|
Activity Number:
|
294
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, July 31, 2007 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #308338 |
|
Title:
|
Probabilistic Plan Tracking and Detection for Intelligence Analysis
|
|
Author(s):
|
Sinjini Mitra*+ and Paul Cohen and Aram Galstyan
|
|
Companies:
|
The University of Southern California and The University of Southern California and The University of Southern California
|
|
Address:
|
Information Sciences Institute, Marina del Rey, CA, 90292,
|
|
Keywords:
|
Hidden Markov Model ; plan recognition ; tracking ; belief state
|
|
Abstract:
|
Plan recognition is the problem of inferring an agent's unobservable state of plans or intentions based on observations of its interaction with the environment. In this paper, we present a rigorous theoretical framework for detecting and tracking malicious plans based on Abstract Hidden Markov Models that uses a Dynamic Bayesian Network representation of the plan hierarchy. The problem is to determine the top-level policy along with those at the lower levels given the current sequence of observations by updating the belief state using the posterior distribution at each time point. Unlike most existing methods, our model does not assume the known identity of the agent and is capable of tracking a very big network of agents carrying out different types of transactions and accurately detecting groups that intend to cause harm. We test our method on a virtual society of agents, called Hats.
|
