Workshop on Data-Driven Decision Making

School of Operations Research and Information Engineering

Cornell University

October 14-15, 2016

Program Overview

Thursday, October 13, 20167:00 pm Welcome Reception, Ithaca Beer Co., 122 Ithaca Beer Drive, Ithaca, NY

Friday, October 14, 2016(All sessions are in Rhodes Hall 253 and all coffee breaks are in Rhodes Hall 258)

9:00 am Breakfast and Welcome Remarks by Professor David Shmoys, ORIE Ph.D. Lounge, Rhodes Hall 258

9:30 am Session I • A Miscellany (Chair: Shane Henderson)• Assortment Optimization Under Consider-then-Choose Choice Models,

Ali Aouad, MIT• Portfolio Liquidity Estimation and Optimal Execution,

Kai Yuan, Columbia University 10:30 am Coffee Break

11:00 am Session II • Pricing (Chair: Madeleine Udell)• Dynamic Pricing in Social Networks: The Word of Mouth Effect,

Amir Ajorlou, MIT• Markdown Pricing with Quality Perception and Consumer Optimism:

From Experiment to Theory, Rim Hariss, MIT• Dynamic Learning and Pricing with Online Product Reviews,

Dongwook Shin, Columbia University 12:30 pm Lunch, Rhodes Hall 258

2:00 pm Session III • Matching (Chair: Sid Banerjee)• Strategic Stable Marriage, James Bailey, Georgia Tech• Empty-car Routing in Ridesharing Systems, Anton Braverman, Cornell University • Matching while Learning, Vijay Kamble, Stanford University

3:30 pm Coffee Break

4:00 pm Session IV • Robust Optimization I (Chair: Damek Davis)• Using data-driven DRO to optimally choose regularization parameter

in machine learning, Karthyek Murthy, Columbia University• On deterministic reformulations of distributionally robust joint chance

constrained optimization problems, Weijun Xie, Georgia Tech

5:30 pm Poster Session • Appel Lobby, Herbert F. Johnson Museum of Art7:00 pm Dinner • Lynch Conference Room, Herbert F. Johnson Museum of Art

Saturday, October 15, 2016(All sessions are in Rhodes Hall 253 and all coffee breaks are in Rhodes Hall 258)

9:00 am Session V • Learning (Chair: Kris Iyer)• Learning Combinatorial Structures, Swati Gupta, MIT• Learning Preferences with Side-Information: Near Optimal Recovery

of Tensors from Noisy Observations, Andrew Li, MIT• Learning and Pricing using Bundles, Will Ma, MIT

10:30 am Coffee Break

11:00 am Session VI • Robust Optimization II (Chair: Yudong Chen)• Distributionally Robust Stochastic Optimization with Wasserstein

Distance, Rui Gao, Georgia Tech• Statistics of Robust Optimization, Hongseok Namkoong, Stanford University

12:00 pm Box Lunch, Rhodes Hall 258

Abstracts

Session I • A Miscellany (Chair: Shane Henderson) - 9:30-10:30 am

Assortment Optimization Under Consider-then-Choose Choice ModelsAli Aouad, MIT

Consider-then-choose models, borne out by empirical literature in marketing, explain that customers choose among alternatives in two phases, by first screening products to decide which alternatives to consider, before then ranking them. In this paper, we develop a unified algorithmic framework, based on dynamic programming, to study the computational tractability of assortment optimization under ranking- based choice models posited on consider-then-choose premises. Although ranking-based models often lead to computationally intractable assortment optimization problems, we show that for many practical and empirically vetted assumptions on how customers consider and then choose, the resulting optimization model is tractable. Our analysis permits expressing quantitative tradeoffs between modeling complexity and computational tractability. As a byproduct, our dynamic program unifies under a common algorithmic approach several specialized settings analyzed in previous literature. Empirically, the running time of our algorithm outperforms a state-of-the-art MIP solver on synthetic instances, in several parameter regimes of interest. Finally, we demonstrate the versatility and predictive power of our modeling approach, making use of a combination of synthetic and real-world datasets, where prediction errors are significantly reduced against common parametric choice models.

Bio: Hong is a Ph.D. candidate in the Management Science and Engineering department at Stanford University where he works under the joint supervision of John Duchi and Peter Glynn. His research interests lie in the span of statistics, optimization, simulation and machine learning. Before Stanford, Hong earned a B.S. degree in Industrial Engineering and Mathematics from KAIST.

...................................................................................................................................................................................

Portfolio Liquidity Estimation and Optimal ExecutionKai Yuan , Columbia University

Accurately estimating liquidity is an important ingredient in portfolio management. Traditionally, liquidity costs are estimated with single asset models. This ignores the fact that, fundamentally, liquidity is a portfolio problem since asset prices are correlated. We develop a model to estimate portfolio liquidity costs through a multi-dimensional generalization of the optimal execution model of Almgren and Chriss (1999). Our model allows for the trading of standardized liquid bundles of assets (e.g., ETFs or indices). We show that the hedging benefits of trading with many assets significantly reduces cost when liquidating a large position. In a “large universe” asymptotic limit, where the correlations across a large number of assets arise from relatively few underlying common factors, the liquidity cost of a portfolio is essentially driven by its idiosyncratic

risk. Moreover, the additional benefit of trading standardized bundles is roughly equivalent to increasing the liquidity of individual assets. Our method is tractable and can be easily calibrated from market data.

Bio: Kai Yuan is currently a fifth-year Ph.D. candidate in Decision, Risk and Operations of Columbia Business School. His research focuses on the applications of optimization and stochastic models to problems in financial engineering including algorithmic trading and optimal execution, risk management, and liquidity management. Kai received a B.Eng. in Electronic Engineering and a B.A. in Economics from Tsinghua University in 2012.

Session II • Pricing (Chair: Madeleine Udell) • 11:00 am-12:30 pm

Dynamic Pricing in Social Networks: The Word of Mouth EffectAmir Ajorlou, MIT

We study the problem of optimal dynamic pricing for a monopolist selling a product to consumers in a social network. In the proposed model, the only means of spread of information about the product is via word of mouth communication; consumers’ knowledge of the product is only through friends who already know about the product’s existence. Both buyers and non-buyers contribute to information diffusion while buyers are more likely to spread the news about the product. By analyzing the structure of the underlying endogenous dynamic process, we show that the optimal dynamic pricing policy for durable products with zero or negligible marginal cost drops the price to zero infinitely often. The firm uses free-offers to attract low-valuation agents and to get them more engaged in the spread. As a result, the firm can reach potential high-valuation consumers in parts of the network that would otherwise remain untouched without the price drops. We provide evidence for this behavior from the smartphone app market, where price histories indicate frequent zero-price sales. We demonstrate the validity of our results in face of forward-looking consumers and homophily in word of mouth engagement.

Bio: Amir Ajorlou is a postdoctoral research fellow at the Institute for Data, Systems and Society at MIT. He received his BS from Sharif University of Technology in Tehran, Iran, and his PhD in electrical and computer engineering from Concordia University in Montreal, Canada, in 2013. He has been the recipient of several prestigious awards, including two gold medals in the International Mathematical Olympiad (IMO), Concordia University Doctoral Prize in Engineering and Computer Science, Governor General of Canada Academic Gold Medal, and NSERC Postdoctoral Fellowship. His current research lies at the intersection of network and information economics, where he applies tools and techniques from optimization and game theory to decision making problems in social and economic networks.

...................................................................................................................................................................................

Markdown Pricing with Quality Perception and Consumer Optimism: From Experiment to TheoryRim Hariss, MIT

We integrate a controlled laboratory experiment and an analytical model to investigate how consumers’ price-based quality perception and optimism/pessimism about future markdown influence a retailer’s markdown pricing strategy. Consumers often perceive higher-priced products to have higher quality. Less is known on how quality perception is affected by simultaneously presenting multiple price-related information. Furthermore, it is unknown whether early consumers to the market versus late consumers use price information differently to form quality perception. Our experimental results demonstrate that early and late consumers form distinctive quality perception influenced by different sets of price information. We subsequently develop a consumer model that incorporates this consumers’ price-based quality perception experimental relationship and consumers’ potential gain/loss emotions due to pessimistic/optimistic expectations on the future markdown. We embed this consumer model into the retailer’s markdown optimization and examine the impact of these behavioral factors on the retailer’s optimal strategy. We show that the retailer would be better off if it could preannounce and commit to a markdown strategy to eliminate consumers’ optimism/pessimism. In addition, consumers’ price-based quality perception has a more significant impact on the retailer’s optimal strategy and revenue than optimism does. Ignoring such quality perception results in an average revenue loss of over 8\%.

Bio: Rim Hariss is currently a third-year Ph.D. doctorate candidate at the Operations Research Center at MIT. She received a B.S. in Applied Mathematics and a Master in Sciences in Applied Mathematics from École Polytechnique in France. Rim’s research focuses on investigating and incorporating the impact of consumer’s behavioral in revenue management, in particular on a firm’s operational decisions in a retail setting and in developing a data-driven approach for limited-time offers and rewards programs.

...................................................................................................................................................................................

Dynamic learning and pricing with online product reviewsDongwook Shin , Columbia University

We investigate how the presence of the product review system affects a dynamic-pricing monopolist who is operating without knowing the demand model. A salient feature of our problem is that the demand function evolves over time in conjunction with the dynamics of the review system. We find the optimal pricing policy in a closed-form using fluid, mean-field model, which is a good approximation when the sales volume is large. We first assume that sellers are relatively well-informed about the parameters of the demand function, in which case we show that a certain form of myopic policy works well. Then we consider a case with more significant uncertainty, where the myopic policy’s performance is strictly suboptimal, because the sellers need to implement price experimentation to counter the added uncertainty in the demand model.

Bio: Dongwook Shin is a Ph.D. candidate of Decision, Risk, and Operations in the Columbia Graduate School of Business. His research centers on pricing and revenue management with applications in e-commerce, machine learning and sequential decision making in operations research context, and sports analytics.

We make significant progress towards resolving this open problem and closing this gap. In particular, we prove that the optimality gap of the same constant-order policy actually converges exponentially fast to zero. We also derive simple and explicit bounds for the optimality gap, which make the result and methodology practical for realistic lead time values.  Session III • Matching (Chair: Sid Banerjee) • 2:00-3:30 pm

Strategic Stable MarriageJames Bailey, Georgia Tech

We study stable marriage where individuals strategically submit private preference information to a publicly known stable marriage algorithm. We prove that no stable marriage algorithm ensures actual stability at a Nash equilibrium when individuals are strategic. Thus the set of Nash equilibria provides no predictive value nor guidance for mechanism design. We propose the following new minimal dishonesty equilibrium refinement, supported by experimental economics results: an individual will not strategically submit preference list L if there exists a more honest L0 that yields as preferred an outcome. Then for all marriage algorithms satisfying monotonicity and IIA, every minimally dishonest equilibrium yields a sincerely stable marriage. This result supports the use of algorithms less biased than the (Gale-Shapley) man-optimal, which we prove yields the woman-optimal marriage in every minimally dishonest equilibrium. However, bias cannot be totally eliminated, in the sense that no monotonic IIA stable marriage algorithm is certain to yield the egalitarian-optimal marriage in a minimally dishonest equilibrium – thus answering an open question of Gusfield and Irving’s in the negative. Finally, we show that these results extend to student placement problems, where women are polyandrous and honest. but not to admissions problems, where women are both polyandrous and strategic.

Bio: James P. Bailey is a Ph.D. candidate at the Georgia Institute of Technology. He received a B.S. in industrial engineering, a B.S. in mathematics, and a M.S. in industrial engineering from Kansas State University. His current research involves the study of manipulation in mechanism design and path finding in tessellations of a space. More generally, his research interests include game theory, combinatorial optimization and graph theory. ...................................................................................................................................................................................

Empty-car Routing in Ridesharing SystemsAnton Braverman, Cornell University

We consider a queueing network model of a ridesharing system such as Lyft or Uber. Each time a car drops off a passenger at her/his destination, a routing decision needs to be made. Should the car stay and wait for the next customer at its current location, or should it drive empty to another part of town to meet demand there?

The way this decision is made greatly affects the supply of cars across a city; a bad routing policy can cause supply shortages in certain parts of the city. Using fluid-model analysis, we develop an asymptotically optimal centralized empty-car routing policy. We then evaluate its performance in finite systems using real-world data obtained from Didi Chuxing, China’s leading ridesharing company.

Bio: Anton Braverman is a fifth-year student at Cornell’s ORIE department working with Jim Dai. His main research is on the use of Stein’s method for steady-state diffusion approximations of stochastic systems. Apart from diffusion approximations and queueing theory, he is also interested in Markov decision processes and stochastic control. Anton is currently exploring application domains that include ridesharing networks such as Lyft or Uber, and is also interested in healthcare applications. ...................................................................................................................................................................................

Matching while LearningVijay Kamble, Stanford University

Online service platforms contend with two simultaneous challenges. On the one hand, supply must be matched with demand on an ongoing basis. On the other hand, new participants are constantly arriving to the platform, and the characteristics of these arrivals can often only be learned through match outcomes. Thus matching efficiently requires management of supply and demand while also learning about new participants.

We consider a benchmark model to study this challenge. In particular we consider a service system with heterogeneous workers and jobs; we assume job types are known, but that there is a fixed arrival rate of jobs of each type. On the other hand worker types are unknown and must be learned through matches.

The typical approach to addressing an exploration-exploitation tradeoff is the multi-armed bandit model. However, in our setting, capacity constraints play a first-order role in determining the learning goals of an optimal algorithm. Our main contribution is to demonstrate how the shadow prices from the optimal matching problem in the setting where all worker attributes are known at arrival can be utilized to determine these learning goals, leading to a policy that is asymptotically regret-optimal as the worker lifetimes in the market become longer. Informed by our theory, we propose several market-implementable heuristics that have demonstrably near-optimal performance in practical settings.

Bio: Vijay Kamble is currently a postdoctoral researcher in the Social Algorithms Lab at the Management Science and Engineering Dept. at Stanford University. He obtained his Ph.D. in Electrical Engineering and Computer Sciences from the University of California, Berkeley in December 2015. He is interested in designing algorithms and mechanisms for optimizing systems that leverage human expertise and aid decision-making, using tools and techniques from the areas of decision-making under uncertainty, applied probability, optimization and game theory. Session IV • Robust Optimization I (Chair: Damek Davis) • 4:00-5:00 pm

Using data-driven DRO to optimally choose regularization parameter in machine learningKarthyek Murthy, Columbia University

The objective of this talk is to introduce a novel, distributionally robust optimization (DRO) based inference procedure called RWPI (Robust Wasserstein Profile-based Inference). The proposed procedure recasts some of the popular regularization based machine learning algorithms (such as generalized Lasso, regularized logistic regression, etc.) as particular examples of a class of data-driven distributionally robust optimization problems based on Wasserstein distances. Most importantly, we shall see that an asymptotic analysis of a suitably defined profile function allows us to optimally select the regularization parameter for these machine learning algorithms without performing cross-validation.

This is based on joint work with Jose Blanchet and Yang Kang.

Bio: Karthyek Murthy is a post-doctoral research scientist in the Department of Industrial Engineering & Operations Research at Columbia University. He completed his Ph.D. at Tata Institute of Fundamental Research, Mumbai, where his Ph.D. work on rare events was awarded with best Ph.D. dissertation award for the year 2015. His research interests lie broadly in applied probability & stochastic processes, with special emphasis on models that arise in operations research, insurance and mathematical finance. Building on his Ph.D. work on rare events, he has been recently investigating stochastic modeling techniques that are robust to model risk, which has led to some interesting connections to statistics and applications to machine learning algorithms. His works have been recognized with IBM International Ph.D. fellowship and TCS Research fellowships. ...................................................................................................................................................................................

On deterministic reformulations of distributionally robust joint chance constrained optimization problemsWeijun Xie, Georgia Tech

A joint chance constrained optimization problem involves multiple uncertain constraints, i.e., constraints with stochastic parameters that are jointly required to be satisfied with probability exceeding a prespecified threshold. In a distributionally robust joint chance constrained optimization problem (DRCCP), the joint chance constraint is required to hold for all probability distributions of the stochastic parameters from a given ambiguity set. In this work, we consider DRCCP involving convex nonlinear uncertain constraints and an ambiguity set specified by convex moment constraints. We investigate deterministic reformulations of such problems and conditions under which such deterministic reformulations are convex. In particular we show that a DRCCP can be reformulated as a convex program if one the following conditions hold: (i) there is a single uncertain constraint, (ii) the ambiguity set is defined by a single moment constraint, (iii) the ambiguity set is defined by linear moment constraints, and (iv) the uncertain and moment constraints are positively homogeneous with respect to uncertain parameters. We further show that if the decision variables are binary and the uncertain constraints are linear then a DRCCP can be reformulated as a deterministic mixed integer convex program.

Bio: Weijun Xie is a Ph.D. candidate in the School of Industrial & Systems Engineering (ISyE) at Georgia Institute of Technology. He received his M.S. from the University of Illinois at Urbana-Champaign in 2013 and his B.Eng. from Tsinghua University, Beijing, China in 2010. His research interests are theory and applications of optimization under uncertainty, in particular, chance constrained stochastic optimization. He has also worked on stochastic network design problems.

Session V • Learning (Chair: Kris Iyer) • 9:00-10:30 am

Learning Combinatorial StructuresSwati Gupta, MIT

We consider two popular online learning algorithms to learn over combinatorial strategies like spanning trees, rankings, subset of experts: multiplicative weights update (MWU) algorithm and online mirror descent (OMD) algorithm. Although the regret of the MWU scales logarithmically in the number of strategies, every round of the MWU requires a large number of weight-updates. This especially becomes a problem when the number of pure strategies is exponential in the representation of the problem. We give a general recipe to simulate the MWU algorithm over vertices of combinatorial polytopes in R^n in time poly(n), whenever there exists an efficient generalized counting oracle (even if approximate) over the vertex set.

Next, the OMD algorithm requires the computation of a certain Bregman projection in each round whenever the learner’s strategies belong to a bounded set. We give a novel algorithm, Inc-Fix, to minimize separable convex functions over base polymatroids (e.g. spanning trees, permutations, k-subset of experts). For cardinality-based functions, our algorithm can be implemented in O(E^2) time (E is the size of the ground set) under certain Bregman divergences. We further show that Inc-

Fix can be terminated early to provide approximate solutions with a provable gap from optimality, when computing projections.

This is joint work with Michel Goemans and Patrick Jaillet.

Bio: Swati Gupta is a doctoral candidate at the Operations Research Center and LIDS, Massachusetts Institute of Technology, advised by Michel Goemans and Patrick Jaillet. She received her Bachelors and Masters in Computer Science from IIT Delhi. Her research interests are in optimization, learning, pricing and revenue management, machine learning, heuristics and approximation algorithms. Swati has been selected as a finalist for the INFORMS Service Science student paper competition 2016, and has won the Google Women in Engineering Award in 2011. ...................................................................................................................................................................................

Learning Preferences with Side-Information: Near Optimal Recovery of Tensors from Noisy ObservationsAndrew Li, MIT

A number of recent problems of great interest in e-commerce — such as demand learning with side information, context and location aware recommendations, personalized ‘tag’ learning, etc. — can be cast as large-scale problems of matrix recovery, with side information in the form of additional matrices of conforming dimension. Viewing the matrix we seek to recover and the side information we have as slices of a tensor, we consider the problem of Slice Recovery, which is to recover specific slices of ‘simple’ tensors from their noisy observations. We propose a definition of simplicity that is motivated by a compelling generative model and subsumes low-rank tensors for the most popular definitions of tensor rank. We provide an efficient algorithm for slice recovery that is practical for gigantic datasets and provides a significant performance improvement over state of the art incumbent approaches to tensor recovery. Further, we establish near-optimal recovery guarantees that in an important regime represent an order improvement over the best available results for this problem. Experiments on data from an online music streaming service demonstrate the performance and scalability of our approach.

Joint work with Vivek Farias.

Bio: Andrew Li is a Ph.D. candidate in the Operations Research Center at MIT, advised by Vivek Farias. Before joining MIT, he earned a B.S. in Operations Research from Columbia University. His primary research interest is in the design and analysis of data-driven solutions to contemporary, large-scale problems in operations management.

...................................................................................................................................................................................

Learning and Pricing using BundlesWill Ma MIT

A multi-product firm wants to learn its customers’ valuations for its products from sales data, and price its products given learned valuation distributions to maximize profit. We introduce new ways in which bundling helps with both problems. An example of bundling is Mixed Bundling (MB): the firm offers a couple of goods at individual prices, and the package containing all those goods at a discount. We show that sales data from MB contains richer information about latent valuations than individual sales data, allowing the firm to learn both mean valuations and price elasticities without price experimentation. We develop an iterative algorithm that reconstructs the distributions from noisy observations with high-probability convergence guarantees, assuming independence. For pricing, bundling is well-known to be effective in selling independent zero-cost goods, but not high-cost goods, since bundling encourages overconsumption. We solve this problem by allowing customers to return goods for their production cost, calling this mechanism Pure Bundling with Disposal for Cost (PBDC). We apply and improve techniques from mechanism design to prove a finite-item, distribution-free guarantee for PBDC. Finally, we show that MB and PBDC are complementary pricing schemes, and our learning results also apply to PBDC. Therefore, PBDC can be used for learning customer valuations while earning high profits. We verify this methodology in simulations.

This is joint work with David Simchi-Levi.

Bio: Will Ma is a fourth-year Ph.D. student in the MIT Operations Research Center, advised by David Simchi-Levi. Ma is interested in data-driven revenue management, helping businesses make adaptive operational decisions based on potentially censored data. He has been working on bundle pricing, learning from bundle sales data, and online personalized assortment planning. Ma spent 2013-2015 as a co-founder of Lunarch Studios, the start-up that makes the strategy game Prismata. During these years, he has also been a research/software engineering intern with the AdX group at Google New York, and a trading intern at Jane Street Capital. Ma completed his undergraduate degree in 2010 from the University of Waterloo, majoring in Pure Mathematics. During this time, he competed in many international poker tournaments and teaches the for-credit poker class at MIT.

Session VI • Robust Optimization II (Chair: Yudong Chen) • 11:00 am-12:00 pm

Distributionally Robust Stochastic Optimization with Wasserstein DistanceRui Gao, Georgia Tech

Optimization under uncertainty is often formulated as a stochastic optimization problem. In many settings, a “true” probability distribution may be unknown, or the notion of a true distribution may not even be applicable. We consider distributionally robust stochastic optimization (DRSO) approach, in which one hedges against all probability distributions that are within a chosen Wasserstein distance from a nominal distribution. Comparing to the popular phi-divergences, Wasserstein distance results in more reasonable worst-case distributions. We derive a dual reformulation of the DRSO problem and construct the worst-case distribution explicitly.Our contributions are three-fold. (i) We show that the worst-case distributions have a concise structure and a clear interpretation. By which we show data-driven DRSO problems can be approximated to any accuracy by robust optimization problems, and thereby many DRSO problems become tractable with tools from robust optimization. (ii) To the best of our knowledge, our proof of strong duality is the first constructive proof for DRSO problems, which is also useful in other contexts. (ii) Our strong duality result holds in a very general setting, and can be applied to infinite dimensional process control problems, for which the classic sample average approximation method fails to provide a meaningful solution.

Bio: Rui Gao is a fourth-year Ph.D. student in Operations Research at Georgia Tech, working with Professor Anton Kleywegt. His current research is focused on data-driven decision making under uncertainty, arising in the context of revenue management, systems design and machine learning. He received his Bachelor’s degree in Mathematics from Xi’an Jiaotong University in 2013.

...................................................................................................................................................................................

Statistical Theory of Robust OptimizationHongseok Namkoong, Stanford University

In this work, we study robust solution methods for stochastic optimization problems and show how robust solutions give optimality certificates. By developing an empirical likelihood theory for stochastic optimization, we give statistical and inferential guarantees for robust optimization. First, we show that robust solutions give a calibrated confidence bound on the optimal risk. Second, we prove that robust optimization is a variance regularization and give fast convergence guarantees. Finally, we develop an efficient solution method that can solve the minimax problem as fast as stochastic gradient descent up to log factors. Our experiments with real datasets show that robust optimization outperforms SAA on the harder instances.

This is joint work with John Duchi and Peter Glynn.

Bio: Hongseok Namkoong is a Ph.D. candidate in the Management Science and Engineering department at Stanford University where he works under the joint supervision of John Duchi and Peter Glynn. His research interests lie in the span of statistics, optimization, simulation and machine learning. Before Stanford, Hong earned a B.S. degree in Industrial Engineering and Mathematics from KAIST.

Name TitleChen, Bangrui Dueling Bandits with Dependent Arms

Choi, Michael Metropolis-Hastings reversiblizations of non-reversible Markov chains

Daw, Andrew Club Queue

Dong, James Multiproduct Robust Newsvendor Problem under a Global Budget of Uncertainty

Eckman, David Challenges in applying ranking and selection after search

Freund, Daniel Optimizing the Planning of Bike-Share Systems

Girard, Cory Dynamic control of a two-class queueing system with a waiting time constraint

Guo, Jiayi Comparing Nonsmooth Quasi-Newton Methods

Hu, Weici Bayes-Optimal Effort Allocation in Crowdsourcing: Bounds and Index Policies

Lo, Venus Assortment optimization under a synergistic version of the multinomial logit model

Ma, Sijia Estimation of Running Time of Ranking and Selection Procedures

Pallone, Steve Bayes-Optimal Entropy Minimization for Active Learning in Conjoint Analysis

Steele, Pat Problems in Air Ambulance Routing

Sumida, Mika Fare Locking : Options in Airline Revenue Management

Vera, Alberto Constrained shortest-paths in large-scale networks

Wang, Tiandong Multivariate Regular Variation of Discrete Mass Functions with Applications to Preferential Attachment Networks.

Wu, Jian The Parallel Knowledge Gradient Method for Batch Bayesian Optimization

Wylie, Calvin Designing a Minimization Algorithm that Identifies Smooth Substructure

Yang, Pu Mean Field Equilibria for Competitive Exploration in Resource Sharing Settings

ORIE Ph.D. Student Poster Session • Appel Lobby • Herbert F. Johnson Muesum of Art

Getting Around Campus

JohnsonMuseum

RhodesHall