Graph Oriented Unsupervised And Supervised Class Distribution Learning And Experimental Research

The graph is a representation of how things or state relationships are expressed.Since many problems can be merged into graph problems,there are many algorithms related to graphs.The graph theory algorithm provides an effective,simple,and systematic way of modeling for many problems.The graph theory model can be analyzed and solved by matrix description and knowledge of linear algebra and matrix theory.Its expression form is concise but general,and it is easy to carry out in-depth theoretical analysis.Many problems can be transformed into graph theory problems,and then solved by the basic algorithm of graph theory.It has been widely used in unsupervised clustering and supervised learning of machine learning.The specific work of this paper is as follows:Due to its easy implementation,the graph-based relaxed optimization indeed provides an effective analytical solution for non-approximation iterative methods.However,due to the inverse of the matrix,such an optimization will run slowly and even become impractical for large-scale data.In this paper,we develop two general approaches for fast clustering based on graph relaxed optimization.One is based on k-means clustering,and the other is based on random projection tree.Extensive experiments show that these two proposed approaches can achieve significant acceleration without degrading the clustering accuracy a lot.In particular,our approaches have better clustering performance than the classical k-mean algorithm on largescale data,and run faster than the graph-based relaxed optimization clustering algorithms,with comparable accuracy.It is worth noting that the proposed approaches in this paper allow a single machine to cluster millions of data samples within minutes.At present,most existing facial expression recognition methods assume that each facial expression in the training set corresponds to a single emotion,and which is typically cast as a classification problem.However,in actual situations,a face expression rarely expresses pure emotion,but often a mixture of different emotions.Therefore,samples of similar expression have a correlation at the emotional level,and this correlation also often leads to ambiguity in the sample's expression labels.Namely,each emotion sample is associated with a latent label distribution.For this reason,we propose a totally data-driven label distribution learning approach to adaptively learn the latent label distributions,without any pre-hypothesized label distribution,we can obtain the relationship between each expression and its multiple emotions.This method can get the specific description of each emotion contained in the expression and the mapping of the expression image to the emotion distribution.Experimental results show that the algorithm has high accuracy in facial expression recognition and could effectively deal with the facial emotion analysis problem.