Thesis (S.M.)--Massachusetts Institute of Technology, Sloan School of Management, Operations Research Center, 2004. / Includes bibliographical references (p. 117-121). / This electronic version was submitted by the student author. The certified thesis is available in the Institute Archives and Special Collections. / Stem (cell research is one of the most promising and cutting-edge fields i the miedical sciences. It is believed that this innovative research will lead to life-saving treatments in the coming years. As part of their work, stem cell researchers must first determine which of their stem cell colonies are of sufficiently high quality to be suitable for experimental studies and therapeutic treatments. Since colony texture is a major discriminating feature in determining quality. we introduce a non-invasive, semi-automated texture-based stem cell colony classification methodology to aid researchers in colony quality control. We first consider the general problem of textural image segmentation. In a new approach to this problem. we characterize image texture by the subband energies of the image's wavelet decomposition, and we employ a non-parametric support vector machine to perform the classification that yields the segmentation. We also adapt a parametric wavelet-based classifier that utilizes the Kullback-Leibler distance. We apply both methods to a set of benchmark textural images, report low segmentation error rates and comment on the applicability of and tradeoffs between the non-parametric and parametric segmentation methods. / (cont.) We then apply the two classifiers to the segmentation of stem cell colony images into regions of varying quality. This provides stem cell researchers with a rich set of descriptive graphical representations of their colonies to aid in quality control. From these graphical representatiolns, we extract colony-wise textural features to which we add colony-wise border features. Taken together, these features characterize overall colony quality. Using these features as inputs to a multiclass support vector machine, we successfully categorize full stem cell colonies into several quality categories. This methodology provides stem cell researchers with a novel, non-invasive quantitative quality control tool. / by Christopher G. Jeffreys. / S.M.

Thesis: Ph. D., Massachusetts Institute of Technology, Sloan School of Management, Operations Research Center, 2017. / This electronic version was submitted by the student author. The certified thesis is available in the Institute Archives and Special Collections. / Cataloged from student-submitted PDF version of thesis. / Includes bibliographical references (pages 137-143). / In this thesis, we study mixed-integer convex optimization, or mixed-integer convex programming (MICP), the class of optimization problems where one seeks to minimize a convex objective function subject to convex constraints and integrality restrictions on a subset of the variables. We focus on two broad and complementary questions on MICP. The first question we address is, "what are efficient methods for solving MICP problems?" The methodology we develop is based on outer approximation, which allows us, for example, to reduce MICP to a sequence of mixed-integer linear programming (MILP) problems. By viewing MICP from the conic perspective of modern convex optimization as defined by Ben-Tal and Nemirovski, we obtain significant computational advances over the state of the art, e.g., by automating extended formulations by using disciplined convex programming. We develop the first finite-time outer approximation methods for problems in general mixed-integer conic form (which includes mixed-integer second-order-cone programming and mixed-integer semidefinite programming) and implement them in an open-source solver, Pajarito, obtaining competitive performance with the state of the art. The second question we address is, "which nonconvex constraints can be modeled with MICP?" This question is important for understanding both the modeling power gained in generalizing from MILP to MICP and the potential applicability of MICP to nonconvex optimization problems that may not be naturally represented with integer variables. Among our contributions, we completely characterize the case where the number of integer assignments is bounded (e.g., mixed-binary), and to address the more general case we develop the concept of "rationally unbounded" convex sets. We show that under this natural restriction, the projections of MICP feasible sets are well behaved and can be completely characterized in some settings. / by Miles Lubin. / Ph. D.

Thesis: Ph. D., Massachusetts Institute of Technology, Sloan School of Management, Operations Research Center, 2015. / This electronic version was submitted by the student author. The certified thesis is available in the Institute Archives and Special Collections. / Cataloged from student-submitted PDF version of thesis. / Includes bibliographical references (pages 147-153). / Online markets are becoming increasingly important in today's world as more people gain access to the internet. Furthermore, the explosion of data that is collected via these online markets provides us with new opportunities to use analytics techniques to design markets and optimize tactical decisions. In this thesis, we focus on two types of online markets -- peer-to-peer networks and online retail markets -- to show how using analytics can make a valuable impact. We first study scrip systems which provide a non-monetary trade economy for exchange of resources; their most common application is in governing online peer-to-peer networks. We model a scrip system as a stochastic game and study system design issues on selection rules to match trade partners over time. We show the optimality of one particular rule in terms of maximizing social welfare for a given scrip system that guarantees players' incentives to participate, and we investigate the optimal number of scrips to issue under this rule. In the second part, we partner with Rue La La, an online retailer in the online flash sales industry where they offer extremely limited-time discounts on designer apparel and accessories. One of Rue La La's main challenges is pricing and predicting demand for products that it has never sold before. To tackle this challenge, we use machine learning techniques to predict demand of new products and develop an algorithm to efficiently solve the subsequent multi-product price optimization. We then create and implement this algorithm into a pricing decision support tool for Rue La La's daily use. We conduct a controlled field experiment which estimates an increase in revenue of the test group by approximately 10%. Finally, we extend our work with Rue La La to address a more dynamic setting where a retailer may choose to change the price of a product throughout the course of the selling season. We have developed an algorithm that extends the well-known multi-armed bandit algorithm called Thompson Sampling to consider a retailer's limited inventory constraints. Our algorithm has promising numerical performance results when compared to other algorithms developed for the same setting. / by Kris Johnson. / Ph. D.

Includes bibliographical references (leaves 111-118). / Thesis (Ph. D.)--Massachusetts Institute of Technology, Sloan School of Management, Operations Research Center, 2008. / (cont.) However, these randomized algorithms can never provide proven upper or lower bounds on the number of objects they are counting, but can only give probabilistic estimates. We propose a set of deterministic algorithms for counting such objects for three classes of counting problems. They are interesting both because they give an alternative approach to solving these problems, and because unlike MCMC algorithms, they provide provable bounds on the number of objects. The algorithms we propose are for special cases of counting the number of matchings, colorings, or perfect matchings (permanent), of a graph. / Multiclass queueing networks are used to model manufacturing, computer, supply chain, and other systems. Questions of performance and stability arise in these systems. There is a body of research on determining stability of a given queueing system, which contains algorithms for determining stability of queueing networks in some special cases, such as the case where there are only two stations. Yet previous attempts to find a general characterization of stability of queueing networks have not been successful.In the first part of the thesis, we contribute to the understanding of why such a general characterization could not be found. We prove that even under a relatively simple class of static buffer priority scheduling policies, stability of deterministic multiclass queueing network is, in general, an undecidable problem. Thus, there does not exist an algorithm for determining stability of queueing networks, even under those relatively simple assumptions. This explains why such an algorithm, despite significant efforts, has not been found to date. In the second part of the thesis, we address the problem of finding algorithms for approximately solving combinatorial graph counting problems. Counting problems are a wide and well studied class of algorithmic problems, that deal with counting certain objects, such as the number of independent sets, or matchings, or colorings, in a graph. The problems we address are known to be #P-hard, which implies that, unless P = #P, they can not be solved exactly in polynomial time. It is known that randomized approximation algorithms based on Monte Carlo Markov Chains (MCMC) solve these problems approximately, in polynomial time. / by Dmitriy A. Katz-Rogozhnikov. / Ph.D.

Thesis: Ph. D., Massachusetts Institute of Technology, Sloan School of Management, Operations Research Center, 2016. / This electronic version was submitted by the student author. The certified thesis is available in the Institute Archives and Special Collections. / Cataloged from student-submitted PDF version of thesis. / Includes bibliographical references (pages 204-209). / Modern business decisions exceed human decision making ability: often, they are of a large scale, their outcomes are uncertain, and they are made in multiple stages. At the same time, firms have increasing access to data and models. Faced with such complex decisions and increasing access to data and models, how do we transform data and models into effective decisions? In this thesis, we address this question in the context of four important problems: the dynamic control of large-scale stochastic systems, the design of product lines under uncertainty, the selection of an assortment from historical transaction data and the design of a personalized assortment policy from data. In the first chapter, we propose a new solution method for a general class of Markov decision processes (MDPs) called decomposable MDPs. We propose a novel linear optimization formulation that exploits the decomposable nature of the problem data to obtain a heuristic for the true problem. We show that the formulation is theoretically stronger than alternative proposals and provide numerical evidence for its strength in multi-armed bandit problems. In the second chapter, we consider to how to make strategic product line decisions under uncertainty in the underlying choice model. We propose a method based on robust optimization for addressing both parameter uncertainty and structural uncertainty. We show using a real conjoint data set the benefits of our approach over the traditional approach that assumes both the model structure and the model parameters are known precisely. In the third chapter, we propose a new two-step method for transforming limited customer transaction data into effective assortment decisions. The approach involves estimating a ranking-based choice model by solving a large-scale linear optimization problem, and solving a mixed-integer optimization problem to obtain a decision. Using synthetic data, we show that the approach is scalable, leads to accurate predictions and effective decisions that outperform alternative parametric and non-parametric approaches. In the last chapter, we consider how to leverage auxiliary customer data to make personalized assortment decisions. We develop a simple method based on recursive partitioning that segments customers using their attributes and show that it improves on a "uniform" approach that ignores auxiliary customer information. / by Velibor V. Mišić. / Ph. D.

Thesis (Ph. D.)--Massachusetts Institute of Technology, Sloan School of Management, Operations Research Center, 2008. / Includes bibliographical references (p. 125-129). / Annual influenza outbreaks incur great expenses in both human and monetary terms, and billions of dollars are being allocated for influenza pandemic preparedness in an attempt to avert even greater potential losses. Vaccination is a primary weapon for fighting influenza outbreaks. The influenza vaccine supply chain has characteristics that resemble the Newsvendor problem, but possesses several characteristics that distinguish it from many other supply chains. Differences include a nonlinear value of sales (caused by the nonlinear health benefits of vaccination that are due to infection dynamics) and vaccine production yield issues. In this thesis we present two models in the interface of operations and supply chain management and public health policy. In the first model, we focus on a supply chain with a government and a manufacturer. We show that production risks, taken currently by the vaccine manufacturer, lead to an insufficient supply of vaccine. Several supply contracts that coordinate buyer (governmental public health service) and supplier (vaccine manufacturer) incentives in many other industrial supply chains can not fully coordinate the influenza vaccine supply chain. We design a variant of the cost sharing contract and show that it provides incentives to both parties so that the supply chain achieves global optimization and hence improves the supply of vaccines. In the second mode, we consider the influenza vaccine supply chain with multiple countries. / (cont.) Each government purchases and administers vaccines in order to achieve an efficient cost-benefit tradeoff. Typically different countries have different economics sensitivities to public outcomes of infection and vaccination. It turns out that the initiating country, while having a significant role in the spread of the disease, does not receive enough vaccine stockpiles. Our model indicates that lack of coordination results in vaccine shortfalls in the most needed countries and vaccine excess in the regions where are not as effective, if the governments in the model act rationally. We show the role of contracts to modify monetary flows that purchase vaccination programs, and therefore modify infectious disease flows. / by Hamed Mamani. / Ph.D.

Thesis (Ph. D.)--Massachusetts Institute of Technology, Sloan School of Management, Operations Research Center, 2003. / Includes bibliographical references (p. 127-130). / Traditional inventory models focus on effective replenishment strategies and typically assume that a commodity's price is exogenously determined. In recent years, however, a number of industries have used innovative pricing strategies to manage their inventory effectively. These developments call for models that integrate inventory control and pricing strategies. Such models are clearly important not only in the retail industry, where price-dependent demand plays an important role, but also in manufacturing environments in which production/distribution decisions can be complemented with pricing strategies to improve the firm's bottom line. To date, the literature has confined itself mainly to models with variable ordering costs but no fixed costs. Extending some of these models to include a fixed cost component is the main focus of this thesis. In this thesis, we start by analyzing a single product, periodic review joint inventory control and pricing model, and characterizing the structure of the optimal policy under various conditions. Specifically, for the finite horizon periodic review case, we show, by employing the classical k-convexity concept, that a simple policy, called (s, S, p), is optimal when the demand functions are additive. For the model with more general demand functions, we show that an (s, S, p) policy is not necessarily optimal. We introduce a new concept, the symmetric k-convex functions, and apply it to provide a characterization of the optimal policy. Surprisingly, in the infinite horizon periodic review case, the concept of symmetric k-convex functions allows us to show that a stationary (s, S, p) policy is optimal for both discounted and average profit models even for general demand functions. / (cont.) Our approach developed for the infinite horizon periodic review joint inventory control and pricing problem is then extended to a corresponding continuous review model. In this case, we prove that a stationary (s, S, p) policy is optimal under fairly general assumptions. Finally, the symmetric k-convexity concept developed in this thesis is employed to characterize the optimal policy for the stochastic cash balance problem. / by Xin Chen. / Ph.D.

Thesis (S.M.)--Massachusetts Institute of Technology, Sloan School of Management, Operations Research Center, 2010. / Cataloged from PDF version of thesis. / Includes bibliographical references (p. 163-166). / Data mining is a versatile and expanding field of study. We show the applications and uses of a variety of techniques in two very different realms: Emergency department (ED) length of stay prediction and visual analytics. For the ED, we investigate three data mining techniques to predict a patient's length of stay based solely on the information available at the patient's arrival. We achieve good predictive power using Decision Tree Analysis. Our results show that by using main characteristics about the patient, such as chief complaint, age, time of day of the arrival, and the condition of the ED, we can predict overall patient length of stay to specific hourly ranges with an accuracy of 80%. For visual analytics, we demonstrate how to mathematically determine the optimal number of clusters for a geospatial dataset containing both numeric and categorical data and then how to compare each cluster to the entire dataset as well as consider pairwise differences. We then incorporate our analytical methodology in visual display. Our results show that we can quickly and effectively measure differences between clusters and we can accurately find the optimal number of clusters in non-noisy datasets. / by Ashley M. Snyder. / S.M.

Thesis (Ph. D.)--Massachusetts Institute of Technology, Sloan School of Management, Operations Research Center, 2007. / Includes bibliographical references (p. 139-150). / Investors commonly use stopping rules to help them get in and out of their investment positions. Despite their widespread use and support from behavioral finance, there has been little discussion of their impact on portfolio performance in classic portfolio choice theory. In this thesis, I remedy this situation by discussing the performance impact of stopping rules, highlighting the stop-loss rule. Stop-loss rules-predetermined policies that reduce a portfolio's exposure after reaching a certain threshold of cumulative losses-are commonly used by retail and institutional investors to manage the risks of their investments, but have also been viewed with some skepticism by critics who question their efficacy. I develop a simple framework for measuring the impact of stop-loss rules on the expected return and volatility of an arbitrary portfolio strategy, and derive conditions under which stop-loss rules add or subtract value to that portfolio strategy. I show that under the Random Walk Hypothesis, simple 0/1 stop-loss rules always decrease a strategy's expected return, but in the presence of momentum, stop-loss rules can add value. To illustrate the practical relevance of this framework, / (cont.) I provide an empirical analysis of a stop-loss policy applied to a buy-and-hold strategy in U.S. equities, where the stop-loss asset is U.S. long-term government bonds. Using monthly returns data from January 1950 to December 2004, I find that certain stop-loss rules add 50 to 100 basis points per month to the buy-and-hold portfolio during stop-out periods. By computing performance measures for several price processes, including a new regime-switching model that implies periodic "flights-to-quality," I provide a possible explanation for our empirical results and connections to the behavioral finance literature. Consistent with the traditional investor's problem, I discuss a generalization of this approach to general stopping rules, which are superimposed on arbitrary portfolio strategies. I define a stopping utility premium and discuss how uncertainty about the true stochastic process can explain a potential value added or value lost by the use of stopping rules in practice. / by Kathryn Margaret Kaminski. / Ph.D.

Thesis (Ph. D.)--Massachusetts Institute of Technology, Sloan School of Management, Operations Research Center, 2012. / Cataloged from PDF version of thesis. / Includes bibliographical references (p. 169-174). / This thesis studies the price of anarchy in supply chains, congested systems and joint ventures. It consists of three main parts. In the first part, we investigate the impact of imperfect competition with nonlinear demand. We focus on a distribution channel with a single supplier and multiple downstream retailers. To evaluate the performance, we consider several metrics, including market penetration, total profit, social welfare and rent extraction. We quantify the performance with tight upper and lower bounds. We show that with substitutes, while competition improves the efficiency of a decentralized supply chain, the asymmetry among the retailers deteriorates the performance. The reverse happens when retailers carry complements. We also show that efficiency of a supply chain with concave (convex) demand is higher (lower) than that with affine demand. The second part of the thesis studies the impact of congestion in an oligopoly by incorporating convex costs. Costs could be fully self-contained or have a spillover component, which depends on others' output. We show that when costs are fully self-contained, the welfare loss in an oligopoly is at most 25% of the social optimum, even in the presence of highly convex costs. With spillover cost, the performance of an oligopoly depends on the relative magnitude of spillover cost to the marginal benefit to consumers. In particular, when spillover cost outweighs the marginal benefit, the welfare loss could be arbitrarily bad. The third part of the thesis focuses on capacity planning with resource pooling in joint ventures under demand uncertainties. We distinguish heterogeneous and homogeneous resource pooling. When resources are heterogeneous, the effective capacity in a joint venture is constrained by the minimum individual contribution. We show that there exists a unique constant marginal revenue sharing scheme which induces the same outcome in a Nash equilibrium, Nash Bargaining and the system optimum. The optimal scheme rewards every participant proportionally with respect to his marginal cost. When resources are homogeneous, we show that the revenue sharing ratio should be inversely proportional to a participant's marginal cost. / by Wei Sun. / Ph.D.