Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

MD01

3 - Automated IMRT Treatment Planning using Expedited Constrained Hierarchical Optimization (echo): Algorithm Development and Clinical Implementation Masoud Zarepisheh, Memorial Sloan Kettering, 485 Lexington Ave, #2039, New York, NY, 10017, United States, Linda Hong, Ying Zhou, Sovanlal Mukherjee, James G. Mechalakos, Margie A. Hunt, Joseph O. Deasy This study automates IMRT treatment planning by formulating that as a hierarchical constrained optimization problem (also known as prioritized optimization). We introduce an idea to speed up the optimization process by sparsifying the influence matrix and then solving an extra unconstrained optimization problem to compensate for the inaccuracy caused by sparsity. The algorithm is equipped with an effective heuristic to handle DVH constraints. The algorithm has been implemented in the clinical environment and has been used to treat more than 300 patients in about 9 months. 4 - Optimization of Fractionated Radiotherapy Treatments under Dose Delivery Uncertainty David Papp, North Carolina State University, Raleigh, NC, United States, Melissa R. Gaddy, Sercan Yildiz, Jan Unkelbach Radiotherapy treatments are fractionated, meaning that the treatment is administered in several sessions, each delivering a small dose to the patient. In current clinical practice, the fractions are identical, but recent work has demonstrated that there is substantial benefit to be had from delivering different dose distributions in each fraction—-a concept known as spatiotemporal fractionation. One concern about such treatments, however, is that they might be more susceptible to the various sources of uncertainty in dose delivery, such as patient positioning errors. In this talk, we investigate how the potential benefit from spatiotemporal fractionation is affected by uncertainty. n Keynote- Monday West Bldg 301AB Keynote: Scaling Transportation Capacity in the Age of E-Commerce Keynote Session Chair: Georgia Perakis, Massachusetts Institute of Technology, Sloan School of Management, 100 Main Street Rm E62-565, Cambridge, MA, 02142-1347, United States 1 - Scaling Transportation Capacity in the Age of E-Commerce Presenter Russell Allgor, Amazon, Seattle, WA, United States Even as Amazon experiences year-over-year growth in package volume, it’s been able to speedup delivery times and shorten the time it takes between when an item is ordered and the moment that item arrives on a customer’s doorstep. Amazon’s fulfillment business - along with that of other online retailers - has increased the pressure on parcel delivery transportation capacity both in the US and overseas. Continuing to satisfy customers’ needs will require more efficient use of existing networks and the creation of additional capacity for package sortation, line haul trucking, air freight, and last mile delivery. The challenge of designing a network that can meet these dynamic needs requires us to develop technical solutions implemented through design tools that incorporate the latest process innovations. There are both technical and organizational challenges in creating network design decisions, which involve inventory placement, facility operations, facility locations, network connections, and scheduling. While the technical challenges are at least partially met through solving large-scale combinatorial optimization problems, the organizational challenges require innovative mechanisms that motivate network-wide cooperation while also allowing decentralized, scalable operation. Monday, 3:10PM - 4:00PM

n Keynote – Monday West Bldg 301D

Keynote: Stochastic Optimization, Statistical Modeling and Distributed Processing Applied to Energy Planning (IFORS Distinguished Lecture) Keynote Session 1 - Stochastic Optimization, Statistical Modeling and Distributed Processing Applied to Energy Planning Mario Veiga Pereira, PSR, Praia de Botafogo 228, Rio de Janeiro, Brazil The explosive growth of renewables has compounded the modeling challenges for investment and operational decisions in the energy area: (i) joint representation of hydro, wind, solar as a multivariate, multiscale stochastic process; (ii) solving the co-optimization of storage (hydro, batteries etc.), renewable-driven generation reserves and power transmission as a multistage nonconvex stochastic optimization; and (iii) because renewable integration benefits from regional interconnections (portfolio effect), probabilistic simulations are carried out for very large systems. In this talk, we show that these challenges can be successfully addressed by the combination of: new nonconvex stochastic optimization algorithms such as SDDiP; advanced geoprocessing and Bayesian networks; distributed computational and data management tools; and new optimization- centered programming languages such as Julia/JUMP. The application of these tools will be illustrated with real planning studies for the US Pacific Northwest and energy integration of South America. Emerging Topic: Keynote Emerging Topic Session Chair: Robert Dell, Naval Postgraduate School, Monterey, CA, United States 1 - Innovative Collaboration Between Industry and Academics to Meet Future Analytics Talent Requirements Melissa R. Bowers, University of Tennessee-Knoxville,Statistics, Operations, & Management Science, Knoxville, TN, 37996, United States, Mike Galbreth, Bryan Noreen, Madeleine Beatty In 2018, the Master of Science in Business Analytics in the Haslam College of Business at the University of Tennessee received the UPDS George D. Smith Prize. This award acknowledges our collaboration with industry partners and our success in preparing students to be effective practitioners in industry. In the spirit of the prize, this representation shares our experience in creating and maintaining the program. Key aspects of the program are its focus on real-world problems at nearly every touchpoint and a continuing interaction with industry. Students develop technical skills, business subject matter expertise, communication, teamwork, and leadership skills needed to impact decision making in practice. n MD01 North Bldg 121A Data Driven Optimization and Learning Sponsored: Optimization/Optimization Under Uncertainty Sponsored Session Chair: Andrew Lim, National University of Singapore, Singapore, 119245, Singapore 1 - When Can We Improve on Sample Average Approximation? Edward James Anderson, University of Sydney, Discipline of Business Analytics, Room 487, Merewether Bld H04, Sydney, NSW 2006, Australia We want to minimize the expected value of an objective function f(x,y) with a decision variable, x and a random variable Y, where we have N samples drawn from Y. The standard approach of sample average approximation simply minimizes the average value over the observed samples. We consider how to use information on the underlying distribution (e.g. that it has a continuous density). We also consider robust approaches that minimize the worst outcome achieved after a small change to the sample points or the weights applied to them. We explore this through looking at a univariate decision variable, and simple functions f. We demonstrate problems for which robustification makes the SAA solution worse. Monday, 4:30PM - 6:00PM n Keynote – Monday West Bldg 301C Keynote: UPS George D. Smith Prize

213