Using Temporal Probabilistic Rules to Learn Group Behavior

JOHN P. DICKERSON; GERARDO I. SIMARI; V.S. SUBRAHMANIAN

Handbook of Computational Approaches to Counterterrorism

SPRINGER

Lugar: New York - USA; Año: 2012; p. 245 - 266

Many applications require logical reasoning about situations that involve temporal uncertainty. This can be done with statements of the form "a formula G becomes true with 90% probability 4 time units after a formula F became true." In this chapter, we overview temporal probabilistic (TP) logic, through which programmers can formally express such rules that have both temporal and probabilistic aspects. Traditionally, logic programmers wrote temporal probabilistic rules manually; however, recent advances allow us to automatically learn the rules from historical data. We provide a general method to derive TP rules from databases of categorical and numerical variables. We conclude with a discussion of a successful, large-scale application of these techniques to model Lashkar-e-Taiba, an active militant terrorist group. Using standard integer programming techniques, we are able to provide policy suggestions based on these automatically learned rules. We also briefly discuss a recent extension to TP logic called annotated probabilistic temporal (APT) logic, which increases the expressiveness of the rules; we can easily extend our architecture to make use of this added power.