Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

TB74

3 - Asymptotically Optimal Exact Solution of Sparse Linear Systems via Left-Looking Roundoff-Error-Free LU Factorization Christopher Lourenco, Texas A&M University, 999 West Villa Maria Road, Apt 607, Bryan, TX, 77801, United States LU factorizations are the key tool used to solve sparse systems of linear equations (SLEs) arising in linear/integer programming. However, roundoff errors, accrued while solving SLEs, may lead solvers to claim suboptimal bases as optimal, feasible problems as infeasible, or vice versa. To address this, we develop an exact LU factorization with 3 key properties: 1) only integral operations, 2) polynomially bounded entry size in L and U, and 3) asymptotically optimal computational complexity. Also we computationally show that the new LU factorization outclasses a modern rational solver for SLEs. n TB72 West Bldg 211A Security Informatics General Session Chair: Victor Benjamin, University of Arizona, Tempe, AZ, United States 1 - Detecting Cyber Threats in Non-english Dark Net Markets: A Cross-lingual Deep Learning Approach Mohammadreza Ebrahimi, University of Arizona, Tucson, AZ, 85719, United States, Hsinchu Chen Dark Net Markets (DNMs) provide hackers with highly-specialized tools that may not be found in other platforms in hacker community. While text classification techniques have been used for cyber threat detection in English DNMs, the task is hindered in non-English platforms due to the language barrier and lack of ground-truth data. Current methods use machine translation with monolingual models to address these challenges. However, the translation errors can deteriorate the classification results. We show that a cross-lingual model that uses two languages, significantly outperforms a monolingual model learned on machine translated data for detecting cyber threats in non-English DNMs. 2 - Avatar Image Role in Developing Trust in an Intelligent Cybersecurity Agent Troy Adams, University of Arizona, Tempe, AZ, United States, Gondy Leroy Intelligent agents are used in a variety of applications and have made their appearance in modern networking and security technologies. In cybersecurity, these agents provide alerts and solutions for mitigating threats to a network, as well as training. This study aims to understand the development of trust in an intelligent cybersecurity agent (ICA), based on the image displaying its avatar. Results indicate that an avatar image had an impact on trust development for an ICA, especially for women. This research adds to IS literature by highlighting the role appearance has in improving HCI. Dark Net Markets (DNMs) provide hackers with highly-specialized tools that may not be found in other platforms in hacker community. While text classification techniques have been used for cyber threat detection in non-English DNMs, the task is hindered in non-English platforms due to the language barrier and lack of ground-truth data. Current methods use machine translation with monolingual models to address these challenges. However, the translation errors can deteriorate the classification results. We show that a cross-lingual model that uses two languages, significantly outperforms a monolingual model learned on machine translated data for detecting cyber threats in non-English DNMs. n TB73 West Bldg 211B JFIG Panel Discussion: Becoming a More Productive Writer Sponsored: Junior Faculty JFIG Sponsored Session Chair: Chrysafis Vogiatzis, North Carolina A&T State University, Greensboro, NC, 27411, United States Co-Chair: Gokce Palak, Shenandoah University,Winchester, VA, 22601, United States 1 - JFIG Panel Discussion: Becoming a More Productive Writer Chrysafis Vogiatzis, North Carolina A&T State University, 1601 East Market Street, McNair 405, Greensboro, NC, 27411, United States Panelists will share their experiences, and provide insights on how to become a more productive writer. The goal of the session is to introduce tips and pitfalls for 3 - Deep Learning for Text-based Social BOT Detection Victor Benjamin, Arizona State University, Tempe, AZ, United States, Raghu Santanam

new assistant professors (and other researchers in the beginning of their careers) when it comes down to academic writing, broadly defined (research publications, grant proposal writing, outreach and general population). Panelists Laura Albert, University of Wisconsin-Madison, Industrial & Systems Engineering, 1513 University Avenue, Madison, WI, 53706, United States Halit Uster, Southern Methodist University, Lyle School of Eng., Dept. EMIS, Dallas, TX, 75275-0123, United States Lawrence V. Snyder, Lehigh University, Mohler Lab 200 West Packer Avenue, Bethlehem, PA, 18015-1582, United States Elise Miller-Hooks, George Mason University, 208 Rosalie Cove Ct, Silver Spring, MD, 20905, United States n TB74 West Bldg 212A Joint Session MCDM/Practice Curated: Multi-Objective Simulation Optimization and its Applications Sponsored: Multiple Criteria Decision Making Sponsored Session Chair: Susan R. Hunter, Purdue University, West Lafayette, IN, 47907- 2023, United States 1 - Digital Twin for the Smart System Loo Hay Lee, National University of Singapore, 10 Kent Ridge Cresent, Industrial and Systems Engineering, Singapore, 119260, Singapore In this talk, we will share the digital twin system we have built for the port system and warehouse system, and will discuss how the simulation optimization technique is able to help to design the system. 2 - Indifference Zones in Multi-Objective Ranking & Selection Juergen Branke, University of Warwick, UK, Warwick Business School, Orms Group, Coventry, CV4 7AL, United Kingdom, Wen Zhang Multi-objective ranking and selection so far has mostly focused on probability of correct selection as criterion. However, this creates problems if there are two very similar solutions, as many samples would be needed to correctly classify them. Thus, people have proposed to allow for an indifference zone, describing a decision maker’s indifference between very similar solutions. We demonstrate that a previously proposed concept of indifference zones in multi-objective problems has some drawbacks, and propose an alternative definition. We also show how this can be used as part of a myopic ranking and selection technique. 3 - Bi-objective Simulation Optimization on Integer Lattices using the Epsilon-constraint Method in a Retrospective Approximation Framework Susan R. Hunter, Purdue University, School of Industrial Engineering, Grissom Hall, 315 N. Grant Street, West Lafayette, IN, 47907-2023, United States, Kyle Cooper, Kyle Cooper, Kalyani S. Nagaraj We propose the Retrospective Partitioned Epsilon-constraint with Relaxed Local Enumeration (R-PeRLE) algorithm to solve the bi-objective simulation optimization problem on integer lattices. R-PeRLE uses a retrospective approximation (RA) framework to repeatedly call the PeRLE algorithm at a sequence of increasing sample sizes. Within an RA iteration, the PeRLE algorithm uses the e-constraint method to add new sample-path local efficient points to the estimated local efficient set (LES). Then, it certifies the estimated LES is approximately a sample-path LES. As the number of RA iterations increases, R- PeRLE provably converges to a LES wp1. 4 - Multiple Objective Probabilistic Branch and Bound for Multi-fidelity Simulation Optimization Hao Huang, Yuan Ze University, IEM Department, Yuan Ze University, 135 Yuan-Tung Road, Chung-Li,, Taoyuan, 32003, Taiwan, Multi-fidelity framework involves simulation optimization problems involving multiple models with different fidelities, such as an accurate but costly simulation and a quick approximation. A multiple objective simulation algorithm is proposed in this study for the multi-fidelity framework, named Multi-Fidelity and Multi- Objective Probabilistic Branch and Bound (MFOPBnB). MFOPBnB incorporates Multiple Objective Probabilistic Branch and Bound (MOPBnB), a partition-based simulation optimization algorithm approximates the Pareto optimal set. Using information from low fidelity model, MFOPBnB approximates the a Pareto optimal set with statistical quality derived.

301