Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

SB09

capture key features of various policy issues with relatively simple “first-strike” models. Problem selection and formulation thus compete with the mathematics of solution methods in determining successful applications. I will review some personal adventures in policy modeling selected from public housing, HIV/AIDS prevention, bioterror preparedness, counterterrorism, and immigration policy. n SB11 North Bldg 125B Operational Issues in Agriculture Sponsored: Manufacturing & Service Oper Mgmt/iFORM Sponsored Session Chair: Buket Avci, Singapore Management University, 50 Stamford Road, Singapore, 178899, Singapore 1 - Optimal Procurement from Multiple Contracts in Agricultural Processing: Implications for Biomass Commercialization Buket Avci, Singapore Management University, 50 Stamford Road, Singapore, 178899, Singapore, Onur Boyabatli, Bin LI We study the optimal procurement strategy of an agricultural processor that procures a commodity input from multiple contracts to produce a commodity output in the presence of input and output spot price uncertainties. We investigate the impact of spot price correlation on the procurement strategy and profitability. We also examine the impact of commercializing output waste as biomass on the profitability and greenhouse gas emissions. Calibrating our model with data from palm industry, we show that failing to update procurement strategy after commercialization leads to a significant profit loss and commercialization of biomass can increase emissions due to change in the procurement strategy. 2 - Integrated Optimization of Fertilizer Procurement, Cultivation, and Harvesting We consider the integrated optimization of three key operational decisions in farm planning: fertilizer procurement via both contract and spot markets; cultivation and harvesting with limited respective resources in the presence of uncertain farm yield and labor cost. We model this problem as a three-stage stochastic program and characterize the optimal decisions in each stage. Using a numerical study calibrated to real data, we characterize the value of the integrated optimization and examine how the uncertainties in the farm yield and labor cost affect the farm profitability, cultivation decisions, and two measures of food loss: un-harvested yield and the yield gap. 3 - Food Safety Control in the Developing Economies. Centralization Versus Decentralization Duo Shi, The Chinese University of Hong Kong, Shenzhen, Shenzhen, China, Lingxiu Dong, Iva Petrova Rashkova We analytically investigate how the food safety outcome is affected by the structure of government auditing agencies in the developing economies. We find that a decentralized control structure may beat a centralized control structure, and higher auditing capabilities from the agencies may lead to a worse food safety outcome. We also propose a responsibility reallocation approach to further improve the current decentralized structure. Yangfang (Helen) Zhou, Assistant Professor, Singapore Management University, Singapore, 178899, Singapore, Onur Boyabatli, Lusheng Shao

n SB09 North Bldg 124B First Order Methods for Saddle Point Problems Sponsored: Optimization/Nonlinear Programming Sponsored Session Chair: Necdet Serhat Aybat, University Park, PA, 16802, United States 1 - Primal-dual Stochastic Gradient Method Yangyang Xu, Rensselaer Polytechnic Institute, Department of Mathematical Sciences, 110 8th Street, Troy, NY, 12118, United States Stochastic gradient (SG) method has been used for problems with objective that is stochastic or an average of many functions. Most works on SG assume that the problem is unconstrained or has an easy-to-project constraint set. In this talk, I consider problems with a stochastic objective and also many functional constraints. For these problems, we propose a novel SG method based on augmented Lagrangian function. Each update, it inquires a stochastic subgradient of the objective, a subgradient and function value of one sampled constraint function, and function value of another sampled constraint function. Its convergence rate is shown. Numerical test on QCQP is conducted to show its efficiency. 2 - Accelerated Stochastic Algorithms for Nonconvex Finite-sum and Multi-block Optimization Yu Yang, ISyE Ga Tech, Atlanta, GA, 30332, United States, Guanghui Lan We first introduce a randomized accelerated proximal gradient (RapGrad) method for solving a class of nonconvex optimization problems consisting of the sum of m component functions, and show that it can significantly reduce the number of gradient computations especially when the condition number, i.e., the ratio between the Lipschitz constant and negative curvature, is large. Inspired by RapGrad, we also develop a new randomized accelerated proximal dual (RapDual) method for solving a class of multi-block nonconvex optimization problems coupled with linear constraints. We demonstrate that RapDual can also save up to a factor of ${\cal O}(\sqrt{m})$ projection subproblems than its deterministic counterpart, where $m$ denotes the number of blocks. 3 - Primal-dual Methods for General Convex-concave Saddle Point Problems Erfan Yazdandoost Hamedani, Pennsylvania State University, State College, PA, 16801, United States, Necdet Serhat Aybat In this research we propose a primal-dual algorithm with a momentum term that can be viewed as a generalization of the method proposed by Chambolle and Pock in 2016 to solve saddle point problems defined by a convex-concave function L(x,y)=f(x)+F(x,y)-h(y) with a general coupling term F(x,y) that is not assumed to be bilinear. We derive error bounds in terms of Lagrangian function for the ergodic sequence of iterates; in particular, we show O(1/k) rate when the problem is merely convex. Furthermore, assuming F(x,.) is linear in y for each fixed x and f is strongly convex, we can obtain the ergodic convergence rate of O(1/k2). 4 - Complexity of a Quadratic Penalty Accelerated Inexact Proximal Point Method for Solving Linearly Constrained Nonconvex Composite Programs Renato Monteiro, School of ISyE, Georgia Tech, Atlanta, GA, 30332, United States This talk discusses the complexity of a quadratic penalty accelerated inexact proximal point method for solving a linearly constrained nonconvex composite program. Its objective function is of the form f+h where f is a differentiable function whose gradient is Lipschitz continuous and h is a closed convex function with bounded domain. The method consists of applying an accelerated inexact proximal point method to solve a sequence of quadratic penalized subproblems associated to the linearly constrained problem. Each subproblem is then approximately solved by an accelerated composite gradient method. Numerical results showing the efficiency of the proposed method are given. n SB10 North Bldg 125A MSOM Distinguished Fellow Talk Sponsored: Manufacturing & Service Oper Mgmt Sponsored Session Chair: Gad Allon, University of Pennsylvania, 3730 Walnut Street, Philadelphia, PA, 19104, United States 1 - Selected Adventures in Policy Modeling Edward H. Kaplan, Yale University, Yale School of Management, Box 208200, New Haven, CT, 06520-8200, United States Policy Modeling refers to the application of operations research, statistics, and other quantitative methods to model policy problems. Recognizing that analyses of all sorts often exhibit diminishing returns in insight to effort, the hope is to

n SB12 North Bldg 126A Joint Session OR Frontiers/Integer Programming/ICS: Machine Learning and

Discrete Optimization I Emerging Topic: OR Frontiers Emerging Topic Session Chair: Bistra Dilkina, University of Southern California, Los Angeles, CA, 90089, United States 1 - Deep Learning for MIP Branch and Bound Bistra Dilkina, University of Southern California, 941 Bloom Walk, Los Angeles, CA, 90089, United States, Aaron Ferber, Elias B. Khalil Effective branching decisions are one of the key components to designing more scalable MIP branch and bound algorithms. We leverage a deep learning approach to obtain a predictive model that works well over distributions of similar instances. The features used as input to the model are generic so that transfer to new instances is seemless. Experimental results are encouraging.

34