III: Small: A Scalable Search Tool for Interesting Patterns in Scientific Data

点击翻译按钮获取中文摘要

Across many scientific disciplines, the availability of very largeamounts of data is creating a paradigm shift. The goal of this projectis to develop Scolopax, a tool for finding interesting patterns inclassification and prediction models trained on large high-dimensionaldata. These patterns can capture previously unknown relationshipsbetween the variables of a complex process, hence are essential forexploratory analysis and scientific discovery.This project explores several research directions to lay thefoundations for Scolopax: (1) Design of a new query language that canexpress all common pattern search preferences. (2) Algorithms forlearning a query so that even non-technical users can formulatenon-trivial queries through an interactive process. (3) New rewriterules and efficient data management approaches to automaticallytransform queries into fast implementations on a cluster or Cloud. (4)Design of new semi-parametric data mining techniques that are amenableto scalable training, evaluation, and pattern confidence computation.User-friendly query writing functionality makes Scolopax accessible toscientists and citizen scientists alike. Its planned deploymentthrough popular Web sites, e.g., those hosted by the Cornell Lab ofOrnithology, has the potential to enable new scientificdiscoveries. By letting citizen scientists not only contribute data,but also make their own discoveries using the data, Scolopax alsoserves as an important enabler and motivator for outreach programs andgreater involvement of citizen scientists. For further information seethe project web site at the URL:http://www.ccs.neu.edu/home/mirek/Projects/Scolopax