Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

TB65

2 - Using Omnichannel Sales Data Analytics to Decide Between Store and Distribution Center Fulfillment Options Jingran Zhang, New Jersey Institute of Technology, University Height, Newark, NJ, 07102, United States, Jingran Zhang, Marshall University, Lewis College of Business, One John Marshall Drive, Corbly Hall 423, Huntington, WV, 25755, United States, Sanchoy Das A brick-and-mortar retailer can fulfill online customer orders in two ways (i) Buy Online Fulfill from Store (BOFS) - Picked from store inventory, and (ii) Fulfill from Distribution Center (FDC) - Picked from DC or warehouse inventory. The fulfillment decision is made in real time for each order, with the primary goal of maximizing the revenue value of the store inventory. Analysis of sales data in both online and store channels is used to forecast the value of the dispersed inventory, and then develop a prescriptive model for making a fulfillment decision. 3 - Google Tells What Happens When Shadow Economy Meets Bitcoin Zheshi Chen, First and Responding, Harbin Institute of Technology, In this study, we find striking positive correlations between Bitcoin’s key financial indicators and the level of shadow economy, based on the geographic distribution of interests on Google Trend keyword “Bitcoin”, along with the corresponding Shadow Economy Index data. Our results indicate that the shadow economy may be one of the key implicit factors driving Bitcoin’s prosperity. Our study also shows that Google Trend can be an effective detector of economic and social drivers of Bitcoin. 4 - Identifying Companies with Prospects of Adopting Blockchain Hamidreza Ahady Dolatsara, Auburn University, 227 Lowder Hall, This research identifies companies that may adopt blockchain technology based on their similarities to the companies that have already demonstrated an intention to use this technology. Long term financial performance of companies that intended to use blockchain technology is investigated for understanding the characteristics of such companies. Semi-supervised classification method is subsequently employed to identify similar companies. 5 - Smart Contract Execution – The (+-)-Biased Ballot Problem Lin Chen, University of Houston, 4800 Calhoun Road, Houston, TX, 77004, United States, Lei Xu, Zhimin Gao, Nolan Shah, Yang Lu, Weidong Shi Transaction system build on top of blockchain, especially smart contract, is becoming an important part of world economy. We consider a blockchain-based smart contract system and study how the behavior of users would affect the consensus. We connect this problem to the classical Ballot problem in combinatorics and establish a more general (+-)-Biased Ballot Problem to model it. We give an asymptotic analysis of this new model. 405 W. Glenn Ave., Auburn, AL, 36830, United States, Ashish Gupta, Alireza Farnoush, Gelareh Ahadi Dolatsara Chair: Mohammadreza Soltani, Iowa State University, 3201 Coover Hall, Ames, IA, 50011, United States 1 - The Role of Community-based Norms in Carbon Emissions: A Machine Learning Exploration Zhasmina Tacheva, SUNY at Buffalo, Buffalo, NY, 14210, United States, Anton Ivanov Through the use of advanced machine learning and text mining techniques, this study analyzes a massive body of unstructured Twitter data in order to uncover patterns of common needs and values across all 3,007 counties in the U.S. The detected personality characteristics are then used to predict household carbon emissions at the county level. Concrete policy implications and important insights about ways to reduce carbon emissions through modifying consumer behavior at the local level are made. 2 - Embedding Machine Learning Into SAP Hana Applications Sricharan Poundarikapuram, SAP America, 2996 N. 83rd Place, Scottsdale, AZ, 85251, United States, Doug Freud The next wave of computing builds upon the Tabulation era that started in the 1900’s and the Programmatic Computing era that started in the 1950’s. The new class of analytic solutions leverage data combined with ML and other technologies to transform and automate tasks typically performed by humans. The SAP point of view is that next generation applications need to evolve from simple rules based programs into solutions that are powered by ML. Instead of hard coding Management Building 612, West Dazhi Street, Harbin, Heilongjiang China, Harbin, 150001, China, Wenjun Sun, Qiang Ye n TB65 West Bldg 104B Data Science and Deep Learning I Sponsored: Data Mining Sponsored Session

rules into applications, next generation solutions will be a combination of rules and probabilities dynamically generated via ML. In this session we will cover an overview of ML powered application that we built in SAP HANA and Cloud Platform. 3 - Data Driven Portfolio Optimization Utilizing Machine Learning Meng-Chen Hsieh, Rider University, 2083 Lawrenceville Rd, Lawrence Township, NJ, 08648, United States In practice, data-driven optimal portfolio decisions are derived using the time series data of underlying asset returns.Such data-driven optimal portfolio tends to have inferior out-of-sample performance due to estimation errors of parameters.In the ‘big data’ era, correlations between asset returns and auxiliary variables are frequently observed. In this talk, several machine learning methods are applied to derive the optimal portfolio leveraging the association between the underlying asset returns and auxiliary variables. A comparison study on the out- of-sample performance of the optimal portfolio with and without utilizing machine learning methods is conducted. 4 - A DQN Method for Progressive Generation of Solutions of VRPTW Xiaodong Zhang, Artificial Intelligence Department, Zhejiang Cainiao Supply Chain Management Co., Ltd, Hang Zhou, China We present a ‘like-heuristic’ framework for solving Vehicle Routing Problem with Time Window using Deep Q-learning Network. In this approach, instead of defining a specific heuristic policy to search the huge solution space, we make decisions according to the probability distribution of all customers to make progressive generation, by observing the reward signals and following feasibility rules. 5 - Fast and Provable Algorithms for Learning Two-layer Polynomial Neural Networks Mohammadreza Soltani, Iowa State University, 3201 Coover Hall, Ames, IA, 50011, United States, Chinmay Hegde We study the problem of (provably) learning of a two-layer neural network with quadratic activations. We focus on the under-parametrized regime where the number of neurons in the hidden layer is smaller than the dimension of the input. Our main approach is to “lift the learning problem into a higher dimension, which enables us to borrow techniques from low-rank matrix estimation. Using this intuition, we propose three novel, non-convex algorithms. We support our algorithms with rigorous theoretical analysis, and show that all three enjoy a linear convergence, fast running time per iteration, and near-optimal sample complexity. We complement our theoretical results with some experiments. n TB66 West Bldg 105A Joint Session AI/Practice Curated: Social Media and Surveillance: Detection and Prediction Sponsored: Artificial Intelligence Sponsored Session Chair: Yongcheng Zhan, The University of Arizona, Tuscon, AZ 1 - Learning User Age via Locality and its Applications in Regulatory Science: A Deep Learning-Based Approach Zhipeng Chen, The University of Arizona, Tuscon, AZ, United States, Daniel Dajun Zeng The development of regulatory science requires researchers to utilize multiple data sources for assessing the safety, efficacy, quality and performance of all regulated products. The large quantity of social media data provides an opportunity to characterize public attitudes in real time. However, the lack of demographic data induces data biases and diminishes the utility of social media data. Our research mitigates this weakness by building a deep learning model to predict user age. The experiment shows the locality is a strong predicator and achieve a satisfiable prediction accuracy. Then we apply the prediction model to an e-cigarette dataset in an e-cigarette brand analysis. The analysis results indicate strong public health implications of demographics prediction. 2 - Illicit and Prescription Drug Street Name Detection using Social Media Yongcheng Zhan, The University of Arizona, Tucson, AZ, 85719, United States, Zhu Zhang, Scott J. Leischow, Daniel Dajun Zeng Recent years have observed the spread of America’s drug overdose epidemic. Much effort has been made in public health surveillance to realize the potential of early detection and targeted intervention. However, because of the ever-changing drug street names, it is increasingly difficult for researchers and public health practitioners to keep up with the terms. We proposed an innovative method to automatically detect illicit and prescription drug street names by using rich text information from social media. Our method showed its potential by enriching NIH opioid street name list with emerging terms.

298