Dr. W. Matthew Carlyle is a Professor and Chair of the Operations Research Department at the Naval Postgraduate School. He received his bachelors in Information and Computer Science from Georgia Tech in 1992 and his doctorate in Operations Research from Stanford University in 1997. He was in the Department of Industrial Engineering at Arizona State University from 1997-2002, and joined the NPS faculty in 2002. His research and teaching interests include network optimization, integer programming, network interdiction and spreadsheet models in OR. He receives support from the Office of Naval Research for military applications of optimization, including attack and defense of critical infrastructure, delaying large industrial projects and weapons programs, theater ballistic missile defense, sensor mix and deployment, naval logistics, and Navy mission planning. Carlyle was awarded the Navy Superior Civilian Service Award in 2006 for his work in support of OEF and OIF.
There are two “classic” areas of military problems that I see as always remaining relevant. The first of these, military logistics, occupies a large part of my own work in collaboration with students and other faculty right now. We see the potential for needing to stand up major operations in key places on the globe, and, as always, the question is whether we can get the right platforms, personnel, materiel, and fuel to the right place at the right time. The second is in manpower planning; we are keenly aware of the problems all of the services have to address in this general area. New promotion, retention, and retirement policies will have impacts on the future force structure, and we have a history of building models to analyze the impacts of these types of changes.
Newer areas of work include cybersecurity and cyber operations, and the rise of autonomous system. As just one example of the many OR contributions to be made in cyber, we need to be able to develop and use statistical models that can help determine if a cyber system has been breached. Autonomous systems applications present a host of OR-related problems and opportunities in the areas of information sharing and real-time analysis of that information, and determining the degree of autonomy needed for different mission types. Artificial intelligence, and, in particular, machine learning algorithms, can have significant impact on the scope and scale of problems we can solve. In addition, many of these topics need rigorous, statistically valid plans for experimentation and for analyzing the results of those experiments, including the use of analytical wargames; these are all fruitful areas of research for our faculty.
One of our overarching strategies is to make sure we hire people with deep analytical skillsets, and then mentor them in the art of military modeling and analysis. Topics in the Pentagon change frequently based on changes in our adversaries, changes in technology, and changes in our own goals. However, the research that can advance those topics always comes down to two basic things: our ability to help decision makers frame questions so that the answers are useful, and our core set of analytical tools and their mathematical and computational foundations. Our focus over the last decade on emphasizing computation throughout our curriculum has allowed us to develop the skills to create both relevant analytical models and to deliver decision support tools to decision makers throughout the Navy and the DOD. The fact that we are home to the statisticians, optimizers, and stochastic and simulation modelers on campus means that we have the expertise to contribute to all of these emerging analytical problems.
To specifically address those four topics, for example, statistics and optimization are at the heart of almost every artificial intelligence and machine learning algorithm. Integrated networks need to be functional, but also robust to deliberate damage, a type of modeling that is central to much of our work in infrastructure defense. Hypersonics change offensive and defensive capabilities, and we can incorporate those changes into our combat modeling, run computer experiments to determine the effect of this new technology on current tactics, and even suggest new tactics that account for those changes. Our SEED Center for Data Farming specializes in this sort of combat modeling. Finally, microelectronics as a manufacturing problem has plenty of classic OR modeling and analysis supporting it. The innovative ways in which microelectronics can transform military operational decision making at all levels is a new set of challenges that OR is well prepared to address.
The biggest issue of this decade will be the management and analysis of enormous amounts of information. AI and ML have been identified as important tools for that task because of the speed at which they can process large amounts of data, but we have seen in recent years that fully automated systems can make mistakes, sometimes with severe consequences. The operations research areas of statistics and optimization are at the core of most AI and machine learning algorithms, and one of the things we bring to the table is the ability to ask questions about the recommendations these systems make. “Why does the model end up favoring these types of decisions? What can we do to remove its bias against this type of assignment? How can we avoid mis-classifying this type of target?” Fully automated systems don’t pause to ask these questions, but modelers and analysts can, and should, continue to develop tools not only to improve the recommendations these automated systems make, but to also understand what underlying features are guiding those decisions and how to make recommendations that truly reflect what the decision makers want. Better tools can help keep humans an important part of the decision process. This makes the algorithms and, more importantly, the decisions they suggest more reliable, more transparent, and therefore more efficient to employ.
The most basic (and accurate) description I’ve heard of operations research is “allocating scarce resources.” If there is one thing that every one of our military officer students understands, it is being given a task or mission with insufficient resources to fully accomplish it. They also have a deep appreciation for how difficult uncertainty can be; planning for a known shortfall is a well understood problem, but if the actual resources you will have available is still unknown you have to dramatically increase the scope of your analysis and figure out what you might do for a range of values. We have a long history of incorporating uncertainty in the models we build and analyze for many of our research sponsors. Sometimes we plan for worst-case outcomes, sometimes we try to plan for the “average” result, and sometimes we experiment by building specific, relevant scenarios to see what might happen. The choices we make for how to model uncertainty depends on the type and priority of the problem being addressed, the specific question we need to answer, and how impactful the result of our decisions will be if the dice fall on a bad outcome. We teach all of this throughout our courses and incorporate it into our thesis advising and our sponsored research.
