Performance Metrics for the Global Nuclear Detection Architecture Abbreviated Version (2013) / Chapter Skim
Currently Skimming:
Appendix D: Model-Based Approaches for the Gnda
Appendix D: Model-Based Approaches for the Gnda
Pages 75-88
The Chapter Skim interface presents what we've algorithmically identified as the most significant single chunk of text within every page in the chapter.
Select key terms on the right to highlight them within pages of the chapter.
From page 75... ...
D.2 RATIONALE FOR GNDA MODELING Mathematical models are developed for many different reasons. For example, sometimes models are derived as compact and precise statements of basic truths (e.g., physics)
|
From page 76... ...
Constructing such models serves the purpose of linking GNDA resource inputs and program activities to the primary outcomes of interest -- interdiction and the risk consequences of failing to detect radiological or nuclear material. Second, such models can help identify appropriate performance measures for evaluating the effectiveness of the GNDA by identifying (via model analysis)
|
From page 77... ...
Traditional game theory models adopt a "worst case" viewpoint that essentially grants adversaries perfect foresight, while less pessimistic approaches presume that potential attackers know some things but not others about GNDA activities (or know about resource deployments of different assets with different probabilities)
|
From page 78... ...
Such models are descriptive in nature, are meant to help understand how the system in question actually works, and also serve as building blocks for downstream decision-oriented models that address resource allocation or other issues. In thinking about the GNDA, one set of descriptive models would seek to answer the following basic question: Given a particular physical deployment of agents and sensors in a particular setting (e.g., a port, border crossing, along a highway)
|
From page 79... ...
. D.5.3 False Positive Versus False Negative Errors When we model sensors we need to consider false negative and false positive errors.
|
From page 80... ...
false detection; that is, the sensor indicates detection when the material is not present in sufficient quantity. While GNDA is primarily concerned with minimizing false negative errors for preventing the illicit transport of nuclear or radiological material, false positive errors can significantly increase the detection cost and impose a burden on the organization whose vehicle or container created the a false detection.
|
From page 81... ...
Continuing with our simple example, suppose that a geographic area is subdivided into two zones A and B, and the participating GNDA agency is trying to decide how many of its 10 sensors it should deploy in zone A versus zone B From current intelligence assessments, the agency believes that if an adversary were to attempt to bring illicit nuclear or radiological material into the area of interest, there is a conditional probability a that entry would occur in zone A and a complementary conditional probability b = 1 – a that entry would occur in zone B
|
From page 82... ...
To maximize cost-effectiveness in this case means to minimize the cost per detected infiltration, which of course is achieved by maximizing the probability of detection. The example shows that placing three sensors in zone A and seven in zone B maximizes detection probability at 70 percent, thus, the cost per detected event equals $1M/.7 = $1.43 million per case de
|
From page 83... ...
FIGURE D-4 Detection probability in two-zone, 10-sensor example illustrating the sensitivity of detection probability to intelligence estimates of probability of attack within a particular zone.
|
From page 84... ...
. This turns out to be a much more general proposition -- when defending against intelligent adversaries, worst-case analysis requires defenders to minimize the terrorists' maximum probability of success (or more generally the expected risk consequences of terrorist success including, for example, morbidity, mortality, economic, and political damage)
|
From page 85... ...
, so the probability of detection would shrink below 50 percent. D.5.7 Extension to Risk Consequences and Randomization Defense To illustrate how the ideas above extend beyond the probability of de tection to risk consequences, consider Table D-2.
|
From page 86... ...
. D.5.8 Resource Allocation Modeling In the examples above, the allocation decisions faced by defenders all involved the placement of different numbers of otherwise equivalent sensors or agents in different zones.
|
From page 87... ...
To see how the same resource allocation logic can apply when there are multiple resources with different unit costs available for GNDA use, suppose that there are some number m of different resource types (e.g., sensors with different costs and different sensitivity and specificity [equivalently different likelihoods of committing false negative and false positive errors as discussed previously]
|
From page 88... ...
A proposed resource allocation plan as implied by assigned numerical values of the decision variables is said to be feasible if its total cost resides within the available budget and if no other constraints on resource availability are violated. Now, in similar spirit to the simple models discussed earlier, the likelihood of interdicting an attempted infiltration with nuclear material out of regulatory control (or more generally the expected risk consequences of any terrorist infiltration plan)
|
Key Terms
This material may be derived from roughly machine-read images, and so is provided only to facilitate research.
More
information on Chapter Skim is available.