Research On Pattern Recognition For New Learning Scenarios Combined With User Requirements

With the rapid progress of modern society and constantly updated technology,people explore more and more learning scenario from actual demands,followed by the corresponding learning methods in new learning scenarios.This paper explores how to use the transfer learning framework to solve the challenges brought by these new learning scenarios.In addition,most of the traditional clustering algorithms,including transfer clustering algorithm,can only get a single result,however,when dealing with the complex data,it is likely that there are multiple reasonable clustering results.This feature is particularly evident in high-dimensional data,such as text,image,gene data,etc.These data have many characteristics,and different feature subspaces tend to get completely different clustering results,and each result reflects the different structural information of the data.In summary,how to combine transfer learning to improve the traditional clustering algorithm,and get a lot of clustering results for users to choose,has become an urgent problem.In this paper,we study following research of transfer learning and multiple alternative clustering mining:1.A novel technique of maximum entropy clustering(MEC)based on knowledge transfer is studied in this paper.There are two challenges needing to be addressed:1)How can the knowledge be appropriately selected from the source domain so as to enhance the clustering performance in the target domain via transfer learning.2)How does transfer clustering conduct if the numbers of clusters between the source and target domains are inconsistent.To this end,a new transfer clustering mechanism,i.e.the cluster center-based matching transfer,is designed.Furthermore,the knowledge-transfer-based maximum entropy clustering(KT-MEC)algorithm is developed by incorporating this mechanism into classic MEC.Extensively experimental results reveal that the proposed KT-MEC algorithm has higher accuracy and better noise immunity than many existing methods on the application of texture image segmentation with different transfer scenarios.2.Current transfer learning model study the source data for future target inferences within a major view that the whole source data should be used to explore the shared knowledge structure.However,due to the limited availability of human ranked source domain,this assumption may not hold due to the fact that not all prior knowledge in the source domain is correlative to the target domain in most real-world applications.In this paper,we propose a general framework referred to discriminative knowledge-leverage(KL)based on generalized empirical risk minimization(GERM)transfer learning,where the ERM principle is generalized to the transfer learning setting.Additionally,we theoretically show the upper bound of the generalized empirical risk minimization for the practical discriminative transfer learning.The proposed method can alleviate negative transfer by automatically discovering useful objects from source domain.Extensive experiments verify that the proposed method can significantly outperform the state-of-the-art transfer learning methods on several artificial public datasets.3.Most clustering algorithms typically just find a single result of the data.Considering the complexity of the data is generally high,and allowing the data to be viewed from different perspectives,on the basis of ensuring the reasonableness,clustering results are often not unique.We present a new algorithm named RLPP for an alternative clustering generation.The objective of RLPP is to balance between clustering quality and dissimilarity by using subspace manifold learning technique through the new subspace such that a variety of clustering results are generated.Experimental results about both linear and nonlinear data sets show that RLPP successfully provides a variety of alternative clustering results,and able to outperform or at least be comparable to a range of existing methods.