Learning Compositional Simulation Models
Abstract
Effective and proactive decisions about intelligence gathering depend on accurate models of an adversary. Specifically, such models need to accurately reflect the cause-and-effect dependencies within the systemic behavior of the adversary. Such models can be created based entirely on the knowledge of experts, or they can be created or augmented based on the analysis of data. However, creating causal models from data will require advances in the fundamental science and technology of discovering causal knowledge. Our project focused on creating such advances. Specifically, we focused on automating the application of quasi-experimental designs, a set of manual analysis techniques developed by social scientists, economists, and medical researchers over the past four decades. Quasi-experimental designs (QEDs) are templates for causal discovery from observational (non-experimental) data. QEDs identify naturally occurring experiments that support inferences about causal dependencies within large bodies of observational data. Our work has shown that many potential designs exist for realistic tasks that those designs can increase the accuracy with which causal inferences can be made from small amounts of data, and that such designs can be automatically identified. This lays the groundwork for powerful tools with which analysts can examine observational data of complex organizations and system to improve their causal understanding of those systems.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 2010
- Accession Number
- ADA513896
Entities
People
- David Jensen
Organizations
- University of Massachusetts Amherst