INFORMS 2021 Program Book

INFORMS Anaheim 2021

TD09

TD09 CC Room 303D In Person: Learning for Queueing and Beyond General Session Chair: Weina Wang, Carnegie Mellon University, Pittsburgh, PA, 15213, United States 1 - Learning While Playing in Mean-field Games: Convergence and Optimality Qiaomin Xie, U. Wisconsin-Madison, Madson, WI, 14850, United States, Zhuoran Yang, Zhaoran Wang, Andreea Minca We study reinforcement learning in mean-field games. To achieve the Nash equilibrium, which consists of a policy and a mean-field state, existing algorithms require obtaining the optimal policy while fixing any mean-field state. In practice, however, the policy and the mean-field state evolve simultaneously, as each agent is learning while playing. To bridge such a gap, we propose a fictitious play algorithm, which alternatively updates the policy (learning) and the mean-field state (playing) by one step of policy optimization and gradient descent, respectively. Despite the non-stationarity induced by such an alternating scheme, we prove that the proposed algorithm converges to the Nash equilibrium with an explicit convergence rate. To the best of our knowledge, it is the first provably efficient algorithm that achieves learning while playing. 2 - Average Cost Deep Reinforcement Learning for Queuing Networks Mark Gluzman, Cornell University, Ithaca, NY, 14850-6311, United States, Jim Dai In queueing control problems, the long-run average cost is a natural performance measure instead of commonly used discounted objectives. We propose a novel bound on the difference of the discounted returns for two policies and show that it converges to a bound for average costs. We further generalize the bound on MDPs with unbounded cost functions, infinite state spaces, and long-run average cost objectives. Our bound leads to a version of Proximal policy optimization algorithm that can monotonically improve controls of queueing networks. Numerical results for multi-class queueing networks, a ride-hailing model, and a We investigate fluid scaling of single server queues operating under a version of shortest job first (SJF) where the priority level undergoes aging. That is, a job’s priority level is initialized by its size and varies smoothly in time according to an ordinary differential equation. Linear and exponential aging rules are special cases of this model. This policy can be regarded as an interpolation between FIFO and SJF. We use the measure-valued Skorokhod map to characterize the fluid model and establish convergence under fluid scale. We treat in detail examples of linear and exponential aging rules and provide a performance criterion based on our main result. 4 - Fluid Limits of Queue-based CSMA, Homogenization and Reflection Eyal Castiel, Technion, Haifa, Israel In this talk, we will discuss the fluid limits of a queueing process with QB-CSMA scheduling policy for general interference graphs. Introduced in 2009, this algorithm aims at mimicking the behavior of the celebrated Max-Weight algorithm by Tassiulas and Ephremides in a fully distributed fashion. The key element of the analysis is a fully coupled stochastic averaging principle where the schedule evolves much faster than queue lengths and we can replace the service rates by an invariant measure ‘adapted’ to the current queue lengths. This approximation fails in a neighborhood of zero but we will be able to overcome this difficulty in the case of a complete interference graph through a coupling argument. hospital inpatient operation model will be provided. 3 - Fluid Limits for Shortest Job First with Aging Yonatan Shadmi, Technion, Haifa, Israel

TD10 CC Room 304B In Person: Analytics in Supply Chain Networks General Session Chair: Deniz Besik, University of Richmond, Richmond, VA, 23220, United States 1 - Vulnerability of Global Supply Chains: Impact of Industrial and Geopolitical Concentration of Upstream Industries Jafar Namdar, University of Iowa, Iowa City, IA, 52246, United States, Gautam Pant, Jennifer Blackhurst In addition to the well-known notion of concentration due to the dominance of a few firms within an industry (i.e., industrial concentration), our study highlights the role of a relatively under-recognized dimension of geopolitical concentration. Using a large data set of firms, we find that the sales growth of firms whose suppliers are operating in high concentrated industries saw a relative drop of 5 percentage points during the pandemic. Our robust findings provide insights for both policy makers and managers for mitigating the supply chain risks stemming from industrial concentrations. 2 - Enhancing a Multi-commodity Supply Chain Network Resilience Through Fairness-based Distribution During Disruptions Andres David Gonzalez, University of Oklahoma, Norman, OK, 73019, United States, Osamah Y. Moshebah The flow of multi-commodities in supply chain networks SCNs can be significantly impacted by any sort of disruption in the network, in particularly the road transportation network. Enhancing the SCN performance can be ensured by implementing a fair-based distribution of commodities while restoring the SCN resilience and maintaining higher satisfaction rate at most of the demand nodes. 3 - An Integrated Multitiered Supply Chain Network Model of Competing Agricultural Firms and Processing Firms Deniz Besik, University of Richmond, Richmond, VA, 23220, United States, Anna B. Nagurney, Pritha Dutta The COVID-19 pandemic has created many disruptions in the agricultural supply chain networks, encompassing production, processing, packaging, storage, and distribution, affecting many stakeholders in the agriculture industry. This presentation shows an integrated multitiered competitive agricultural supply chain network model in which agricultural firms and processing firms compete to sell their differentiated products at the demand markets. The competition among agricultural firms and processing firms at the demand markets is formulated and studied via game theory, where the governing Cournot-Nash equilibrium conditions correspond to a variational inequality problem. We use an algorithm to test our modeling framework through a numerical study consisting of several supply chain disruption scenarios, including ones relevant to the Covid-19 pandemic. TD11 CC Room 304C In Person: Economics and Computation III Award Session Chair: Young-San Lin, Perdue University Co-Chair: Alexander Wei, UC Berkeley 1 - Revenue Maximization under Unknown Private Values with Non Obligatory Inspection Ali Makhdoumi, Duke University, Durham, NC, United States, Azarakhsh Malekian, Saeed Alaei We consider the problem of selling k units of an item to n unit-demand buyers to maximize revenue, where the buyers’ values are independently distributed according to publicly known distributions but unknown to the buyers themselves, with the option of allowing buyers to inspect the item at a cost. This problem can be interpreted as a revenue-maximizing variant of Weitzman’s Pandora’s problem with non-obligatory inspection. We present an approximation mechanism that achieves 1-\frac{1}{\sqrt{k+3}} of the optimal revenue. Our mechanism continues to work in an online setting where buyers arrive in an arbitrary unknown order. 2 - Allocation with Weak Priorities and General Constraints Young-San Lin, Purdue University, West Lafayette, IN, United States, Hai Nguyen, Thanh Nguyen, Kemal Altinkemer With COVID 19 prevalent in the world, efficient social distance seating becomes an option for sport venues, which allows only a fraction of the capacity and necessitates reassigning spectators to games. We model this as a resource allocation problem and develop a mechanism based on a new concept called Competitive Stable Equilibrium. Its novelty is the combination of three features: complex resource constraints, weak priority rankings over the agents, and ordinal preferences over bundles of resources. We empirically apply our mechanism to reassign season tickets to families and show that our method outperforms existing ones in both efficiency and fairness measures.

128