INFORMS 2021 Program Book

INFORMS Anaheim 2021

TD08

2 - Developing Principles for Dei-informed Research in Or/analytics Through an Analysis of Published Journal Articles Michael P. Johnson, University of Massachusetts Boston, Department of Public Policy & Public Aff Mccormack, Boston, MA, 02125-3393, United States, Tayo Fabusuyi We describe a project to develop principles by which researchers in OR/analytics may integrate ideas about diversity, equity and inclusion, racial and social justice and antiracism into research ideas that span application areas, disciplinary modes and analytic methods. These principles are derived from a mixed methods analysis of INFORMS journal publications including thematic analysis and author interviews. This project has the potential to improve the profession (how the work gets done, and the environment within which the work is done) as well as the discipline (the academic and scholarly place within which the work is situated). TD08 CC Room 303C In Person: Simulation, Learning and Queueing Theory General Session Chair: Qiaomin Xie, U-Wisconsin-Madison, Cornell University, Ithaca, NY, 14850, United States 1 - Online Learning and Pricing for Service Systems with Reusable Resources Huiwen Jia, University of Michigan, Ann Arbor, MI, 48105-1181, United States, Cong Shi, Siqian Shen We consider a price-based revenue management problem with finite reusable resources. The arrival and service rates depend on the posted price but the mappings are unknown. The firm makes adaptive pricing decisions to maximize the cumulative revenue. Compared with prior pricing and MAB literature, the salient difficulties are (i) unknown rate-and-price mappings, (ii) the dynamic nature of reusable resources being committed over time, (iii) the transient behavior of the system when price changes, and (iv) unbounded and heavy-tailed observed random variables. Our algorithms contain a Warm-up Phase to eliminate the heavy-tail effects and a Learning Phase to identify the optimal price. We prove that the cumulative regret is O( √ PT log T), where T and P are the number of time periods and candidate prices, and this result matches the lower bound up to a logarithmic factor. 2 - Efficient Algorithms for Online Decision-making with Limited Action Changes Yilun Chen, Cornell, Ithaca, NY, 14850-1854, United States Many of today’s online decision-making tasks (e.g. dynamic pricing, pandemic management) boil down to solving stochastic dynamic programs (DP) with high- dimensional / path-dependent underlying state, suffering from the ``curse of dimensionality’’. We propose a new approach that overcomes this computational barrier for a fairly general class of problems, subject only to a ``limited-action- change’’ constraint (translate to ``limited price change” for dynamic pricing). Our results come with strong theoretical guarantees (both runtime and accuracy) for models with arbitrary state transition and reward structures, guaranteeing an $(1- \epsilon)-$optimal policy in a runtime scaling polynomially in the time horizon and effectively independent of the dimension, The key building block of the approach is our recent algorithmic progress for optimal stopping. 3 - Decentralized Q-learning in Zero-sum Stochastic Games Kaiqing Zhang, MIT, Cambridge, MA, 61801-2307, United States We study reinforcement learning (RL) in infinite-horizon zero-sum stochastic games. We focus on the practical while challenging setting of decentralized multi- agent RL, where the agents are not coordinated by any central controller, and neither the actions nor the payoffs of the opponent agent can be observed. Each agent might be even oblivious to the presence of the opponent. Unlike many existing MARL algorithms, we aim to develop algorithms that are both rational and convergent the learning dynamics are natural to each agent, converging to the opponent’s best-response when the opponent converges to a stationary policy; while the iterates converge to the Nash equilibrium when both agents adopt the learning dynamics. We develop a decentralized Q-learning, with provable convergence guarantees to the Nash equilibrium of the game.

proportional imbalance objective leads to non-integral supplies and demands in the MCNF formulation; we show an alternate MCNF formulation which has an optimal integer solution. 3 - Algorithms and Complexities of Matching Variants in Covariate Balancing Dorit Hochbaum, University of California, Berkeley, Berkeley, CA, 94720, United States, Asaf Levin, Xu Rao We present a comprehensive complexity study of variants of matching problems that arise in covariate balancing. In these problems we seek the largest possible selection of treatment samples to match, or to minimize the assignment costs, while meeting balance requirements. For these problems we provide polynomial time algorithms, or a proof of NP- hardness, where the polynomial time algorithms are combinatorial and use network flow techniques. We further present several fixed-parameter tractable results for problems where the number of covariates and the number of levels of each covariate are seen as a parameter. Hybrid Life as a PhD Student Sponsored: Minority Issues Forum Sponsored Session Chair: Diana Gineth Ramirez-Rios, Rensselaer Polytechnic Institute, Troy, NY, 12180-2050, United States 1 - Life as a PhD Student Diana Gineth Ramirez-Rios, Rensselaer Polytechnic Institute, Troy, NY, 12180-2050, United States Making a decision to start a Ph.D. is not an easy task. Yet, many of us decide to trust our abilities and knowledge to take that step and enroll in the program of interest. Despite of the variety of experiences told to us, for the most part life as a Ph.D. is hard, requires sacrifice, and perseverance. But why do we make this decision? How do we overcome the challenges we face during these years? Is it worth the sacrifice? This panel is intended to cover all these questions about the life of a Ph.D. student. The panel consists of mostly Ph.D. students at the different stages of their research, and one faculty who will share us the lessons learned. 2 - Panelist Sofia Perez-Guzman, Rensselaer Polytechnic Institute, Troy, NY, 12180-3522, United States 3 - Panelist Yifei Sun, Dartmouth College, Lebanon, NH, 03766, United States 4 - Panelist Martha Sabogal, Clemson University, Clemson, SC, United States TD06 CC Room 303A In Person: Diversity/PSOR/MIF Diversity, Equity and Inclusion in OR/MS/Analytics. Innovations in Research and Practice II General Session Chair: Michael P Johnson, University of Massachusetts Boston, Department of Public Policy & Public Aff Mccormack, Boston, MA, 02125-3393, United States 1 - Bringing STEM to underserved Communities Phebe Vayanos, University of Southern California, OHE 310L, University Park Campus, Los Angeles, CA, 90089, United States, Aida Rahmattalabi, Caroline Johnston Industry and academia suffer from lack of full participation in STEM fields, excluding those from traditionally marginalized groups. To help address this problem, we partnered with Code in the Schools, a non-profit in Baltimore city and STEM academy of Hollywood, a high school in Los Angeles to bring students from traditionally under-represented groups in STEM together and engage them in fun AI/OR projects for social good. To achieve this, we held the ExplOR event in November 2020. Guided by mentors from the INFORMS community, the students worked in teams to address a range of problems in areas including public health, conservation, etc. The goal was to raise students’ interest in AI/OR and help them build a network with mentors and fellow students. TD05 CC Ballroom E / Virtual Theater 5

127