Visualization Of High Dimensional Data Based On Optimal Transmission

With the development of technology and hardware,data collection becomes easier,database scale becomes larger and larger,and the complexity becomes higher and higher.A big problem facing human beings is how to get useful information fro m complex and high-dimensional data.As we all know,the visual range that human eyes can recognize is also a three-dimensional coordinate space system,so high-dimensional data is unfamiliar to human beings.The key to solve this problem is dimension reduction visualization technology.Dimensionality reduction visualization technology,as an effective combination of machine learning algorithm and human cognition and experience,presents knowledge to people but hands decision-making power of information knowledge to human.This thesis proposes a new theory based on the optimal quality transmission of high-dimensional data dimension reduction visualization algorithm,combined with quality driven topological perception curve skeleton extraction algorithm.In this thesis,on the basis of the skeleton extraction algorithm was improved,make it suitable for the skeleton extraction of high-dimensional data and a specified number of the output frame.For high-dimensional data dimension reduction algorithm,this thesis will be of high and low dimensional data abstraction for gaussian probability matrix,and USES the symmetrical SNE thoughts,make the probability matrix of the diagonal are equal,to avoid the punishment of outliers value is too big,after using out distance instead of KL divergence as energy function,and to add energy function a fter global distance of the regularization,the dimension reduction algorithm of this thesis pays more attention to in the distance of the outliers and maintain the basic shape of the global.This visualization method for dimension reduction algorithm and skeleton extraction algorithm to cooperate with each other,intentions to skeleton to help users better understand the classification,at the same time guarantee the global dimension reduction algorithm users better understand the global effect of the distance.This thesis shows the visualization effect of our visualization method on different 3D and high-dimensional data sets,and compares it with the basic linear and nonlinear dimensional reduction algorithms,which shows the effectiveness of the visualization method in this thesis.