2016 INFORMS Annual Meeting Program

MA03

INFORMS Nashville – 2016

SD93 Davidson Ballroom C-MCC Panel: Trends in Service Systems Research Funded by NSF: Overview of Opportunities for the Human-technology Frontier Panel Session Moderator: Alexandra Medina-Borja, NSF/ UPRM, PFI: BIC - Smart Service Systems, Falls Church, VA, United States, amedinab@nsf.gov 1 - Trends In Service Systems Research funded byNSF: Overview Of Opportunities For The human-technology frontier Alexandra Medina-Borja, US National Science Foundation, Arlington, VA, 2, United States, alexandra.medinaborja@upr.edu An overview of interdisciplinary funding opportunities for researchers modeling the interaction between humans and engineered systems that could enable the smart service systems of the future. Requirements and opportunities will be discussed by one of the NSF cognizant program officers in this program. 2 - Panelist David Mendonca, Rensselaer Polytechnic Institute, New York, NY, 0, United States, mendod@rpi.edu Monday, 8:00AM - 9:30AM Chair: Neng Fan, University of Arizona, Engineering Bldg, Room 312, 1127 E. James E. Rogers Way, Tucson, AZ, 85721, United States, nfan@email.arizona.edu 1 - Multiclass Support Vector Machines With Labeling Uncertainty Wanlu Gu, University of Arizona, wanlugu@email.arizona.edu The multiclass support vector machines (SVMs) is an extension of the conventional SVM in machine learning, and it builds the classification hyperplanes based on a set of training data points. In practice, the real collected data may have noise or uncertainty. In this talk, we consider the observed data with noise on the labels, and construct models and algorithms to learn from this type of uncertain data. To model the uncertainty, the noise probability is assumed to the labeling noise from one class to the others. Then some novel optimization models are proposed and also validated through numerical experiments to check the difference with noise free models. 2 - Sparse Support Vector Machines With Data Uncertainty Ammon Washburn, University of Arizona, wammonj@email.arizona.edu Data with high dimensionality and uncertainty comes about when there are too many features and the data is unreliable, replicated or missing. Without taking these issues properly into account, classification models will overfit the training data. In order to deal with these two problems, sparse representations and chance-constrained programming have emerged separately. We will show how to implement both ideas by modifying Support Vector Machines in a way that is not overly conservative which we call Decoupled Margin-Moment SVM. Numerical experiments are performed on collected pancreatic cancer data. 3 - Graph Clustering Of Data With Uncertainties Yujia Zhang, University of Arizona, yujiazhang@email.arizona.edu In this talk, we will review models and algorithms for clustering of data with uncertainties. First, the methods to model data uncertainty will be reviewed. Second, we mainly concentrate on the graph models for clustering. Finally, algorithms for solving these models will be reviewed and compared. 4 - Constrained Clustering Of Uncertain Data Derya Dinler, PhD Candidate, Middle East Technical University, ODTU Endustri Muhendisligi Bolumu, Cankaya, Ankara, 06800, Turkey, dinler@metu.edu.tr, Mustafa Kemal Tural We consider a constrained clustering problem where the locations of the data objects are subject to uncertainty. Each uncertainty set is assumed to be either a closed convex bounded polygon or a closed disk. The final clustering is expected to be in accordance with a given number of instance level constraints. We propose a mixed-integer second order cone programming formulation for the considered clustering problem which is only able to solve small-size instances. For larger instances, approaches from the semi-supervised (constrained) clustering literature are modified and compared in terms of computational time and quality. MA01 101A-MCC Learning from Uncertain Data with High Dimensionality Sponsored: Data Mining Sponsored Session

MA02 101B-MCC New Advancements in Using Data Analytics for Healthcare Applications Sponsored: Data Mining Sponsored Session Chair: Talayeh Razzaghi, Clemson University, 100 McAdams Hall, Clemson, SC, 29634, United States, talayeh.razzaghi@gmail.com 1 - Using Density To Identify Fixations In Gaze Data: Optimization-based Formulations And Algorithms Andrew C Trapp, Worcester Polytechnic Institute, atrapp@wpi.edu Eye tracking is an increasingly common technology with a variety of practical uses. Eye-tracking gaze data can be categorized into two main events: fixations, which represent attention, whereas saccades occur between fixation events. We propose a novel manner to identify fixations based on their density, which concerns both the fixation duration as well as its inter-point proximity. We develop two mixed-integer nonlinear programming formulations and corresponding algorithms to recover the densest fixations in a data set. Our approach is parameterized by a unique value that controls for the degree of desired density. We conclude by discussing computational results and insights on real data sets. 2 - Leveraging Longitudinal Healthcare Data For Inverse Classification Michael Lash, University of Iowa, michael-lash@uiowa.edu, Nick Street Inverse classification is the process of manipulating a test point to minimize the predicted probability of a specific class label. Such a process has been shown to be beneficial to healthcare-related problems such as lifestyle modification and treatment recommendations. Past work in this area has focused on single snapshots in time, which does not account for the history or behavioral changes of patients. In this work we incorporate longitudinal information into our inverse classification model and demonstrate its effectiveness in mitigating the long-term risk of cardiovascular disease. 3 - Stability And Performance Of Healthcare Access Supply-demand Systems Affected by Stochastic Time Delays Sara Nourazari, California State University-Long Beach, Bellflower Boulevard, Long Beach, CA, 90840, United States, Sara.Nourazari@csulb.edu, Rifat Sipahi, James Benneyan Time delays are an inevitable aspect of many healthcare supply-demand systems and can potentially lead to undesirable outcomes and decision making challenges. We propose an approach to characterize “expected” stability maps of healthcare access supply-demand systems affected by random time delays following a known probability distribution. This study aims to enable broader insight into the effects of random process delays, and across a wide range of test applications in healthcare queue management, demonstrates minimized undesirable oscillatory behaviors and improved system performance.

MA03 101C-MCC

Daniel H. Wagner Prize Competition I Invited: Daniel H. Wagner Prize Competition Invited Session

Chair: C. Allen Butler, Daniel H Wagner Associates, Inc., 2 Eaton Street, Hampton, VA, 23669, United States, Allen.Butler@va.wagner.com 1 - Calibrated Route Finder – Social, Environmental And Cost-effective Truck Routing Mikael Ronnqvist, Professor, Universite Laval, Pavillion Adrien- Pouliot, Bureau 3345, 1065 Avenue De La Medecine, Quebec, QC, Finding the best route with many conflicting objectives is very difficult. The online system Calibrated Route Finder has been developed in collaboration among many companies and organizations and successfully addresses the problem. A key component is an inverse optimization process that establishes more than 100 weights to balance social values, environmental impacts, traffic safety, stress, fuel consumption, CO2 emissions, and costs. In addition, methodological and analytic developments now enable measurement and inclusion of perceived hilliness and curviness as well as strict rules where to drive. The system has been in operations since 2009 and is today used by about 100 companies. G1V OA6, Canada, mikael.ronnqvist@gmc.ulaval.ca Gunnar Svenson, Patrik Flisberg, Lars-Erik Jönsson

121