Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

TC67

2 - A New Algorithm with Real-time Smoothing for Predicting Blood Glucose Concentrations Based on Wavelet Filters Lei Li, Beihang University, Beijing, China, Jun Yang Based on the Continuous Glucose Monitoring (CGM) data, we aim at predicting future blood glucose levles so that appropriate actions can be taken in advance to prevent hyper/hypoglycemia. Due to the small fluctuations of CGM data, an ARMA model with a wavelet filter is proposed in the prediction framework. To verify the performance of the proposed method, we conduct the proposed method with different wavelet function, different decomposition levels and threshold methods of wavelet denoise in a case study based on the CGM data of 5 diabetics. Results show that the proposed methods with db8 wavelet function and minimaxi threshold method has most satisfactory and robust performance. 3 - The Recognition Method of the Equipment State Based on the MTS Modified by FDA Ning Wang, Chang’an University, Middle-section of Nan’er Huan Road, Xi’an, ShaanXi, 710064, China, Dawei Hu, Yingbin Fu Mahalanobis-Taguchi System (MTS) is a kind of data classification and reduction method using Mahalanobis distance (MD) as the measurement scale to identify the system state with multidimensional characteristics. In this paper,against the imbalanced classification by the model to identify the sample when the benchmark and abnormal space constructed by the traditional MTS have a serious overlap, a modified MTS amended by Fischer linear discriminant analysis (FDA) is proposed, and to be used to recognize the running state of equipment. The result proves the effectiveness and superiority of the modified model. 4 - Scheduling of Maintenance Teams and Activities for Nuclear Power Plant Subsystems Meng-Yu Du, Tsinghua University, Beijing, China, Yan-Fu Li The existing research works on nuclear power plant (NPP) maintenance scheduling normally assume that activities are performed on the exact scheduled times. However, due to budget limit, the shortage of maintenance workers cannot be ignored and thus the plans can be missed. To deal with this problem, we propose an integrated scheduling of maintenance teams and activities. A mixed integer program model is built with the objective of minimizing maintenance cost. The lower bound of system unavailability, limited maintenance workers, etc. are regarded as constraints. The proposed optimization model is applied to an NPP subsystem and solved by a mathematical programming solver. Joint Session ISS/AI/Practice Curated: Applications of Deep Learning Methods in Information Systems Sponsored: Information Systems Sponsored Session Chair: Xiao Liu, University of Utah 1 - How Are You Feeling Today? Let Online Cancer Communities Help You Feel Better Anqi Xu, University of Utah, Salt Lake City, UT, United States, Paul Hu Cancer is devastating and has profound effects on patients’ well-beings, both physical and emotional. Cancer patients need strong social support in different aspects and their emotions can greatly benefit from effective support. Toward that, online communities are crucial because they can provide the needed support to cancer patients. This study analyzes patients’ different social support needs, then develops and empirically evaluates a deep learning-based, multi-label classification method to scrutinize their respect effects on a patent’s emotion change, with a focus on the fit between the patient’s particular need and the specific support by an online cancer community. Social bots hold immense power, as they are excellent broadcasters of good/bad messages in high velocity. Some bots are useful because they can assist other users in several ways, such as providing news article summaries. However, other bots act maliciously by spreading misinformation or exploiting users. As these bots become more prevalent on social media, platform owners must identify strategies to manage them. One key capability that platforms need is the ability to distinguish between beneficial and harmful social bots that exist within their platform. In this research, crowd reactions to known social bots are scrutinized through a deep learning framework to assess the intent of the bot. n TC67 West Bldg 105B 2 - Using Crowd Reactions to Detect Social BOT Intent Victor Benjamin, Arizona State University, Tempe, AZ, United States, Kunal Sahay, Raghu Santanam

n TC65 West Bldg 104B Data Science and Deep Learning II Sponsored: Data Mining Sponsored Session Chair: James Patrick Bailey, Georgia Institute of Technology, Atlanta, GA, 30309, United States 1 - Relaxed Wasserstein and Applications to Generative Adversarial Networks Tianyi Lin, University of California, Berkeley, 4174 Etcheverry Hall, 2521 Hearst Ave, Berkeley, CA, 94709, United States, Xin Guo, Johnny Hong, Nan Yang We propose a class of statistical divergences called \textit{Relaxed Wasserstein} (RW) divergence. We establish for RW divergence several properties: 1) RW divergence is dominated by Total Variation and Wasserstein-$L^2$ divergence; 2) RW divergence has continuity, differentiability and duality representation; 3) RW divergence has a non-asymptotic moment estimate and a concentration inequality. Experiments on the image generation demonstrate that RWGANs possess better convergence than the existing WGANs with competitive inception scores. To the best of our knowledge, this is the first to provide both the flexibility to design effective GANs and the possibility to study different losses. 2 - Probabilistic Mixture Models for Predicting Unbalanced Thunderstorm-induced Power Outages Elnaz Kabir, University of Michigan, 1205 Beal Ave., Ann Arbor, MI, 48109, United States, Seth Guikema, Steven Quirirng Severe storms have substantial impacts on power systems, posing risks and inconveniences due to power outages. Statistical prediction models have promise for supporting utility response, but power outage data are highly zero-inflated. In this study, we propose probabilistic mixture models for predicting the spatial distribution of power outages based on zero-inflated data. The result shows that these models offer strong predictions with zero-inflated data. 3 - Multiplicative Weights Update in Zero-sum Games James Patrick Bailey, Postdoctoral Researcher, Singapore University of Technology and Design, Singapore We study the classic setting where two agents compete in a zero-sum game by applying Multiplicative Weights Update (MWU). In a twist of the standard approach of Freund, we show the K-L divergence from an equilibrium increases and that strategies convergence to the boundary of the smallest face containing all equilibria. Our results come in stark contrast with the standard interpretation of the behavior of MWU in zero-sum games, which is typically referred to as “converging to equilibrium”. If equilibria are indeed predictive even for the benchmark class of zero-sum games, agents in practice deviate robustly from the axiomatic perspective of optimization driven dynamics captured by MWU. Joint Session QSR/Practice Curated: Prognostics and Health Management Sponsored: Quality, Statistics and Reliability Sponsored Session Chair: Tongdan Jin, Texas State University, San Marcos, TX, 78666, United States Co-Chair: Yan-Fu Li, Tsinghua University, Beijing, 100084, China 1 - Joint Optimization of Maintenance Planning and Workforce Routing for Networked Infrastructures Chi Zhang, Tsinghua University, Department of Industrial Engineering, Tsinghua University, Beijing, 100084, China, Chuanzhou Jia The economic development and social well-being of modern societies are highly dependent on networked infrastructures. Thus, it is necessary to timely and effectively maintain them to ensure the reliability of their continuous operation. However, their components are geographically distributed and the time required to transport between the components needs to be considered in order to make achievable maintenance plans. To address this problem, we transfer the time of each component being maintained into their sequence of being visited by each team of workforce, and propose a new approach to jointly optimize the maintenance planning and workforce routing for a networked infrastructure. n TC66 West Bldg 105A

331