Schnelleinstieg Reader


Startseite FSU

Visually Guided Parameter Choice in Point Cloud Data Analysis

Joint project with Klaus Mueller at SUNY Stony Brook. Funded by the Deutsche Forschungsgemeinschaft within the priority program Scalable Visual Analytics.

Point cloud data in Euclidean space have become ubiquitous. Sometimes the point cloud data representation is genuine like in physical and bio-chemical measurements and simulations, otherwise given data like images, shapes or text can be easily transformed into point cloud data: an image can be represented as a vector of gray-scale or color values, and a text document can be represented as a term frequency vector that stores for each term (in a collection of terms) how often it appears in the document. Similar transformations are known for shapes, audio- and video data. Hence analyzing point cloud data in Euclidean space is a core task in machine learning and various algorithms have been developed to that means. Many of these algorithms have freely adaptable parameters whose setting can heavily influence the results of the analysis. Hence there is a well recognized need to explore the parameter settings systematically. Interestingly, there are quite a number of algorithms known for which applying the the algorithm for all parameter settings (computing the solution path parameterized by the parameter) is not much more expensive than computing a single solution. Much less is known on how to exploit this for choosing a good parameter setting or identifying interesting parameter regions. In this project we plan to develop and validate the following paradigm: compute the whole parameter solution path of an analysis algorithm and visually represent this path in a way that allows the data analyst to identify interesting parameter settings interactively.