Informs Annual Meeting 2017

MC31

INFORMS Houston – 2017

MC31

4 - Social Learning Strategies for Matters of Taste Pantelis Pipergias-Analytis, Cornell University, Ithaca, NY, United States, pantelispa@gmail.com, Daniel Barkoczi, Stefan Herzog Most choices people make are about “matters of taste”, where no universal, objective truth about the options’ quality exists. Still, people can learn from the experiences of individuals who have already evaluated available options. We mapped recommender system algorithms to social learning strategies for “matters of taste”. Using computer simulations, we studied how people can leverage the opinions of (similar) others to make better decisions. We show that experienced individuals can benefit from relying more on the opinions of seemingly similar people, whereas inexperienced individuals are better off picking the mainstream option despite inter-individual differences in taste.

351A INFORMS Internal Activities & 2018 INFORMS International Conference Overview Panel Session 1 - Moderator Grace Lin, Institute for Information Industry, 1F, No. 133, Sec. 4, Minsheng E. Rd., Taipei, Taiwan, gracelin.ny@gmail.com We will present our effort in integrating big data information including market trend, industry talent needs, and recruiting website information with the institution internal information to guide university course and program design and to support students’ Major selection. The corresponding big data lake, natural language processing and multi-source algorithms will also be discussed. 2 - Panel: INFORMS Internal Activities & 2018 INFORMS International Conference Overview Grace Lin, Institute for Information Industry, 1F, No. 133, Sec. 4, Minsheng E. Rd., Taipei, Taiwan, gracelin.ny@gmail.com 3 - Panelist Kathryn Stecke, University of Texas at Dallas, Richardson, TX, United States, kstecke@utdallas.edu 4 - Panelist Mary Magrogan, INFORMS, Catonsville, MD, United States, mary.magrogan@informs.org 351B Simplicity Versus Complexity in Forecasting and Prediction Invited: Forecasting and Prediction Invited Session Chair: Nathaniel Phillips, University of Basel, Basel, 4001, Switzerland, nathaniel.phillips@unibas.ch 1 - FFTrees: How to Easily Create and Visualise Fast-and-frugal Decision Trees to Make Accurate Decisions Based on Minimal Information Nathaniel Phillips, Basel, 4053, Switzerland, Nathaniel.D. Phillips.is@gmail.com, Hansjoerg Neth, Wolfgang Gaissmaier, Jan Woike Fast and frugal decision trees (FFTs) are simple decision algorithms that allow people to make efficient and effective decisions based on limited information. We introduce an R package called FFTrees that allows anyone to create, visualize, and use FFTs from data with minimal programming. We explain how FFTs work and provide a 5 step tutorial for using the FFTrees package to create them. We then present a simulation across a 10 datasets that show how FFTs created by FFTrees can predict data as well as complex machine learning algorithms while remaining simple enough for anyone to understand and use. 2 - Recency: Predicting with Smart Data Florian Artinger, Max Planck Institute for Human Development, Berlin, Germany, artinger@mpib-berlin.mpg.de Since the early 1910s, managers have been using a simple recency-based decision strategy, the hiatus heuristic, to identify valuable customers. This study analyzes the role of recency using a library of 60 data sets from business and other areas including weather, sports, and medicine. We find that the hiatus heuristic can perform on par with complex algorithms such as from machine learning and taking into account further variables apart from recency does not improve performance. We show that the results are not so much driven by limited sample size than by the dominant role that recency plays in most of these environments. 3 - Simple Rules for Complex Decisions Daniel G.Goldstein, Microsoft Research, New York, NY, United States, dan@dangoldstein.com, Jongbin Jung, Connor Concannon, Ravi Shroff, Sharad Goel We present a new method, select-regress-and-round, for constructing simple rules that perform well for complex decisions. These rules take the form of a weighted checklist and can be applied mentally. Importantly, our method for creating these rules is itself simple, and can be carried out by practitioners with basic statistics knowledge. Across 22 domain, including a key example of bail decision making, we find that these simple rules are nearly as good as state of the art models trained on all available features. MC32

MC33

351C AAS Special Presentation Sponsored: Aviation Applications Sponsored Session

Chair: Virginie Lurkin, Ecole Polytechnique Fédérale de Lausanne, Route Cantonale, Lausanne, 1015, Switzerland, vlurkin@ulg.ac.be 1 - Probabilistic Decision Making and Airline Operations: Why it Pays to Play Games of Chance Tim Niznik, Director, Operations Systems and Decision Support, American Airlines Decisions made in the absence of probability-based analytics will not just usually be wrong, they will consistently be wrong, especially when those decisions involve weather and ATC. One of the objectives of the Decision Support Vision for the Operations Control Center at American Airlines is to improve how we respond to weather events, specifically in terms of the determining the timing, level of risk, and specific actions to take. We need to move away from deterministic decision making that is primarily ad-hoc and experience-based to a more data-driven process that is built on predictive analytics and accounts for uncertainty. This talk looks at some examples where probabilistic decision making could be applied, some examples in the industry of where these techniques are already being applied, and finally, our current efforts to develop this capability internally. 351D Applications of Cyber Defense Sponsored: Military Applications Sponsored Session Chair: Natalie M Scala, Towson University, Towson, MD, 21252, United States, natalie.scala@gmail.com 1 - Reinforcement Learning for Autonomous Cyber Defense Ahmad Ridley, Department of Defense, Washington, DC, United States, adridle@tycho.ncsc.mil Our research goal is to build an autonomous cyber defense system that will make an enterprise network, and its associated missions/services, more resilient to cyber-attack. Such a cyber defense system should make decisions and implement appropriate responses to anticipate, withstand, recover from and/or evolve in the presence of cyber-attacks. Since attack methods are constantly evolving, this defense system must be adaptive, and also operate in real-time and at network scale. Therefore, in our research, we apply reinforcement learning to train the system to perform sequential decision making under uncertainty to defend the network. 2 - Performance Benchmarking of Cyber Streaming Analytics Caleb Baechtold, Janine Bainger, BAH, Baltimore, MD, baechtold_caleb@bah.com This research effort explores streaming platform behaviors to build models predicting performance metrics of concern as they relate to combinations of system environments and analytic configurations. The research focuses on WaterSlide, an open-source streaming analytic technology commonly used to perform Cyber operations by processing metadata using its event-at-a-time architecture. MC34

214