Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

TB03

4 - Two-Stage Robust Optimization with Decision-Dependent Information Discovery Phebe Vayanos, University of Southern California, OHE 310L, University Park Campus, USC, Viterbi School of Engineering, Los Angeles, CA, 90089, United States, Angelos Georghiou, Jiachuan Chen We consider two-stage robust optimization problems in which part of the first stage variables decide on the uncertain parameters that will be observable in the second stage (as in sensor positioning, patrolling). The state of the art formulates the problem as a two-stage robust problem with decision-dependent non- anticipativity constraints and uses binary decision rules over a preselected partition of the uncertainty set resulting in very conservative solutions. We propose a novel min-max-min-max formulation and a solution method based on the K-adaptability idea. We reformulate the problem as an MILP solvable with off-the-shelf solvers and demonstrate its effectiveness on stylized problems. n TB02 North Bldg 121B Sparsity in Regression Sponsored: Optimization/Optimization Under Uncertainty Sponsored Session Chair: Dimitris Bertsimas, Massachusetts Institute of Technology, Cambridge, MA, 02139, United States 1 - A Scalable Algorithm for Sparse and Robust Portfolios Ryan Cory-Wright, PhD Student, Massachusetts Institute of Technology, Cambridge, MA, United States, Dimitris Bertsimas We present a cutting-plane method which solves the sparse Markowitz portfolio problem to provable optimality at scale, by exploiting a dual representation of the continuous problem to obtain a closed form representation of the problem’s subgradients. We refine the cutting-plane method by deriving an efficient local- search heuristic which exploits these subgradients, embedding the heuristic within the cutting-plane method, and exploiting a correspondence between the convexified dual problem and a rotated QCQP to obtain a strong apriori lower bound. We illustrate the method’s effectiveness by obtaining optimal sparse frontiers for major indices including the S&P 500 and the Wilshire 5000. 2 - Sparse Regression Over Clusters: Sparclur Lea Kapelevich, Massachusetts Institute of Technology, 77 Massachusetts Avenue, Bldg. E40-103, Cambridge, MA, United States, Dimitris Bertsimas, Jack W. Dunn, Rebecca Zhang We aim to develop machine learning models that combine state of the art accuracy and interpretability. Sparse regression models and decision trees are machine learning methods that aspire to achieve both these properties. In this work, we generalize ideas from new developments in sparse regression and optimal regression trees. We present an integer programming approach for regression tasks arising in tree-based machine learning models, and apply it to prediction problems in healthcare. 3 - Sparse Regression Scalable Algorithms and Empirical Performance Jean Pauphilet, MIT, Cambridge, MA, 02139, United States, Dimitris Bertsimas, Bart Paul Gerard Van Parys We address the problem of sparse linear and logistic regression from a discrete optimization perspective. We formulate the problem as a convex integer optimization problem and solve it efficiently using a cutting-plane algorithm. We also propose a fast sub-gradient algorithm to solve its Boolean relaxation and compare our approach with L1 regularization and two methods with non-convex penalties. We demonstrate empirically the performance of our methods in terms of accuracy, false detection rate and computational time, for different regimes of noise and correlation. 4 - Optimistic Robust Optimization with Application to Sparse Regression and Classification Matthew Norton, Naval Postgraduate School, 1 University Circle, Monterey, CA, 93943, United States Robust Optimization has traditionally taken a pessimistic, or worst-case viewpoint of uncertainty. We explore an optimistic, or best-case view of uncertainty and show that it can be a fruitful approach to address a wide variety of problems. In the context of robust linear programming, we provide an intuitive method for reducing conservatism. We also find that many problems in machine learning and robust statistics can be interpreted as optimistic robust optimization problems. This includes popular sparsity inducing non-convex regularization schemes and outlier protection methods.

n TB03 North Bldg 121C Managing IT-based Technology and Services Sponsored: Technology, Innovation Management & Entrepreneurship Sponsored Session Chair: Juliana Hsuan, Copenhagen Business School, Copenhagen Modeling Health Technology Adoption by Elderly Women Charles M. Weber, Portland State University, Engineering and Technology Management, P.O. Box 751, Portland, OR, 97207, United States,, Noshad Rahimi Abstract Modeling how patients adopt personal health technology is a challenging problem: Decision-making processes are largely unknown, occur in complex, multi-stakeholder settings, and may play out differently for different products and users. This paper develops a soft analytics approach, based on Fuzzy Cognitive Maps, which is empirically grounded in a case study of how a group of elderly women adopts wearable devices. The approach leads to an adoption model that simulates different product configurations and scenarios that will most likely lead to successful adoption. The model can be used by product developers and rollout Nitin Mayande, Tellagence, 6249 NE Carillion Drive, Unit 201, Hillsboro, OR, 97124-8097, United States, Charles Weber Network Modularity has been used in many previous studies for community detection. The focus of these studies has mostly been on how people connect with each other to form communities. This exploratory study, instead of people, focuses on contextualized communication (knowledge) and explores to see if Network Modularity measure can be used to quantify Knowledge Modularity thereby identifying knowledge structures within a social network. 3 - Uno or Duo? On Creation and Discovery Entrepreneurship Opportunities and Their Interactions John N. Angelis, Elizabethtown College, Elizabethtown, PA, United States, Moren Levesque, Richard Arend While the interaction process between opportunity and the entrepreneur who exploits it has been well researched, research on outcomes has been limited. We provide a formalized explanation of how the two main types of entrepreneurial opportunity (i.e., discovery and creation) interrelate across six specific cases. We provide analysis of a modified model of an established partial-equilibrium production chain. Via theory and simulation, we explain how exploitation of creation opportunities can lead to discovery opportunities; how creation (discovery) opportunities are less (more) likely to be profitable when paired with other opportunity type; and spillover and transfer payment scenarios. 4 - Pricing and Service Bundling at a Smartphone Provider Eric Bentzen, Copenhagen Business School, Frederiksberg, Denmark, Juliana Hsuan Many of the attributes with respect to pricing services are implied by economic theory, could be modeled, and investigated using relevant data from a company. In this paper, we will study relationship that a large smartphone service provider should be aware of and use in the pricing of bundling services. From a smartphone company we have access to customers and their use of cell phone services, Internet access and other services provided by the company. The dataset has been collected during a period and includes the amount that a large number of customers are willing to pay for each service. managers to support technology planning decisions. 2 - Knowledge Modularity in Social Networks Business School, Frederiksberg, DK-2000, Denmark 1 - Soft Data Analytics with Fuzzy Cognitive Maps:

275