The National Academies Logo
Research Associateship Programs
Fellowships Office
Policy and Global Affairs

Participating Agencies - AFRL

  Sign InPrintable View

Opportunity at Air Force Research Laboratory (AFRL)

Information Fusion and Receiver Operating Characteristic Manifolds


Air Force Institute of Technology

RO# Location
13.50.00.B7998 Wright-Patterson AFB, OH 454337542


Name E-mail Phone
Oxley, Mark Edwin 9372553636 x4515


The US Air Force (USAF) has many legacy sensor-exploitation systems and they understand how they work and behave, and how well they perform. The USAF is always seeking ways to improve the performance of these systems for operational use. Improvements might be in the form of (1) upgrading the sensor, (2) changing some algorithm in the exploitation component, (3) inserting a new sensor that has a different modality, or (4) inserting a new exploitation component to work alongside the existing component. Making changes to existing systems not knowing if the end result will be beneficial can be a big waste of time and money. This time and money is due to the current manner in which these improvements are approached. When making changes or adding to the legacy system the procedure typically is:

Step 1. Design a combination method or rule. There may be many possible methods but they usually make a "best" guess.

Step 2. Build the new combined system according to this method.

Step 3. Test the new system. One may have to perform multiple tests in order to get enough data.

Step 4. Perform validation(s) of the combined system.

Step 5. Determine the performance of the new combined system

What is needed is an analysis that could determine (or estimate) the performance of the combined system before Step 2 is begun, and maybe before the method is chosen in Step 1. Department of Defense (DOD) contractors typically submit proposals to the Air Force Research Laboratory (AFRL) for changes to legacy sensor-exploitation systems. If fusion engineers had this analysis in the form of a software program, they could play "what if" scenarios to determine (or estimate) the performance of the new combined system. If the performance was not much better than the existing system, then it may not be worth the cost. If the performance is greatly improved then the steps listed above may well be worth the time and cost. Often (potential) DOD contractors may not share all the details of their methods but would share some performance data. How much data is needed to determine/estimate the performance of the combined systems?

The goal of this research is to generate a mathematical theory of combining a legacy sensor-exploitation (SE) system with (a) new component(s), e.g. sensor, exploiter, or an entire new SE system, and that would predict the performance before physically building it (Step 2). The mathematics needed for this research will Receiver Operating Characteristics (ROC) analysis, e.g. ROC curves for sensor-detection systems and ROC manifolds for sensor-classification systems.



Schubert CM, Thorsen SN, E. Oxley ME: The ROC Manifold for Classification Systems, Pattern Recognition 44(2): 350-362, February 2011.

Thorsen SN, Oxley ME: A Description of Competing Fusion Systems, International Journal of Information Fusion 7: 347-360, December 2006.


Receiver operating characteristics (ROC); Information fusion; ROC curves; ROC manifolds; Category theory; Performance functionals; Detection system; Classification system;


Citizenship:  Open to U.S. citizens
Level:  Open to Postdoctoral and Senior applicants
Copyright © 2014. National Academy of Sciences. All rights reserved. 500 Fifth St. N.W., Washington, D.C. 20001.
Terms of Use and Privacy Statement.