Resource-constrained Data Collection and Fusion for Identifying Weak Distributed Patterns in Networks
Final rept. 15 Jul 2010-14 Jul 2013
CARNEGIE-MELLON UNIV PITTSBURGH PA
Pagination or Media Count:
This project addressed the problems of detection, localization and estimation of weak and distributed patterns of activation in a large-scale network given access to direct, compressive and adaptive noisy node measurements. Precise information-theoretic limits were identified for these problems that provide necessary conditions on how the signal-to-noise ratio required scales as a function of the number of measurements, the graph size, connectivity and properties such as cut-size of the activated vertices, under a graph-structured normal means model. By leveraging highly inter-disciplinary tools from machine learning, statistics, signal processing and optimization, fast methods were developed that nearly achieve the information-theoretic limits, for general graph structures and classes of activation patterns. Development of such state-of-the-art methods that are both computationally and statistically efficient is crucial to advance AFOSRs ability to monitor, understand and secure modern large-scale networks that are vulnerable to covert attacks.
- Statistics and Probability
- Operations Research