INFORMS 2021 Program Book

INFORMS Anaheim 2021

WB31

WB28 CC Room 207B In Person: Stochastic Integer Programming and Its Applications General Session Chair: Yiling Zhang, University of Minnesota, Minneapolis, MN, 55455- 0141, United States 1 - Distributionally Robust Bilevel Programming and Applications of Facility Location Yiling Zhang, University of Minnesota, Minneapolis, MN, 55455- 0141, United States, Chuan He, Akshit Goyal We study distributionally robust two-stage stochastic bilevel programs (DRBPs) in the context of sequential two-player games under uncertainty, where the leader makes a binary here-and-now decision. After observing the leader’s decision and realization of uncertainty, the follower responds with a continuous wait-and-see decision. We show that the DRBP is equivalent to a DR two-stage stochastic integer program with uncertainty in both objective and constraints. Under both moment-based and Wasserstein ambiguity sets, we derive 0-1 semidefinite and copositive programs. Computational study of a facility location problem is conducted to demonstrate the efficiency and effectiveness. 2 - Tight Conic Approximations for Two-Sided Chance-Constrained Optimization In this talk, we focus on developing tight conic approximations for two-sided chance constrained (TCC) programs with an application to AC optimal power flow problem. We present an efficient second-order cone programming (SOCP) approximation of the TCC programs under Gaussian Mixture (GM) distribution. As compared to the conventional normality assumption for forecast errors, the GM distribution adds an extra level of accuracy representing the uncertainties. Moreover, we show that our SOCP formulation has adjustable rates of accuracy and its optimal value enjoys asymptotic convergence properties. Finally, we demonstrate the effectiveness of our proposed approaches with both real historical data and synthetic data on the IEEE 118-bus system. WB29 CC Room 207C In Person: Application of Stochastic Programming to COVID-19 Related Problems General Session Chair: Lewis Ntaimo, Texas A&M University, College Station, TX, 77843, United States 1 - A Distributionally Robust Optimization Approach for Location and Inventory Prepositioning of Disaster Relief Supplies Karmel S. Shehadeh, Lehigh University, Bethlehem, PA, 18015- 1518, United States, Emily L. Tucker We study the problem of disaster relief inventory prepositioning under uncertainty of disaster level and location, demand of relief items, usable fraction of prepositioned items post-disaster, procurement quantity, and arc capacity. We propose distributionally robust optimization (DRO) and stochastic programming (SP) approaches, assuming unknown and known uncertainty distributions, respectively. To illustrate potential applications of our approach, we conduct extensive experiments using a hurricane season and an earthquake as case studies. Our results demonstrate the (1) superior operational performance of the DRO decisions compared to SP decisions, (2) trade-off between DRO pessimism and SP optimism, and (3) computational efficiency of our approaches. 2 - Optimal COVID-19 Vaccine Allocation under Uncertain Transmission Characteristics Using Stochastic Programming Lewis Ntaimo, Texas A&M University, College Station, TX, 77843, United States, Jiangyue Gong, Brittany Segundo, Krishna Reddy Gujjula We present a chance constrained stochastic programming model to determine optimal COVID-19 vaccination policies for a multi-community population under variability in parameters related to vaccines, age-specific effect of SARS-CoV-2, and social behaviors. The optimal solution provides the minimum number of vaccines to prevent epidemics for a specified reliability level for a given community and the results show that around 80% of the population needs to be vaccinated and vaccines should be targeted towards larger household sizes. Abolhassan Fathabad, University of Arizona, Tucson, AZ, United States, Jianqiang Cheng, Kai Pan, Boshi Yang

WB30 CC Room 207D In Person: The Internet of Federated Things: An Overview of Federated Learning and a Vision for the Future Panel Session Chair: Raed Al Kontar, University of Michigan, Ann Arbor, MI, 48109- 2117, United States 1 - Panelist Seokhyun Chung, University of Michigan, Ann Arbor, MI, 48109- 2117, United States 2 - Panelist Naichen Shi, University of Michigan, MI, United States WB31 CC Room 208A In Person: Emerging Topics in Social Media Analytics General Session Chair: Tauhid Zaman, Yale University, Boston, MA, 02114, United States 1 - Time-constrained Data Collection for Seeding Time-critical Interventions M. Amin Rahimian, University of Pittsburgh, Pittsburgh, PA, United States, Sanzeed Anwar, Dean Eckles Seeding strategies rely on knowledge of social network structure to choose local intervention points that maximally spread information or a desired social behavior. In practice, such structural knowledge of social networks is costly and time-consuming to obtain. In this paper, we provide a framework for performing time-constrained structural queries that inform the design of time-critical seeding interventions (where one cares about not only the eventual extent of the spread, but also the speed at which new adopters join the campaign). Our theoretical results address the following question: how much time and sampling resources the researchers need to spend to acquire enough information for designing seeding interventions with appropriate quality guarantees. 2 - To Pay or Not to Pay: Targeting Referral Rewards in the Presence of Voluntary Word-of-Mouth Diffusion Shatian Wang, Columbia University, New York, NY, 10027-6715, United States Marketing campaigns should capitalize on voluntary word-of-mouth (WoM) to avoid paying for diffusion (via referral rewards) that would have otherwise occurred for free. In the presence of voluntary WoM, we study referral rewards targeting and identify conditions where it is optimal to target a strict subset rather than the entire customer population within two settings: 1) a random network model in which the campaign can only access population-level characteristics and 2) explicit network models in which the campaign has full knowledge of individual consumers’ social network. 3 - Text Analysis in Social Media Using Transformers Tauhid Zaman, Yale University, New Haven, CT, United States In this session we will provide a hands-on course on how to use transformer neural networks to measure the sentiment of social media posts. We will begin with an overview of transformer architectures. Then we will show how to use pre-trained transformers to measure a variety of sentiments on social media posts. These include standard positive and negative sentiment, but also more complex sentiments such as toxicity, joy, and optimism. Finally, we will apply these transformers to real Twitter data to find novel connections between sentiment and engagement on social media. This session will be interactive and attendees will be given access to all Python code used.

155