NSF awards prestigious CAREER grant to ORIE Assistant Professor Frazier
The National Science Foundation has awarded ORIE Assistant Professor Peter Frazier a Faculty Early Career Development (CAREER) grant to develop improved methods for optimization via simulation, and to apply these methods to decision-making problems in cardiovascular medicine.
Optimization and simulation are distinct fields in operations research and are typically covered in separate courses and research programs. Frazier's research combines elements from both areas so as to find the best among many options, each of which is expensive to evaluate: determining the quality of each option entails a computer simulation that has random elements and so must be run many times to provide a reliable characterization.
CAREER awards are granted through an NSF-wide activity "in support of junior faculty who exemplify the role of teacher-scholars through outstanding research, excellent education and the integration of education and research," according to the NSF. Frazier has been awarded a five-year grant for a total of $400,000.
The NSF awards about 400 CAREER grants each year. This year four of these grants went to Cornell faculty in Ithaca, including Nate Foster and Ashutosh Saxena, both in Computer Science, and Ricardo Daziano in Civil and Environmental Engineering as well as Frazier in ORIE.
“I am honored to receive the award, and very appreciative to the NSF,” said Frazier. “I’m also excited to do the work proposed, and am glad to have support from the NSF to allow me to do it,” he said.
Frazier joined ORIE in 2009 after receiving a Ph.D. from the Department of Operations Research and Financial Engineering at Princeton. He has received the Sonny Yau '72 Excellence in Teaching award and an Air Force Office of Scientific Research Young Investigator award, and is a participant in several collaborative grant activities.
An enabling technology
If successful, Frazier’s research will yield an enabling general operations research technology, since many applied problems within optimization via simulation share the characteristics for which his proposed approach is valuable. Examples include selecting the best design for a supply chain, selecting the best layout for an assembly line, and selecting the best way to staff a hospital.
Finding optimal configurations for such systems is difficult, because the systems entail random variation (in product demand, processing duration, and workload), and because a particular configuration may be better than similar alternatives but not as good as some that are radically different and therefore less likely to be found through an iterative series of small changes. Improving the speed with which solutions can be found for these problems can make “the difference between the method being practical, and not,” according to Frazier.
Frazier's research will also have specific benefits within cardiovascular medicine. Cardiologists will be able to design improved strategies for patients who have had aortic aneurysms or atherosclerotic blockages repaired, and will be better able to design and place grafts in patients undergoing bypass surgery.
With Dr. Andrew Meltzer and colleagues at the Division of Vascular and Endovascular Surgery at Weill Cornell Medical College, Frazier has considered the application of engineering principles that typically focus on the study of reliability and failure of systems or manufactured products, to the identification of the risk that blockages will recur in patients who have surgery to restore blood flow through a limb.
They have shown that such failures after peripheral surgery follow the typical “bathtub curve” (at right) for product failure, with a shape depending on patient characteristics. Patients are currently given follow-up CT scans or duplex ultrasounds according to a fixed schedule regardless of conditions that would either predispose the patient to having a high risk at certain points in time (in which case scans should be more frequent) or to having a low risk (in which case excessive radiation and cost could be avoided).
Frazier and Meltzer have proposed a goal of improving surveillance schedules to detect life-threatening problems earlier and more reliably, with fewer CT scans or duplex ultrasounds overall. The goal would be accomplished through a computational procedure that determines an optimal surveillance schedule for each patient - one that minimizes the probability of a negative outcome subject to a constraint on the total number of scans. Doing so requires determining the odds of a negative outcome due to a missed detection for each of a large number of possible schedules.
While Frazier and Meltzer have carried out an initial analysis of these odds that takes the type of surgery and the condition of the patient, computing the odds for a particular candidate schedule requires thousands of runs of a computer simulation. Improving the efficiency of such an optimization via simulation could make it feasible to routinely generate optimal surveillance schedules.
Among the coauthors of a paper on Frazier’s work with Meltzer are Shanshan Zhang, a recent ORIE Ph.D. graduate, and Pranav Hanagal, ORIE ’13. Hanagal is currently a Ph.D. student at the University of Minnesota. Paul Liu ORIE ’13, who received his BS with honors in May, remains involved in this work as an ORIE Master of Engineering student.
With Dr. Alison Marsden of the University of California at San Diego, Frazier is applying optimization via simulation to determine the optimal shape and placement of the grafts used in coronary artery bypass surgery, as seen in the idealized model at left. The shape and placement must account in detail for the extent of arterial narrowing and the rate at which blood flows into the artery, both of which can be measured but with some uncertainty. The design and placement also must account for uncertainty in exactly where the graft will end up being placed during the surgery.
Even without taking uncertainty into account, using a fluid flow simulation to predict the results of a given design and placement takes hours of time on a large cluster of computers. Adding uncertainty and the search for an optimal design and placement requires systematic consideration of a large number of possible solutions, so here too, improved efficiency of optimization via simulation can help improve surgical outcomes.
Linking competing approaches
Frazier proposes to make the ranking of options and selecting among them more efficient by linking statistical approaches sometimes considered to be in opposition: Bayesian methods and frequentist methods. Frequentist methods assume observed data has been sampled from a fixed population of possible outcomes, while Bayesian methods analyze observed data in light of prior opinions about the possible outcomes. While Bayesian methods may produce better results on average, frequentist methods can yield guarantees on worst case results. Frazier posits that combining these approaches will produce solutions more quickly and with tighter guarantees on the quality of solutions than existing methods.
He has also proposed to use Bayesian analysis to efficiently allocate computing resources used in simulation experiments based on multiple starting points, allowing those starts that are more likely to have high-quality (local) optima to converge first.
Frazier's grant proposal is based on his overarching research philosophy, which is to employ the most efficient way to collect information in the course of solving a problem. He calls work based in this philosophy "optimal learning" because it entails asking how to learn in an optimal fashion in the course of pursuing answers. His philosophy is discussed in this video.
Students assist the research
As is typically the case with faculty research grants, graduate student research assistants will benefit from the financial support that Frazier’s grant provides. Some current Ph.D. students are already participating in the work. In particular, second year student Weici Hu is working on methods for handling situations in which radically different alternatives may improve on solutions that appear optimal relative to small perturbations. Jing Xie is working on the arterial bypass surgery application. ORIE undergraduate twin brothers Divya and Somya Singhvi are working with M.Eng. student Paul Liu on vascular surveillance project. Other graduate and undergraduate students may participate in Frazier’s CAREER research in the future.
As the NSF states, CAREER grants are for teacher-scholars. The educational impact of Frazier’s grant will extend from course development and undergraduate mentoring to high-school outreach through Cornell’s summer programs. Frazier will develop and teach specific week-long summer modules for the CURIE Academy, which brings high school girls to Cornell for college preparatory work, and Upward Bound, which provides college preparation and enrichment activities to high school students in rural areas around Ithaca, NY and the city of Elmira, NY.
Frazier received a B.S. in Physics and Engineering/Applied Science from the California Institute of Technology in 2000, after which he spent several years in industry as a software engineer, working for two different start-up companies and for the Teradata division of NCR, before undertaking his Ph.D. at Princeton.
Ever since he was a kid, says Frazier, he's wanted to be a professor. He chose Cornell for its collaborative culture. "It seems to be easier to do interdisciplinary work here than other places and that's important to me,” he says,” because my work is statistics, optimization, biology, simulation, all kinds of different stuff."