Interactive Data Visualization and Pattern Discovery with Mirage

摘要

Pattern discovery often occurs in studying correlations of data viewed from multiple perspectives. For high dimensional, low level sensory data such as images, it is especially important that the input can be examined in its raw form, along with features extracted according to the application's concern, plus all intermediate results-of manual or automatic analysis. To facilitate this, flexible and effective exploratory tools that can handle diverse data types, a wide range of objectives, large data volumes, and variable demands on speed are in critical need. I describe our experience with Mirage, a research software tool for open-ended interactive pattern discovery. Applications to analyzing photonics simulations and in the Virtual Observatory raised interesting challenges on the tool's organization and the data visualization algorithms. We describe our progress on designing effective modularization and connection interfaces to local or remote analysis libraries, and approaches to highlighting data relationships between high-dimensional spaces or across contexts of different resolutions and data types.