Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

POSTERS

2 - Clustering with Incomplete Proximity Matrices Samira Karimzadeh, Iowa State University, Ames, IA, 50011, United States, Sigurdur Olafsson In data clustering, we sometimes only have an incomplete proximity matrix available to measure distance. This compromises the quality of the clusters, especially when values are not missing at random. Addressing this, our study proposes an effective graph theory method to complete values that are not missing at random. 3 - Sentence Embedding Module Satisfying a Characteristic of Human Language Recognition Myeongjun Jang, Korea University, Seoul, Korea, Republic of, Pilsung Kang Sentence embedding is an important topic in natural language processing (NLP). It is essential to generate a good embedding vector to enhance performance for many NLP tasks. So far, various models have been proposed and claim their superiorities through the good performance for sentiment analysis and text classification. However, since the performance of those tasks can be enhanced by using a simple sentence representation method, it is not sufficient to claim that they are good embedding methods. In this paper, inspired by human language recognition, we suggest the concept of semantic coherence that a good sentence embedding method should satisfy and propose the model to pursue this property. 4 - Prediction of Performance Deterioration for Solid Oxide Fuel Cell System Jumpei Kawasaki, Tokyo Gas Co Ltd, Minato-ku, Tokyo, 105-8527, Japan Although solid oxide fuel cell is an environmentally friendly system with high power generation efficiency and low carbon dioxide emissions, early fault detection and prediction of performance deterioration of the system components, especially fuel cell stacks, is required for further operation stabilization and maintenance optimization. In this study, the correlation of about 300 system parameters measured for 4000 hours and the power generation performance of the fuel cell stack is examined using machine learning techinique, and performance deterioration prediction method is constructed which is expected to contribute for stabilization and cost reduction of operation. 5 - ELM-SOM: A Continuous Self-organizing Map for Visualization Renjie Hu, Research Assistant Professor, University of Iowa, Iowa City, IA, 52242, United States This paper presents a novel dimensionality reduction technique: ELM-SOM. This technique preserves the intrinsic quality of Self-Organizing Maps (SOM): it is nonlinear and suitable for big data. It also brings continuity to the projection using two Extreme Learning Machine (ELM) models, the first one to perform the dimensionality reduction and the second one to perform the reconstruction. ELM-SOM is tested successfully on six diverse datasets. Regarding reconstruction error, ELM-SOM is comparable to SOM while bringing continuity. 6 - Deep Convolutional Networks for Forgery Classification and Anomaly Detection Zohreh Raziei, Southern Methodist University, Dallas, TX, United States, Xinyi Ding, Eric Larson, Michael Hahsler, Paul Krueger, Eli Olinick We apply Convolutional Neural Networks (CNN) to identify fake photographs. Using medium-to-high-resolution images, we combine an auto-encoder-based neural network with explicit facial modeling to generate swapped faces of celebrities and compare the CNN’s performance to approximate pairwise rankings inferred from judgement of human subjects. 7 - Optimal Design of Experiments on Riemannian Manifolds Hang Li, Pennsylvania State University, State College, PA, 16803, United States, Enrique Del Castillo, George Runger In recent years, scientists and engineers often need to deal with large volumes of high-dimensional data. Sometimes these data are available in a high-dimensional ambient space but they truly lie on a lower-dimensional manifold. The objective of this research is to develop theory of optimal experimental design on manifold data. In particular, we prove a new Equivalence Theorem for continuous optimal design on Riemannian manifold, and also provide a converging algorithm to find the optimal design. 8 - WeCureX Intelligent Psychiatry Assistant Salih Tutun, Binghamton University (SUNY),Johnson City, NY, 13790, United States, Sedat Irgil, Ilker Yesilkaya, Ahmet Aykac, Nilay Aras, Begum Basaran Tutun The aim of this research is to propose a new intelligent system (namely WeCureX) by using artificial intelligence approaches and psychometrics values for detecting major symptoms of social and personal maladjustment, assessing medical patients and design effective treatment strategies with high accuracy rates. Therefore, in this research, we will show how to identify symptoms for the SCL-90 test. We proposed optimized Lasso Logistic Regression and outlier detection. Finally, the results show that it works with 97% accuracy, and this system can be used by experts and patients for better treatment.

9 - A Framework for Five Big V’s of Big Data and Organizational Culture in Firms Thuan Nguyen, University of North Texas, Denton, TX, 76203, United States Based on Cameron and Quinn’s organizational cultural model, this study proposes a theoretical framework that describes how each type organizational culture - hierarchy, clan, adhocracy, and market - has an impact on each Big V of big data. The framework suggests that firms, influenced by their organizational culture, have different views on how important each Big V’s should be. The study argues that organizations should develop, nurture, and maintain an adhocracy organizational culture that has a positive impact on each of the five Big V’s of big data to harness the full potential of big data. 10 - Two Stage Aggregate Production Planning with Flexible Requirement Profiles Setareh Torabzadeh, University of North Carolina-Charlotte, Charlotte, NC, 28262, United States, storabza@uncc.edu, Ertunga Ozelkan This Research investigates on the aggregate production planning problem, using a flexible optimization approach, called: Flexible Requirement Profile, which enforces flexible bounds on production levels in different planning periods to improve the stability of the production plans. Due to the uncertainty of future periods demand estimation, the stochastic programming approach is incorporated into the optimization framework. 11 - Degradation Models of Biopolymers Under Accelerated Weathering Conditions Elias Arias Nava, New Mexico State University, Las Cruces, NM, 88001, United States, Delia Valles-Rosales Research in renewable products with the potential to replace fossilized matter as raw materials for energy and materials use is at the forefront of modern science and engineering. This project designs new degradation models that could provide reliable lifetime data of biopolymers. The goal associated with this study is to present a new degradation model(s). Specimens were manufactured under ASTM standards using extrusion and injection molding; followed by 2000 hours of accelerated degradation test. The accelerating variables were: temperature, humidity, UV exposure and time. Multivariate analysis of five responses were analyzed: tensile, flexural, color change, mass loss, and FTIR analysis. 12 - Berth Scheduling at Marine Container Terminals: Minimizing Carbon Dioxide Emissions Due to Container Handling Maxim A. Dulebenets, Florida A&M University-Florida State University, Tallahassee, FL, 32311, United States, Olumide Abioye, Ren Moses, Eren Erman Ozguven, Arda Vanli, Thobias Sando This study presents a mixed integer programming model for the green berth scheduling problem, which accounts not only the total vessel service costs, but also considers the cost associated with the carbon dioxide emissions produced due to container handling. A set of Hybrid Evolutionary Algorithms are proposed to solve the mathematical model. Numerical experiments demonstrate that introduction of local search heuristics within the developed algorithms improved the quality of the obtained solutions. Furthermore, changes in the carbon dioxide emission cost influence the berth schedule design. 13 - Dynamically Scheduling National Football League Games to Reduce Strength of Schedule Variability Jamie Fravel, Furman University, Greenville, SC, 29613, United States, Elizabeth Bouzarth, Andrew Cromer, Ben Grannan, Kevin Hutson The NFL schedules games where some matchups are based on the previous year’s results. Since team composition changes from year to year, this scheduling policy sometimes benefits teams unfairly, allowing some an easier path to the playoffs than others. Thus, strength of schedules vary between teams and arguments have to be made why some teams make the playoffs and others do not. We propose methods to produce an NFL schedule that combines some of its traditional elements with dynamically-scheduled games aimed at optimizing different objectives, such as reducing the variability of teams’ strengths of schedules or minimizing the number of pairwise comparisons needed to differentiate team quality. 15 - User-based Rebalancing Approach for Free-floating Bike Sharing Systems Yan Wang, The University of Hong Kong, Hong Kong, Hong Kong, Junwei Wang This work studies the rebalancing problem in FFBS, which aims at reconfiguring the spatial and temporal distribution of bikes so that supply matches demand. We propose the first model specially designed for FFBS. Based on our original Radiant Service Theory (RST), the model describes the real-time status of the system. Also, we develop the first user-based rebalancing approach for FFBS. A dynamic incentive mechanism that encourages users to change parking places to problematic locations is presented. Finally, simulation results show the effectiveness of the mechanism and suggest it best fits a system with modest bike amount, high unbalance degree, and users sensitive to incentives.

339