM to construct data,and mines the local kernel-aligned nature among multiple views of the data,and proposes a multi-view clustering method.In this method,a view is constructed using a random feature map of the ELM,and different views correspond to different hidden layer nodes.Then,experiments were conducted to analyze the complementarity and compatibility between these views.Based on this,a multi-view clustering method based on local kernel alignment is proposed.This method has good versatility and extensibility,and can be applied to single-view data learning.The experimental results show that the proposed algorithm can effectively improve the clustering results compared with the contrast algorithm.?3?An l_q norm sample adaptive multi-kernel learning algorithm is proposed.The formulation of existing Sample-adaptive Multiple Kernel Learning?SAMKL?algo-rithm falls to al₁-norm MKL which is not flexible.And,it is restricted to solve MKL problems with pre-specified kernels.To allow for robust kernel mixtures that general-ize well in practical applications,we extend MKL to arbitrary norm and apply it to image classification.In this paper,we formulate a closed-form solution for optimizing the kernel weights based on the equivalence between group-lasso and MKL,and derive an efficientl_q -norm?q?1 and denoting thel_q -norm of kernel weights?SAMKL algorithm.Cutting plane method is used to solve this margin maximization problem.Besides,we propose a framework for solving MKL problems in image classification.Experimental results on multiple data sets show the promising performance of the proposed solution compared with other competitive methods.?4?A feature fusion-based deep clustering method is proposed.Image clustering is one of the challenging tasks in machine learning and has been widely used in various applications.Recently,various depth clustering methods have been proposed by the aca-demic community.These methods generally use a two-stage learning method,sequential or combined use feature learning and clustering.We observe that these tasks mainly fo-cus on the combination of input reconstruction loss and clustering loss,and only a few work studies have further improved the feature representation capability in neural net-work clustering.In this paper we propose a Deep Convolutional Embedded Clustering with Inception-like block?DCECI?using Incept-like blocks.Specifically,an Incept-like block with different types of convolution filters is introduced in a symmetric deep con-volutional network to preserve the local structure of the convolutional layer and to fuse different nonlinear features.In this method we simultaneously minimize the input re-construction error and clustering loss of the convolutional autoencoder.Compared with other comparison algorithms,our proposed method obtains better experimental results on multiple image data sets,which proves the effectiveness of the proposed method.