Multi-Hypothesis Abductive Reasoning for Link Discovery
Nicholas J. Pioch, Dan Hunter, James V. White, Amy Hiumin Kao, Daniel F. Bostwick, Eric K. Jones · 2004
Intelligence agencies are under increasing pressure to “connect the dots ” between fragments of evidence from disparate sources to enable preemption of potential threats such as terrorist attacks. Most systems for threat detection in use today provide only data visualization tools for manual “link analysis, ” leading to methods that do not scale to massive data sets. The CADRE system (Continuous Analysis and Discovery from Relational Evidence) addresses this deficiency by automating the link analysis process. CADRE combines an expressive knowledge representation of threat patterns with efficient, constraint-based abductive reasoning algorithms to automatically infer links and construct coherent threat hypotheses from structured data. A compact, factored representation of multiple hypotheses avoids redundant storage and enables scaling to large data sets. CADRE efficiently manages the growth of the hypotheses using probabilistic evaluation models and a consistency checking algorithm to prune unlikely hypotheses.