Efficient Effects-based Military Planning with

Dynamic Influence Diagram

Sponsored by Army Research Office


Effects-based operations (EBO) has become an increasingly important warfighting concept for military planning, execution, and assessment.  Unlike the conventional destruction-centric and attrition-based approach, EBO is outcome-centric, focusing primarily on the achievement of a desired end state through the application of a full spectrum politico-military resources to conflict resolution. In this research, we propose to apply the EBO concept for military plan modeling and assessment.  Significant technical challenges, however, remain for the systematic implementation of this concept for military planning.  First, a major issue facing military planning is the huge amount of uncertainty. Uncertainties exist in the sensory observations, in the actions, in their outcomes, and in the effect assessment.  A probabilistic framework is therefore needed that can systematically capture these uncertainties, propagate them, and quantify their effects on a military plan.   Second, the utility of a military plan usually varies with time since the effects of its constituent actions vary with time.  It is therefore crucial to explicitly and dynamically model the actions, their outcomes, and their relationships. Third, action effects must be systematically propagated through time and through different levels of military structures in order to evaluate their impact on the campaign objective. Finally, for military planning, it is crucial to identify the optimal campaign strategy in a timely and efficient manner.

In this research, we propose methods to systematically tackle these issues. First, towards the first three challenges, we propose a unified probabilistic framework based on the Dynamic Influence Diagram to systematically represent the causal relationships between actions and their effects, their interactions, their uncertainties, and their dynamics.  In addition, the framework provides a mechanism for systematic propagation of the uncertainties and the action effects throughout the network over time.  Second, given the proposed framework, we introduce two complementary approaches aiming at achieving efficient military plan analysis. Specifically, we first propose a graph-theoretic approach to effectively reduce search space by eliminating a large number of unlikely plans from further consideration based on the synergies among the selected sensors. We then propose a factorization procedure to significantly reduce the evaluation time for each plan by factoring out the common computation so they only need be computed only once.   Below is an example Influence Diagram for Effects-based military planning






        We have developed a software that can easily construct an EBO model, automatically learn its parameters from training data and the related expert knowledge, and perform efficient evaluation and selection of different plans.  Below is an example EBO model constructed by the software. 




An example Influence Diagram for Modeling Effects-based Military Planning, where the leaf nodes represent the various military actions that can be included in a military plan as well as their respective utilities, the intermediate nodes representing the effects of certain military actions as well as means to evaluate the effects, and top node representing the goal of the military campaign. 





Results of ID evaluation using our software with the selected actions shaded green


Related Publications:

1)   Cassio de Campos,  Yan Tong,  Qiang Ji, Constrained Maximum Likelihood Learning of Bayesian

Networks, European Conference on Computer Vision (ECCV), 2008.


2)  Yan Tong and Qiang Ji, Learning Bayesian Networks with Qualitative Constraints, IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2008.


3)  Cassio de Campos and Qiang Ji, Improving Bayesian Network Parameter Learning using Constraints, International Conference in Pattern Recognition (ICPR), 2008.


4) Wenhui Liao and Qiang Ji, Exploiting Qualitative Domain Knowledge for Learning Bayesian Network Parameters with Incomplete Data, International Conference in Pattern Recognition (ICPR), 2008.


Above papers introduce different methods for learning a probabilistic graphical model (e.g. influence diagram) based on a combination of limited training data with any types of linear constraints derived from various sources including domain experts.



5)  Cassio de Campos and Qiang Ji, Strategy Selection in Influence Diagrams using Imprecise Probabilities, the 24th Conference on Uncertainty in Artificial Intelligence (UAI), 2008.


This paper introduces a new method for influence diagram evaluation to identify the optimal strategy (e.g. military plan)


6)  Weihong Zhang and Qiang Ji, A Factorization Approach To Evaluating Simultaneous Influence

Diagrams, IEEE Transactions on Systems, Man, and Cybernetics A, p746-754, Vol. 36, No. 4, July, 2006


This paper introduces a new factorization method for performing efficient influence diagram evaluation. 



        We are actively seeking collaboration from different fields, especially from the military, who are interested in evaluating and applying our model and methods to their specific applications.    If interested, feel free to contact Dr. Qiang Ji at qji@ecse.rpi.edu .