EAGER: Models, analytics, and algorithms for data driven applications

Iowa State University, Ames IA

Investigators

Abstract

Data driven applications are becoming increasingly prevalent as our ability to collect data continue to increase and the cost continues to decrease. Examples of such applications include social networks (twitter, facebook, etc.), bioinformatics (gene regulatory networks, genomic sequence analysis), and environmental monitoring using wireless sensor networks. As more data become available, how to convert data to actionable information to guide decision is a pending problem that will have many applications. In addition to being "big", the data and underlying phenomena also exhibit time-variations which add to the difficulty of gleaning useful information from data. To cope with these challenges, this project will develop an flexible algorithmic framework using Dynamic Bayesian Networks, which can sufficiently describe the physical reality, and at the same time are expressive enough to allow for machine based learning, optimization, inference and adaptation to dynamical data. Such a computational framework can assimilate data continuously, adaptively, and reduce the data to information in a format that is logical, intuitive, and amenable to human interactions.Such features are particularly needed for large-scale, data-intensive engineering systems and networks such as structural health monitoring, surveillance and disease discovery using gene regulatory networks. Intellectual Merit: The proposed project will build on existing work on graphical models and explore the largely open area of learning dynamical graphical models based on measured data. Algorithms will be developed based on convex optimization formulation for online adaptation of the graphical models. Partially known structural information, and issues with noisy, incomplete, corrupted, or missing data will be examined. The proposed research will advance the status of the art of learning from dynamical data, and explore the many under-explored connections between signal processing, and information theory, statistics, and machine learning. Broader Impact: The proposed research will develop generic models and analytical tools for data based applications. It will also generate low-complexity algorithms that can be adapted to a large number of applications such as wireless sensor networks, gene-regulatory networks, and genomic sequence analysis. The research will allow data that are made available through digital technologies to be converted to information automatically by computers, which can then be understood and acted upon by humans. The educationalgoal of this proposed project is to efficientl integrate research with educational activities and to train both undergraduate and graduate students in interdisciplinary areas to produce next-generation engineers. Efforts will be made to invite women, underpresented, and minority undergraduate students to participate in the proposed research.