Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

MB10

3 - The Weighted Target Set Selection Problem on Cycles Rui Zhang, Assistant Professor, University of Colorado Boulder, Boulder, CO, 80309, United States, Subramanian Raghavan The study of viral marketing strategies on social networks has become an area of significant research interest. In this setting, we consider the weighted target set selection (WTSS) problem. Motivated by the desire to develop a better understanding of fundamental problems in social network analytics, we focus on a special case where the underlying graphs are cycles. First, we propose a linear time algorithm for the WTSS problem on cycles. More importantly, we present a tight formulation in the node space. It provides a complete description of this polytope. Our work can be a building block for developing exact methods for tackling this important problem in social network analytics. 4 - A Shortest Path Interdiction Problem with Improvement Timothy Holzmann, Clemson University, Engineering, Computing and Applied Sciences, Freeman Hall, Clemson, SC, 29634, United States, Cole Smith We consider a shortest path interdiction problem with improvement. As in standard interdiction problems a leader and a follower play a Stackelberg game, wherein the leader seeks to maximize the follower’s shortest path between two nodes on a graph. In our variant, in addition to the leader raising edge costs with a budget, the follower also has an opportunity to lower costs within a budget. We employ a multiobjective optimization model for problem where the leader is uncertain of the follower’s budget. We present two solution methods and consider their comparative strengths and weaknesses. 5 - New Integer Programming Formulations for Detecting Critical Nodes Hosseinali Salemi, Oklahoma State University, 15 North University Place, Apartment Number 2, Stillwater, OK, 74075, United States, Austin Buchanan Recently, researchers have studied the problem of identifying so-called “critical nodes in networks, which are small subsets of nodes whose deletion maximally deteriorates a network. In this talk, we consider variants of this problem in which the task is to delete at most B nodes so as to minimize the number of node pairs that remain connected by: (1) a path, or (2) a path of length at most k. We propose new cut-like IP formulations for these problems and compare with previous approaches. n MB09 North Bldg 124B Intersections of Optimization and Statistical Learning Sponsored: Optimization/Nonlinear Programming Sponsored Session Chair: Sam Davanloo Tajbakhsh, The Ohio State University, Columbus, OH, 43210, United States 1 - Rank Minimization Based Sparse Data Reconstruction in Sensory Networks Constantino Lagoa, PSU, State College, PA, United States, Ashkan Jasour In this work, a novel approach to reconstruct a noisy sparse n-dimensional data is proposed. The main idea is to complete the data with least possible complexity. The complexity is defined as the number of exponential signals that could describe the data. We show that the number of exponential signals corresponds to the rank of block Hankel matrix constructed from given data. In this context, the problem of data reconstruction can be reformulated as matrix completion and rank minimization problems where the nuclear norm is used as a convex relaxation of the matrix rank. A first-order augmented Lagrangian algorithm is implemented for solving the resulting optimization problem. 2 - On Stochastic Quasi Newton Methods for Illposed Stochastic Optimization Problems Farzad Yousefian, Oklahoma State University, 317D Engineering North, School of Industrial Engineering & Management, Stil, OK, 74074, United States, Angelia Nedich, Uday Shanbhag We consider stochastic quasi-Newton (SQN) methods for solving unconstrained convex optimization problems. Much of the convergence analysis of SQN methods, in both full and limited-memory regimes, requires the objective function to be strongly convex and have Lipschitzian gradients. We develop SQN methods addressing absence of these assumptions. Particularly, we develop a regularized stochastic limited-memory BFGS algorithm, where the stepsize, regularization parameter, and the Hessian inverse approximate are updated iteratively. We establish convergence of the iterates to an optimal solution of the original problem. A convergence rate is derived in terms of the objective function.

3 - Hierarchical Sparse Modeling Jacob Bien, University of Southern California, Los Angeles, CA, United States It is common in statistics to demand sparsity patterns honoring certain problem- specific constraints. In hierarchical sparse modeling (HSM), these constraints specify that one set of parameters be set to zero whenever another is set to zero. Numerous papers have developed convex regularizers for this sparsity structure, which arises in many areas of statistics including interaction modeling and covariance estimation. We observe that these methods are based on two different types of group lasso that have not been systematically compared in the context of HSM. We investigate the differences both in terms of statistical properties and computational efficiency. Joint work with Xiaohan Yan. 4 - Distributed Proximal Algorithms for Statistical Learning with Structured Sparsity Sam Davanloo Tajbakhsh, The Ohio State University, 210 Baker Systems Building, 1971 Neil Ave, Columbus, OH, 43210, United States In some sparse statistical learning problems, solutions should follow sparsity structures between variables known a priori. Being able to generate such structures helps to obtain more interpretable models. Our study focuses on hierarchical sparsity structures represented as Directed Acyclic Graphs. Designed penalty functions exploit group overlaps to induce solutions with desired hierarchical structures. In this talk, we will present new distributed proximal algorithms to solve the underlying optimization problems with theoretical convergence guarantees. Some numerical results supporting the proposed algorithms will be provided. Saif Benjaafar, University of Minnesota, 111 Church Street SE, Department of Industrial and Systems Engr, Minneapolis, MN, 55419, United States, Jian-Ya Ding, Guangwen Kong, Terry Taylor We study labor welfare in on-demand service platforms that rely on independent agents. The platform chooses wages and prices knowing that customers are sensitive to both price and delay. In contrast to settings where customers are insensitive to delay, we show that an increase in labor supply does not necessarily Ming Hu, University of Toronto, Rotman School of Management, 105 St. George Street, Toronto, ON, M5S 3E6, Canada, Yan Liu We study competition of on-demand platforms, where firms compete on both supply and demand sides. We consider various modes of competition, and study their implications on platforms’ profit, prices on the demand side, wages on the supply side, and matching quantities at equilibrium. We also establish Kreps- Scheinkman equivalence in this type of two-sided markets. 3 - Peak-period Pricing Strategies in the Presence of Customer Impatience and Store and Time Flexibility Steve Yoo, University College London, London, United Kingdom, Christopher S. Tang Should a service firm charge higher prices during peak periods? We examine this question formally by analyzing a stylized duopoly model where firms compete for homogeneously impatient consumers. We consider four settings, defined by whether consumers are (i) flexible in their choice of store (where) and/or (ii) flexible in their choice of shopping time (when). For each setting, we use the concept of a rational expectation equilibrium to characterize how consumers endogenously segment themselves regarding where and when to shop to avoid congestion. We examine how the firms can profitably influence consumers’ self- segmentation process by employing the peak-period pricing strategy. 4 - Rewarding Loyalty: Bonus Design on Ridesharing Platforms Puping Jiang, Olin Business School, Washington University in St. Louis, Washington University Olin Business School, Knight Hall, Campus Box 1156, Saint Louis, MO, 63130-4899, United States, Kaitlin Daniels, Fuqiang Zhang Ridesharing platforms commonly offer drivers a bonus payment that is contingent on the number of rides completed. The bonus constitutes a costly incentive that encourages drivers to drive exclusively for one platform. In this paper, we explain the existence of bonuses in the context of competing platforms and describe the market conditions that lead to their use. result in lower wages and lower labor welfare. 2 - Competition of On-demand Platforms n MB10 North Bldg 125A On-demand Service Platforms Sponsored: Manufacturing & Service Oper Mgmt Sponsored Session Chair: Kaitlin Daniels, Washington University 1 - Labor Welfare in On-demand Service Platforms

151