Causality and Information Dynamics in Networked Systems with Many Agents (ARO 10.3)
Abstract
The proposed work is on identifying causality starting with observational time series data. The motivation for the work comes from the fact that traces of activity from networks -- communication, information, social, biological, etc. -- are available these days, but yet understanding the causal relationship is hard. The PI will work, in this STIR grant, will address approximate algorithms in the context of graphical models used in Machine Learning. The proposed approach is one of making the graph spare using certain L1 norms that preserves the principal components of original graph The PI will also investigate the impact of under sampling the time series data to ensure robustness of causality detection. The algorithms developed will be tested on financial and weather prediction data.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jan 12, 2017
- Source ID
- W911NF1610155
Entities
People
- Ravi Mazumdar
Organizations
- Army Contracting Command
- United States Army
- University of Waterloo