Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

MD02

2 - “Model-based Regularization” in a Context of Choice Modeling Shanshan Huang, National University of Singapore, National University of Singapore, BIZ2 B1-01, Singapore, 117592, Singapore, Andrew Lim, Tong Wang This work defines and explores the notion of “model-based regularization in the context of choice modeling.Regularization is an important tool from Machine Learning that is used to stabilize the solutions of regression problems when there is little data (relative to covariates). We take the more general view that regularization is simply a perturbation of a data-driven estimate towards a sensible default solution so as to remove “variance at the cost of some bias. 3 - Effective Scenarios in Multistage Distributionally Robust Stochastic Programs with Total Variation Distance Guzin Bayraksan, The Ohio State University, 210 Baker Systems, 1971 Neil Avenue, Columbus, OH, 43210-1271, United States, Hamed Rahimian, Tito Homem-de-Mello We consider multistage distributionally robust stochastic programs (DRSP) with a finite number of scenarios, where we use total variation distance to form conditional ambiguity set of distributions. We investigate the question of which scenarios have “effect on the optimal value of this multistage DRSP. To identify the effective scenarios, we conduct perturbation analysis with respect to a collection of scenarios being excluded and propose easy-to-check conditions for them. We explore the effectiveness for a scenario path as well as for a scenario conditional on the history of the stochastic process. Computational experiments Andrew Lim, National University of Singapore, 15 Kent Ridge Drive, Mochtar Riady Building, Singapore, 119245, Singapore, Michael Kim While the optimal policy for the Gittins-Whittle formulation of the multi-armed bandit is fully characterized in terms of the Gittins index, these are notoriously difficult to compute for high dimensional Bayesian bandits. We develop a method for approximating the dynamics of the posterior using the Bayesian Central Limit Theorem and show how it can be used to approximate the Gittins index for high- dimensional Bayesian bandits. Comparisons of the Gittins-Whittle framework to Thompson sampling and the Upper Confidence Bound approach will be discussed, and applications of our approximation to Bayesian bandits where the rewards are mixtures will also be presented. Robust Optimization with Varying Uncertainty Sponsored: Optimization/Optimization Under Uncertainty Sponsored Session Chair: Kartikey Sharma, Northwestern University, Evanston, IL, 60208, United States 1 - Robust Non-stationary Queueing Theory Chaithanya Bandi, Kellogg School of Management, Northwestern University, 2001 Sheridan Rd, 566, Evanston, IL, 60208, United States We consider the problem of obtaining delay bounds for queueing systems with time varying arrival and service rates. We formulate optimization problems that allow tractable calculation of bounds on the waiting time. 2 - Optimization under Decision-dependent Uncertainty Omid Nohadani, Northwestern University, 2145 Sheridan Road, Technological Institute M233, Evanston, IL, 60208-3119, United States, Kartikey Sharma The efficacy of robust optimization spans a variety of settings with predetermined uncertainty sets. In many applications, uncertainties are affected by decisions and cannot be modeled with current frameworks. This talk takes a step towards generalizing robust optimization to problems with decision-dependent uncertainties, which we show are NP-complete. We introduce a class of uncertainty sets whose size depends on decisions, and proposed reformulations that improve upon alternative techniques. In addition, the proactive uncertainty control mitigates over conservatism of current approaches. illustrate the results on finance and environmental problems. 4 - Approximating the Gittins Index for High-dimensional Bayesian Bandits n MD02 North Bldg 121B

3 - Optimization under Connected Uncertainty Kartikey Sharma, Northwestern University, Room C223. 2145 Sheridan Road, Evanston, IL, 60208, United States, Omid Nohadani Robust optimization methods provide a tractable method to address uncertainties in optimization. Typically, the uncertainty models are prespecified. In problems spread across multiple time periods, however, the uncertainty may be influenced by the past. In this presentation, we leverage this influence by, what we call, connected uncertainty sets, where the set parameters at each period depend on previous realizations. To find optimal here-and-now solutions, we reformulate robust constraints for these connected uncertainty sets. We evaluate the performance of this framework on a knapsack problem with different levels and directions of past dependence. 4 - Mixed-integer Recourse via Prioritization Yuanyuan Guo, University of Michigan, 1205 Beal Avenue, Ann Arbor, MI, 48109, United States, Ruiwei Jiang In this talk, we study the transmission expansion planning problem with prioritization (TEP-P), where we rank the candidate circuits by considering the uncertainty of development budget, electricity load, and renewable energy supply. We propose a two-stage robust TEP-P model and solve this problem by using the Benders’ decomposition approach. We validate the performance of the proposed approach via case studies based on real-world wind power data.

n MD03 North Bldg 121C TIMES Distinguished Speaker Sponsored: Technology, Innovation Management & Entrepreneurship Sponsored Session

Chair: Gulru F. Ozkan-Seely, University of Washington Bothell, University of Washington Bothell, bothell, WA, 98011, United States 1 - Alpha Strategies: Sustained Advantage in Entrepreneurial and Established Firms Karl Ulrich, University of Pennsylvania, Philadelphia, PA, United States All innovation begins with some kind of disequilibrium, and may lead to supra- normal profits for the pioneering enterprise. However, all disequilibrium eventually fades. This talk links what we know about strategy, finance, and innovation into a framework for understanding sustained competitive advantage for innovating organizations, whether new entrants or incumbents. I also examine the evidence for the magnitude and duration of sustained profitability above the cost of capital in established public companies, and consider the extent to which such profits are the result of innovation. n MD04 North Bldg 122A Computational Integer Programming Sponsored: Optimization/Integer and Discrete Optimization Sponsored Session Chair: Ezgi Karabulut, Georgia Tech, 2257 Burdett Avenue, Troy, NY, 12180-2406, United States 2 - Using the Membership Linear Program and General Split Disjunctions to Approximate the Split Closure Split cuts are widely used cutting planes to solve general mixed-integer linear programs (MILPs). The split closure has been computationally shown to provide strong bounds for the optimal solutions of MILPs. To approximate the split closure, we generate rank-1 split cuts iteratively by applying general split disjunctions to the membership LP, which was used by Bonami (2012). Our algorithm can be carried out with any LP solver, without needing simplex tableau information. We study the properties of the membership LP (applied with general split disjunctions) and propose some practical strategies for improving the performance of the algorithm. Finally, we present detailed computational results. Junghwan Kwak, Korea Advanced Institute of Science & Technology, Daejeon, Korea, Republic of, Sungsoo Park

214