大数据系列讲座之四： Discovering Drivers of Change in Spatial Systems Through Association Pattern Mining
讲座名称：
大数据系列讲座之四： Discovering Drivers of Change in Spatial Systems Through Association Pattern Mining
讲座时间：
2013-06-17
讲座人：
丁薇
形式：
校区：
兴庆校区
实践学分：
讲座内容：
大数据系列讲座之四： Discovering Drivers of Change in Spatial Systems Through Association Pattern Mining
题目：Discovering Drivers of Change in Spatial Systems Through Association Pattern Mining
主讲人：丁薇 教授 Department of Computer Science University of Massachusetts Boston
时间：2013年6月17日(周一）上午10：00-11：30
地点：管理学院313室
主持人：吴锋 教授
讲座内容：
Advances in gathering spatial data and progress in Geographical Information Science (GIS) allow analysts to monitor and to model dynamics of complex spatial systems pertaining to geographic, economic, ecological, and natural hazards domains. Predictive models of change help to guide economic and political decisions of high social significance. This project seeks to contribute to the science of change analysis by developing a new paradigm of how spatial change is analyzed and predicted. Specifically, the project aims at creating a suite of change analysis tools that are based on data-centric and model-free foundation of association analysis instead of model-oriented tools that are presently used by change analysts. The deliverables of this project are novel algorithms that enable comprehensive and efficient analysis of major factors driving change in spatial systems and provide more accurate prediction of future change. They also represent extension to the present state-of-the-art in association analysis techniques.
The project has an original goal and it proposes novel approaches and algorithms to achieve its purposes. The bulk of existing work on change analysis concentrates on modeling and predicting change; establishing major drivers of change is treated as a byproduct of change prediction. This project proposes a new paradigm in which comprehensive discovery of change drivers through association analysis is a central topic, and change prediction is a byproduct of the discovery process. Novel solutions are required to develop tools based on the new paradigm: (i) Change drivers are not discovered individually, but rather as discriminative patterns of multiple factors; modification of discriminative pattern mining technique to spatial datasets with ambiguous labels is proposed. (ii) Large number of change patterns need to be synthesized into more comprehensive form of knowledge; development of novel pattern similarity measure that enables such synthesis is proposed. (iii) Change prediction is based on discovered patterns of change; a technique of classification by aggregating emerging discriminative patterns is proposed to be applied to spatial datasets. Two application case studies, one pertaining to rural-urban land conversion and another to modeling hurricane risk assessment, are an integral part of this project, designed to demonstrate advantages of the new methodology for analyzing change in spatial systems.
相关视频