Visualizing and Exploring High-dimensional Data

该项目的目的是开发新的方法，以交互式地探索大型高维数据集中的关系，例如那些典型的高通量科学实验。由此产生的工具将为科学家在应用传统的离线数据分析技术（如聚类、分割和分类）之前提供帮助。科学家将能够探索假设，并结合自己的知识，将传统的无监督数据挖掘算法推向更明智、更有前途的方向。可视化工具将帮助许多学科的科学家，包括研究基因功能的生物学家，理解疾病易感性的医生，开发候选药物的化学家，以及分析粒子加速器产生的数据的高能物理学家。这种新方法的一个关键组成部分是能够交互式地探索参数空间和组合高维数据点的属性。可视化工具将提供数据集的两种替代视图：不同矩阵视图，提供对集群的大小、紧凑度、分离和相对接近度的见解；点云视图，提供高维源数据的3-D投影，最好地保留点之间的距离。随着参数的调整，这种双视图方法在通信从一个集群到另一个集群的点的流和迁移方面表现出色。它还允许用户探测数据并与之交互，包括手工聚类数据和检查特定点等任务。随着各种数据集特性的贡献被交互式修改，由此产生的可视化工具将支持动态集群形成和迁移。该项目提供了一个优秀的跨学科教育和研究环境，项目的合作性质也增强了成果传播的潜力。项目网站（http://cs.unc.edu/~tynia/HiDimViewer/index.html）提供了对出版物和可视化工具的访问。

The aim of this project is to develop new methods for interactively exploring relationships within large high-dimensional data sets, such as those typical of high-throughput scientific experiments. The resulting tools will provide an aid to scientists prior to applying traditional offline data-analysis techniques such as clustering, segmentation, and classification. Scientists will be able to explore hypotheses and incorporate their own knowledge to drive traditional unsupervised data-mining algorithms in sensible and more promising directions. The visualization tools will assist scientist in many disciplines, including biologists in studying gene function, medical doctors in comprehending disease susceptibility, chemists in developing candidate drugs, and high-energy physics in analyzing the data generated by particle accelerators. A key component of the novel approach is the ability to interactively explore parameter spaces and combine attributes of high-dimensional data points. The visualization tool will provide two alternate views of the data sets: a dissimilarity-matrix view that offers insights into the size, compactness, separation, and relative proximity of clusters, and a point-cloud view that provides a 3-D projection of the high-dimensional source data that best preserve the distance between points. This dual-view approach excels in communicating the flow and migrations of points from one cluster to another as parameters are tuned. It also allows the user to probe and interact with the data, including such tasks as hand clustering the data, and examining particular points. The resulting visualization tools will support dynamic cluster formation and migration as the contributions of various data set features are interactively modified. The project provides an excellent interdisciplinary education and research environment, and the collaborative nature of the project also enhances the potential for results dissemination. The project Web site (http://cs.unc.edu/~tynia/HiDimViewer/index.html) provides access to publications and visualization tools.