Search

This paper studies a classic maximum entropy sampling problem (MESP), which aims to select the most informative principal submatrix of a prespecified size from a covariance matrix. MESP is widely applied to many areas, including healthcare, power systems, manufacturing, and data science. By investigating its Lagrangian dual and primal characterization, we derive a novel convex integer program for MESP and show that its continuous relaxation yields a near-optimal solution. The results motivate us to study efficient approximation algorithms and develop their approximation bounds for MESP, which improves the best known one in the literature. This paper studies a classic maximum entropy sampling problem (MESP), which aims to select the most informative principal submatrix of a prespecified size from a covariance matrix. By investigating its Lagrangian dual and primal characterization, we derive a novel convex integer program for MESP and show that its continuous relaxation yields a near-optimal solution. The results motivate us to develop a sampling algorithm and derive its approximation bound for MESP, which improves the best known bound in literature. We then provide an efficient deterministic implementation of the sampling algorithm with the same approximation bound. Besides, we investigate the widely used local search algorithm and prove its first known approximation bound for MESP. The proof techniques further inspire for us an efficient implementation of the local search algorithm. Our numerical experiments demonstrate that these approximation algorithms can efficiently solve medium-size and large-scale instances to near optimality. Finally, we extend the analyses to the A-optimal MESP, for which the objective is to minimize the trace of the inverse of the selected principal submatrix. Funding: This work was supported by the National Science Foundation Division of Information and Intelligent Systems [Grant 2246417] and Division of Civil, Mechanical and Manufacturing Innovation [Grant 2246414]. Supplemental Material: The e-companion is available at https://doi.org/10.1287/opre.2023.2488. [ABSTRACT FROM AUTHOR]

Solving Rank Constrained Least Squares via Recursive Importance Sketching In statistics and machine learning, we sometimes run into the rank-constrained least squares problems, for which we need to find the best low-rank fit between sets of data, such as trying to figure out what factors are affecting the data, filling in missing information, or finding connections between different sets of data. This paper introduces a new method for solving this problem called the recursive importance sketching algorithm (RISRO), in which the central idea is to break the problem down into smaller, easier parts using a unique technique called "recursive importance sketching." This new method is not only easy to use, but it is also very efficient and gives accurate results. We prove that RISRO converges in a local quadratic-linear and quadratic rate under some mild conditions. Simulation studies also demonstrate the superior performance of RISRO. In this paper, we propose a recursive importance sketching algorithm for rank-constrained least squares optimization (RISRO). The key step of RISRO is recursive importance sketching, a new sketching framework based on deterministically designed recursive projections, and it significantly differs from the randomized sketching in the literature. Several existing algorithms in the literature can be reinterpreted under this new sketching framework, and RISRO offers clear advantages over them. RISRO is easy to implement and computationally efficient, and the core procedure in each iteration is to solve a dimension-reduced least squares problem. We establish the local quadratic-linear and quadratic rate of convergence for RISRO under some mild conditions. We also discover a deep connection of RISRO to the Riemannian Gauss–Newton algorithm on fixed rank matrices. The effectiveness of RISRO is demonstrated in two applications in machine learning and statistics: low-rank matrix trace regression and phase retrieval. Simulation studies demonstrate the superior numerical performance of RISRO. Funding: Y. Luo and A. Zhang were partially supported by the National Science Foundation [Grant CAREER-2203741]. W. Huang was partially supported by the Fundamental Research Funds for the Central Universities [Grant 20720190060] and the National Natural Science Foundation of China [Grant 12001455]. X. Li was partially supported by the National Natural Science Foundation of China [Grants 62141407 and 12271107], the Chenguang Program by the Shanghai Education Development Foundation and Shanghai Municipal Education Commission [Grant 19CG02], and the Shanghai Science and Technology Program [Grant 21JC1400600]. Supplemental Material: The online appendix is available at https://doi.org/10.1287/opre.2023.2445. [ABSTRACT FROM AUTHOR]

Optimal stopping is a fundamental class of stochastic dynamic optimization problems with numerous applications in finance and operations management. In "A nonparametric algorithm for optimal stopping based on robust optimization," the author introduces an approach based on robust optimization for computing near-optimal solutions for computationally demanding stochastic optimal stopping problems with known probability distributions. Through this new use of robust optimization, the paper develops new algorithms for solving stochastic optimal stopping problems that are practical and can outperform state-of-the-art simulation-based algorithms in the context of pricing high-dimensional financial options. Optimal stopping is a fundamental class of stochastic dynamic optimization problems with numerous applications in finance and operations management. We introduce a new approach for solving computationally-demanding stochastic optimal stopping problems with known probability distributions. The approach uses simulation to construct a robust optimization problem that approximates the stochastic optimal stopping problem to any arbitrary accuracy; we then solve the robust optimization problem to obtain near-optimal Markovian stopping rules for the stochastic optimal stopping problem. In this paper, we focus on designing algorithms for solving the robust optimization problems that approximate the stochastic optimal stopping problems. These robust optimization problems are challenging to solve because they require optimizing over the infinite-dimensional space of all Markovian stopping rules. We overcome this challenge by characterizing the structure of optimal Markovian stopping rules for the robust optimization problems. In particular, we show that optimal Markovian stopping rules for the robust optimization problems have a structure that is surprisingly simple and finite-dimensional. We leverage this structure to develop an exact reformulation of the robust optimization problem as a zero-one bilinear program over totally unimodular constraints. We show that the bilinear program can be solved in polynomial time in special cases, establish computational complexity results for general cases, and develop polynomial-time heuristics by relating the bilinear program to the maximal closure problem from graph theory. Numerical experiments demonstrate that our algorithms for solving the robust optimization problems are practical and can outperform state-of-the-art simulation-based algorithms in the context of widely-studied stochastic optimal stopping problems from high-dimensional option pricing. Supplemental Material: The e-companion is available at https://doi.org/10.1287/opre.2023.2461. [ABSTRACT FROM AUTHOR]

The article "A Decomposition Algorithm with Fast Identification of Critical Contingencies for Large-Scale Security-Constrained AC-OPF" presents the decomposition algorithm used by Team GO-SNIP for the ARPA-E Grid Optimization (GO) Competition Challenge 1, held from November 2018 through October 2019. The algorithm involves unique contingency ranking and evaluation strategies for determining the important contingencies to include in a master problem that approximates the original large-scale security-constrained problem. It also involves efficient strategies for handling the complementarity constraints that appear in the model and for handling the arising degeneracies. Software implementation details are described, and the results of an extensive set of numerical experiments are provided to illustrate the effectiveness of each of the used techniques. Team GO-SNIP received a second-place finish in Challenge 1 of the Go Competition. A decomposition algorithm for solving large-scale security-constrained AC optimal power flow problems is presented. The formulation considered is the one used in the Advanced Research Projects Agency-Energy Grid Optimization Competition, Challenge 1, held from November 2018 through October 2019. Algorithmic strategies are proposed for contingency selection, fast contingency evaluation, handling complementarity constraints, avoiding issues related to degeneracy, and exploiting parallelism. The results of numerical experiments are provided to demonstrate the effectiveness of the proposed techniques as compared with alternative strategies. History: This paper has been accepted for the Operations Research Special Issue on Computational Advances in Short-Term Power System Operations. Funding: This work was supported by the Advanced Research Projects Agency-Energy [Grant DE-AR0001073]. [ABSTRACT FROM AUTHOR]

Traditional clustering algorithms such as k-means and vanilla spectral clustering are known to deteriorate significantly in the presence of outliers. Several previous works in literature have proposed robust variants of these algorithms; however, they do not provide any theoretical guarantees. Extending previous clustering literature on Gaussian mixture models, in their paper "A Robust Spectral Clustering Algorithm for Sub-Gaussian Mixture Models with Outliers," Prateek R. Srivastava, Purnamrita Sarkar, and Grani A. Hanasusanto developed a new spectral clustering algorithm and provided error bounds for the algorithm under a general sub-Gaussian mixture model setting with outliers. Surprisingly, their derived error bound matches with the best-known bound for semidefinite programs under the same setting without outliers. Numerical experiments on a variety of simulated and real-world data sets further demonstrate that their algorithm is less sensitive to outliers compared with other state-of-the-art algorithms. We consider the problem of clustering data sets in the presence of arbitrary outliers. Traditional clustering algorithms such as k-means and spectral clustering are known to perform poorly for data sets contaminated with even a small number of outliers. In this paper, we develop a provably robust spectral clustering algorithm that applies a simple rounding scheme to denoise a Gaussian kernel matrix built from the data points and uses vanilla spectral clustering to recover the cluster labels of data points. We analyze the performance of our algorithm under the assumption that the "good" data points are generated from a mixture of sub-Gaussians (we term these "inliers"), whereas the outlier points can come from any arbitrary probability distribution. For this general class of models, we show that the misclassification error decays at an exponential rate in the signal-to-noise ratio, provided the number of outliers is a small fraction of the inlier points. Surprisingly, this derived error bound matches with the best-known bound for semidefinite programs (SDPs) under the same setting without outliers. We conduct extensive experiments on a variety of simulated and real-world data sets to demonstrate that our algorithm is less sensitive to outliers compared with other state-of-the-art algorithms proposed in the literature. Funding: G. A. Hanasusanto was supported by the National Science Foundation Grants NSF ECCS-1752125 and NSF CCF-2153606. P. Sarkar gratefully acknowledges support from the National Science Foundation Grants NSF DMS-1713082, NSF HDR-1934932 and NSF 2019844. Supplemental Material: The online appendix is available at https://doi.org/10.1287/opre.2022.2317. [ABSTRACT FROM AUTHOR]

In an observational study there are two disjointed groups of samples, one of treatment samples and the other of control samples. Each of the samples is characterized by several observed covariates. Covariate balancing problems arise when estimating causal effects using observational data. It is desirable to replicate a randomized experiment by obtaining treatment and control groups with similar covariate distributions. Even though covariate balancing problems have been studied extensively, the complexity status of many variants has not been established. Some of our results demonstrate that 2-covariate balancing problems are polynomial time solvable, whereas almost all problems are hard for three or more covariates. These results have practical implications, such as justifying the use of implicit enumeration techniques or heuristics for the hard cases. A new approach suggested by our results is to use the 2-covariate polynomial cases and relax the problem by aggregating covariates into two sets to be solved efficiently. (accepted paper title: Algorithms and complexities of matching variants in covariate balancing) Here, we study several variants of matching problems that arise in covariate balancing. Covariate balancing problems can be viewed as variants of matching, or b-matching, with global side constraints. We present here a comprehensive complexity study of the covariate balancing problems providing polynomial time algorithms, or a proof of NP-hardness. The polynomial time algorithms described are mostly combinatorial and rely on network flow techniques. In addition, we present several fixed-parameter tractable results for problems where the number of covariates and the number of levels of each covariate are seen as a parameter. Funding: This work was supported by National Science Foundation [Grants CMMI-1760102 and NSF 2112533]; Israel Science Foundation (308/18). Supplemental Material: The e-companion is available at https://doi.org/10.1287/opre.2022.2286. [ABSTRACT FROM AUTHOR]

A New Family of Valid-Inequalities for Dantzig-Wolfe Reformulation of Mixed Integer Linear Programs In "Consistency Cuts for Dantzig-Wolfe Reformulation," Jens Vinther Clausen, Richard Lusby, and Stefan Ropke present a new family of valid inequalities to be applied to Dantzig-Wolfe reformulations with binary linking variables. They show that, for Dantzig-Wolfe reformulations of mixed integer linear programs that satisfy certain properties, it is enough to solve the linear programming relaxation of the Dantzig-Wolfe reformulation with all consistency cuts to obtain integer solutions. An example of this is the temporal knapsack problem; the effectiveness of the cuts is tested on a set of 200 instances of this problem, and the results are state-of-the-art solution times. For problems that do not satisfy these conditions, the cuts can still be used in a branch-and-cut-and-price framework. In order to show this, the cuts are applied to a set of generic mixed linear integer programs from the online library MIPLIB. These tests show the applicability of the cuts in general. This paper introduces a family of valid inequalities, which we term consistency cuts, to be applied to a Dantzig-Wolfe reformulation (or decomposition) with linking variables. We prove that these cuts ensure an integer solution to the corresponding Dantzig-Wolfe relaxation when certain criteria to the structure of the decomposition are met. We implement the cuts and use them to solve a commonly used test set of 200 instances of the temporal knapsack problem. We assess the performance with and without the cuts and compare further to CPLEX and other solution methods that have historically been used to solve the test set. By separating consistency cuts, we show that we can obtain optimal integer solutions much faster than the other methods and even solve the remaining unsolved problems in the test set. We also perform a second test on instances from the MIPLIB 2017 online library of mixed-integer programs, showing the potential of the cuts on a wider range of problems. [ABSTRACT FROM AUTHOR]

Games with multiplicative payoffs Trade is win-win: The buyer gets a product that the seller can provide more easily. Selling more and buying more benefits both. This mutual benefit can sometimes be represented multiplicatively. But the buyer also pays a price to the seller. This money transfer is win-lose, a zero-sum game. The sum of the two payoff matrices is a matrix of rank one. In "Fast Algorithms for Rank-1 Bimatrix Games," Bharat Adsul, Jugal Garg, Ruta Mehta, Milind Sohoni, and Bernhard von Stengel show new algorithms that analyze such rank-1 games. The computation time for finding one Nash equilibrium is polynomial. Finding all Nash equilibria takes time comparable to finding a single equilibrium of a general bimatrix game. This puts games in computational reach that are economically much more interesting than zero-sum games (which are of rank zero), whereas solving general games is computationally hard. The paper's detailed structural analysis of games based on their rank enhances our understanding of this computational complexity. The rank of a bimatrix game is the matrix rank of the sum of the two payoff matrices. This paper comprehensively analyzes games of rank one and shows the following: (1) For a game of rank r, the set of its Nash equilibria is the intersection of a generically one-dimensional set of equilibria of parameterized games of rank r − 1 with a hyperplane. (2) One equilibrium of a rank-1 game can be found in polynomial time. (3) All equilibria of a rank-1 game can be found by following a piecewise linear path. In contrast, such a path-following method finds only one equilibrium of a bimatrix game. (4) The number of equilibria of a rank-1 game may be exponential. (5) There is a homeomorphism between the space of bimatrix games and their equilibrium correspondence that preserves rank. It is a variation of the homeomorphism used for the concept of strategic stability of an equilibrium component. [ABSTRACT FROM AUTHOR]

Stochastic inventory systems with lead times are often challenging to optimize, including single-sourcing lost-sales and dual-sourcing systems. Recent numerical results suggest that capped policies demonstrate superior performance over existing heuristics. However, the superior performance lacks a theoretical foundation. In "1.79-Approximation Algorithms for Continuous Review Single-Sourcing Lost-Sales and Dual-Sourcing Inventory Models," the author provides a theoretical foundation for this phenomenon in two classical inventory models. First, in a continuous review lost-sales model with lead times and Poisson demand, he proves that a capped base-stock policy has a worst-case performance guarantee of 1.79 by conducting an asymptotic analysis under a large penalty cost and lead time. Second, in a more complex continuous review dual-sourcing model with general lead times and Poisson demand, he proves that a similar capped dual-index policy has a worst-case performance guarantee of 1.79 under large lead time and ordering cost differences. The results provide a deeper understanding of the superior numerical performance of capped policies and present a new approach to proving worst-case performance guarantees of simple policies in hard inventory problems. Stochastic inventory systems with lead times are often challenging to optimize, including single-sourcing lost-sales and dual-sourcing inventory systems. Recent numerical results suggest that capped policies demonstrate superior performance over existing heuristics. However, the superior performance lacks a theoretical foundation, and why such policies generally perform so well remains a major open question. In this paper, we provide a theoretical foundation for this phenomenon in two classical inventory models with lead times. First, in a continuous-review lost-sales inventory model with lead times and Poisson demand, we prove that a so-called capped base-stock policy has a worst-case performance guarantee of 1.79 by conducting an asymptotic analysis under a large penalty cost and lead time. Second, we extend the analysis to a more complex continuous-review dual-sourcing inventory model with general lead times and Poisson demand and also prove that a so-called capped dual-index policy has a worst-case performance guarantee of 1.79 under large lead time and ordering cost differences. Our results provide a deeper understanding of the superior numerical performance of capped policies and present a new approach to proving worst-case performance guarantees of simple policies in hard inventory problems. [ABSTRACT FROM AUTHOR]

This paper studies a novel decomposition scheme, utilizing decision diagrams for modeling elements of a problem where typical linear relaxations fail to provide sufficiently tight bounds. Given a collection of decision diagrams, each representing a portion of the problem, together with linear inequalities modeling other portions of the problem, how can one efficiently optimize over such a representation? In this paper, we model the problem as a consistent path problem, where a path in each diagram has to be identified, all of which agree on the value assignments to variables. We establish complexity results and propose a branch-and-cut framework for solving the decomposition. Through application to binary cubic optimization and a variant of the market split problem, we show that the decomposition approach provides significant improvement gains over standard linear models. The application of decision diagrams in combinatorial optimization has proliferated in the last decade. In recent years, authors have begun to investigate how to use not one, but a set of diagrams, to model constraints and objective function terms. Optimizing over a collection of decision diagrams, the problem we refer to as the consistent path problem (CPP) can be addressed by associating a network-flow model with each decision diagram, jointly linked through channeling constraints. A direct application of integer programming to the ensuing model has already been shown to result in algorithms that provide orders-of-magnitude performance gains over classical methods. Lacking, however, is a careful study of dedicated solution methods designed to solve the CPP. This paper provides a detailed study of the CPP, including a discussion on complexity results and a complete polyhedral analysis. We propose a cut-generation algorithm, which, under a structured ordering property, finds a cut, if one exists, through an application of the classical maximum flow problem, albeit in an exponentially sized network. We use this procedure to fuel a cutting-plane algorithm that is applied to unconstrained binary cubic optimization and a variant of the market split problem, resulting in an algorithm that compares favorably with CPLEX, using standard integer programming formulations for both problems. [ABSTRACT FROM AUTHOR]

Jose L. Walteros and Austin Buchanan This paper focuses on providing a rigorous, worst-case explanation for why the maximum clique problem is apparently so easy in naturally occurring graphs, despite its intractability in the worst case. To this day, there exist unsolved benchmark instances with just 1,000 vertices. In contrast, real-life instances appear to be much easier. Indeed, relatively simple algorithms can solve instances with millions of vertices in just a few seconds. However, these algorithms lack any worst-case guarantees on their running time. Exactly why these real-life instances are so easy has been somewhat of a puzzle. This paper shows that maximum clique can be solved in time polynomial in the size of the graph, but exponential in a graph invariant that we denote by. Typically, is very small for real-life graphs, being 0, 1 or 2 in over half of the instances coming from commonly used testbeds. In these and other cases where can be thought of as a small constant, our algorithm runs in time. To this day, the maximum clique problem remains a computationally challenging problem. Indeed, despite researchers' best efforts, there exist unsolved benchmark instances with 1,000 vertices. However, relatively simple algorithms solve real-life instances with millions of vertices in a few seconds. Why is this the case? Why is the problem apparently so easy in many naturally occurring networks? In this paper, we provide an explanation. First, we observe that the graph's clique number ω is very near to the graph's degeneracy d in most real-life instances. This observation motivates a main contribution of this paper, which is an algorithm for the maximum clique problem that runs in time polynomial in the size of the graph, but exponential in the gap g ≔ (d + 1) − ω between the clique number ω and its degeneracy-based upper bound d+1. When this gap g can be treated as a constant, as is often the case for real-life graphs, the proposed algorithm runs in time O (d m) = O (m 1.5 ) . This provides a rigorous explanation for the apparent easiness of these instances despite the intractability of the problem in the worst case. Further, our implementation of the proposed algorithm is actually practical—competitive with the best approaches from the literature. [ABSTRACT FROM AUTHOR]

In the paper, "Optimistic Monte Carlo Tree Search with Sampled Information Relaxation Dual Bounds," the authors propose an extension to Monte Carlo tree search that uses the idea of "sampling the future" to produce noisy upper bounds on nodes in the decision tree. These upper bounds can help guide the tree expansion process and produce decision trees that are deeper rather than wider, in effect concentrating computation toward more useful parts of the state space. The algorithm's effectiveness is illustrated in a ride-sharing setting, where a driver/vehicle needs to make dynamic decisions regarding trip acceptance and relocations. Monte Carlo tree search (MCTS), most famously used in game-play artificial intelligence (e.g., the game of Go), is a well-known strategy for constructing approximate solutions to sequential decision problems. Its primary innovation is the use of a heuristic, known as a default policy, to obtain Monte Carlo estimates of downstream values for states in a decision tree. This information is used to iteratively expand the tree toward regions of states and actions that an optimal policy might visit. However, to guarantee convergence to the optimal action, MCTS requires the entire tree to be expanded asymptotically. In this paper, we propose a new "optimistic" tree search technique called primal-dual MCTS that uses sampled information relaxation upper bounds on potential actions to make tree expansion decisions, creating the possibility of ignoring parts of the tree that stem from highly suboptimal choices. The core contribution of this paper is to prove that despite converging to a partial decision tree in the limit, the recommended action from primal-dual MCTS is optimal. The new approach shows promise when used to optimize the behavior of a single driver navigating a graph while operating on a ride-sharing platform. Numerical experiments on a real data set of taxi trips in New Jersey suggest that primal-dual MCTS improves on standard MCTS (upper confidence trees) and other policies while exhibiting a reduced sensitivity to the size of the action space. [ABSTRACT FROM AUTHOR]

This paper considers the optimal adaptive allocation of measurement effort for identifying the best among a finite set of options or designs. An experimenter sequentially chooses designs to measure and observes noisy signals of their quality with the goal of confidently identifying the best design after a small number of measurements. Just as the multiarmed bandit problem crystallizes the tradeoff between exploration and exploitation, this "pure exploration" variant crystallizes the challenge of rapidly gathering information before committing to a final decision. The paper proposes several simple Bayesian algorithms for allocating measurement effort and, by characterizing fundamental asymptotic limits on the performance of any algorithm, formalizes a sense in which these seemingly naive algorithms are the best possible. This paper considers the optimal adaptive allocation of measurement effort for identifying the best among a finite set of options or designs. An experimenter sequentially chooses designs to measure and observes noisy signals of their quality with the goal of confidently identifying the best design after a small number of measurements. This paper proposes three simple and intuitive Bayesian algorithms for adaptively allocating measurement effort and formalizes a sense in which these seemingly naive rules are the best possible. One proposal is top-two probability sampling, which computes the two designs with the highest posterior probability of being optimal and then randomizes to select among these two. One is a variant of top-two sampling that considers not only the probability that a design is optimal, but the expected amount by which its quality exceeds that of other designs. The final algorithm is a modified version of Thompson sampling that is tailored for identifying the best design. We prove that these simple algorithms satisfy a sharp optimality property. In a frequentist setting where the true quality of the designs is fixed, one hopes that the posterior definitively identifies the optimal design, in the sense that that the posterior probability assigned to the event that some other design is optimal converges to zero as measurements are collected. We show that under the proposed algorithms, this convergence occurs at an exponential rate, and the corresponding exponent is the best possible among all allocation rules. It should be highlighted that the proposed algorithms depend on a single tuning parameter, which determines the probability used when randomizing among the top-two designs. Attaining the optimal rate of posterior convergence requires either that this parameter is set optimally or is tuned adaptively toward the optimal value. The paper goes further, characterizing the exponent attained on any problem instance and for any value of the tunable parameter. This exponent is interpreted as being optimal among a constrained class of allocation rules. Finally, considerable robustness to this parameter is established through numerical experiments and theoretical results. When this parameter is set to 1/2, the exponent attained is within a factor of 2 of best possible across all problem instances. [ABSTRACT FROM AUTHOR]

In the paper "Robust Dual Dynamic Programming," Angelos Georghiou, Angelos Tsoukalas, and Wolfram Wiesemann propose a novel solution scheme for addressing planning problems with long horizons. Such problems can be formulated as multistage robust optimization problems. The proposed method takes advantage of the decomposable nature of these problems by bounding the costs arising in the future stages through lower and upper cost-to-go functions. The proposed scheme does not require a relatively complete recourse, and it offers deterministic upper and lower bounds throughout the execution of the algorithm. The promising performance of the algorithm is shown in a stylized inventory-management problem in which the proposed algorithm achieved the optimal solution in problem instances with 100 time stages in a few minutes. Multistage robust optimization problems, where the decision maker can dynamically react to consecutively observed realizations of the uncertain problem parameters, pose formidable theoretical and computational challenges. As a result, the existing solution approaches for this problem class typically determine suboptimal solutions under restrictive assumptions. In this paper, we propose a robust dual dynamic programming (RDDP) scheme for multistage robust optimization problems. The RDDP scheme takes advantage of the decomposable nature of these problems by bounding the costs arising in the future stages through lower and upper cost-to-go functions. For problems with uncertain technology matrices and/or constraint right-hand sides, our RDDP scheme determines an optimal solution in finite time. Also, if the objective function and/or the recourse matrices are uncertain, our method converges asymptotically (but deterministically) to an optimal solution. Our RDDP scheme does not require a relatively complete recourse, and it offers deterministic upper and lower bounds throughout the execution of the algorithm. We show the promising performance of our algorithm in a stylized inventory management problem. Supplemental material is available at https://doi.org/10.1287/opre.2018.1835. [ABSTRACT FROM AUTHOR]

In many sequential decision-making settings where there is uncertainty about the reward of each action, frequent selection of specific actions may reduce expected reward while choosing less frequently selected actions could lead to an increase. These effects are commonly observed in settings ranging from personalized healthcare interventions and targeted online advertising. To address this problem, the authors propose a new class of models called ROGUE (reducing or gaining unknown efficacy) multiarmed bandits. In the paper, the authors present a maximum likelihood approach to estimate the parameters of these models, and we show that these estimates can be used to construct upper confidence bound algorithms and epsilon-greedy algorithms for optimizing these models with strong theoretical guarantees. The authors conclude with a simulation study to show that these algorithms perform better than current nonstationary bandit algorithms in terms of both cumulative regret and average reward. Many settings involve sequential decision making where a set of actions can be chosen at each time step, each action provides a stochastic reward, and the distribution for the reward provided by each action is initially unknown. However, frequent selection of a specific action may reduce the expected reward for that action, whereas abstaining from choosing an action may cause its expected reward to increase. Such nonstationary phenomena are observed in many real-world settings such as personalized healthcare adherence–improving interventions and targeted online advertising. Though finding an optimal policy for general models with nonstationarity is PSPACE-complete, we propose and analyze a new class of models called reducing or gaining unknown efficacy (ROGUE) bandits, which we show in this paper can capture these phenomena and are amenable to the design of policies with provable properties. We first present a consistent maximum likelihood approach to estimate the parameters of these models and conduct a statistical analysis to construct finite sample concentration bounds. Using this analysis, we develop and analyze two different algorithms for optimizing ROGUE models: an upper confidence bound algorithm (ROGUE-UCB) and an ɛ-greedy algorithm (ɛ-ROGUE). Our theoretical analysis shows that under proper conditions, the ROGUE-UCB and ɛ-ROGUE algorithms can achieve logarithmic in time regret, unlike existing algorithms, which result in linear regret. We conclude with a numerical experiment using real-world data from a personalized healthcare adherence–improving intervention to increase physical activity. In this intervention, the goal is to optimize the selection of messages (e.g., confidence increasing versus knowledge increasing) to send to each individual each day to increase adherence and physical activity. Our results show that ROGUE-UCB and ɛ-ROGUE perform better in terms of aggregated regret and average reward when compared with state-of-the-art algorithms, and in the context of this intervention, the use of ROGUE-UCB increases daily step counts by roughly 1,000 steps a day (about a half-mile more of walking) compared with other algorithms in a simulation experiment. [ABSTRACT FROM AUTHOR]

Markov decision processes are commonly used to model forward-looking behavior. However, cyclic terms, including seasonality, are often omitted from these models because of the increase in computational burden. This paper develops a cyclic value function iteration (CVFI), an adjustment to the standard value function iteration. By updating states in a specific order, CVFI allows cyclic variables to be included in the state space with no increase in the computational cost. This result is proved theoretically and shown to hold closely in Monte Carlo simulations. In this paper I develop a cyclic value function iteration, which is an adjustment to the standard value function iteration. When using this algorithm, the inclusion of cyclic variables of any size into the state space of an infinite horizon Markov decision process does not increase the computational complexity of solving for the value function. This result is proven theoretically and shown to closely hold in practice using Monte Carlo simulations. [ABSTRACT FROM AUTHOR]

Sensitivities for Stochastic Optimization In optimization, sensitivities of optimal values with respect to the data of the problem are of practical as well as theoretical interest. In the paper "Envelope Theorems for Multistage Linear Stochastic Optimization," Terça and Wozabal propose a framework to compute derivatives of optimal values for a certain class of stochastic optimization problems. The results make use of classical envelope theorems and almost sure smoothness properties for optimal values of linear optimization problems, which are derived using tools from real algebraic topology. The authors show that derivatives can be sampled based on solutions obtained from stochastic dual dynamic programming and discuss two numerical case studies, demonstrating that the approach is superior, both in terms of accuracy as well as computationally, to naïve methods of computing derivatives that are based on difference quotients. We propose a method to compute derivatives of multistage linear stochastic optimization problems with respect to parameters that influence the problem's data. Our results are based on classical envelope theorems and can be used in problems directly solved via their deterministic equivalents as well as in stochastic dual dynamic programming for which the derivatives of the optimal value are sampled. We derive smoothness properties for optimal values of linear optimization problems, which we use to show that the computed derivatives are valid almost everywhere under mild assumptions. We discuss two numerical case studies, demonstrating that our approach is superior, both in terms of accuracy and computationally, to naïve methods of computing derivatives that are based on difference quotients. [ABSTRACT FROM AUTHOR]

Transforming Dynamic Optimization Problems for Enhanced Computational Efficiency Dynamic programming is one of the core algorithms for solving optimization problems in operations research, economics, finance, computer science, and numerous other fields. Although the theory itself is relatively straightforward, implementation can be computationally expensive, inhibiting application to interesting and realistic problems. This research considers a variety of transformations that can be applied to optimization problems and the key equations associated with the dynamic programming algorithm. It investigates the range of possible transformations, clarifies their links with optimality, and applies specific transformations to generate large speed gains in high-dimensional problems. Several economic applications are considered. Some approaches to solving challenging dynamic programming problems, such as Q-learning, begin by transforming the Bellman equation into an alternative functional equation to open up a new line of attack. Our paper studies this idea systematically with a focus on boosting computational efficiency. We provide a characterization of the set of valid transformations of the Bellman equation, for which validity means that the transformed Bellman equation maintains the link to optimality held by the original Bellman equation. We then examine the solutions of the transformed Bellman equations and analyze correspondingly transformed versions of the algorithms used to solve for optimal policies. These investigations yield new approaches to a variety of discrete time dynamic programming problems, including those with features such as recursive preferences or desire for robustness. Increased computational efficiency is demonstrated via time complexity arguments and numerical experiments. [ABSTRACT FROM AUTHOR]

When dealing with dynamic optimization problems, time consistency is a desirable property as it allows one to solve the problem efficiently through a backward recursion. The mean-risk problem is known to be time inconsistent when considered in its scalarized form. However, when left in its original bi-objective form, it turns out to satisfy a more general time consistency property that seems better suited to a vector optimization problem. In "Time Consistency of the Mean-Risk Problem," Kováĉova and Rudloff introduce a set-valued version of the famous Bellman principle and show that the bi-objective mean-risk problem does satisfy it. Then, the upper image, a set that contains the efficient frontier on its boundary, recurses backward in time. Kováĉova and Rudloff present conditions under which this recursion can be exploited directly to compute a solution in the spirit of dynamic programming. This opens the door for a new branch in mathematics: dynamic multivariate programming. Choosing a portfolio of risky assets over time that maximizes the expected return at the same time as it minimizes portfolio risk is a classical problem in mathematical finance and is referred to as the dynamic Markowitz problem (when the risk is measured by variance) or, more generally, the dynamic mean-risk problem. In most of the literature, the mean-risk problem is scalarized, and it is well known that this scalarized problem does not satisfy the (scalar) Bellman's principle. Thus, the classical dynamic programming methods are not applicable. For the purpose of this paper we focus on the discrete time setup, and we will use a time-consistent dynamic convex risk measure to evaluate the risk of a portfolio. We will show that, when we do not scalarize the problem but leave it in its original form as a vector optimization problem, the upper images, whose boundaries contain the efficient frontiers, recurse backward in time under very mild assumptions. Thus, the dynamic mean-risk problem does satisfy a Bellman's principle, but a more general one, that seems more appropriate for a vector optimization problem: a set-valued Bellman's principle. We will present conditions under which this recursion can be exploited directly to compute a solution in the spirit of dynamic programming. Numerical examples illustrate the proposed method. The obtained results open the door for a new branch in mathematics: dynamic multivariate programming. [ABSTRACT FROM AUTHOR]

In "Improving the Chilean College Admissions System," Rios, Larroucau, Parra, and Cominetti describe the design and implementation of a new system to solve the Chilean college admissions problem. The authors develop an algorithm that (i) obtains all applicant/program pairs that can be part of a stable allocation when preferences are not strict and when all students tied in the last seat of a program (if any) must be allocated and (ii) efficiently incorporates affirmative action, which is part of the system to correct the inefficiencies that arise from having double-assigned students. By unifying the regular admission with affirmative action, the solution proposed and implemented by the authors has improved the allocation of approximately 2.5% of students assigned every year since 2016, helping to improve the overall efficiency of the system. In this paper we present the design and implementation of a new system to solve the Chilean college admissions problem. We develop an algorithm that obtains all applicant/program pairs that can be part of a stable allocation when preferences are not strict and when all students tied in the last seat of a program (if any) must be allocated. We use this algorithm to identify which mechanism was used in the past to perform the allocation, and we propose a new method to incorporate the affirmative action that is part of the system to correct the inefficiencies that arise from having double-assigned students. By unifying the regular admission with the affirmative action, we have improved the allocation of approximately 2.5% of students assigned every year since 2016. From a theoretical standpoint, we show that some desired properties, such as strategy-proofness and monotonicity, cannot be guaranteed under flexible quotas. [ABSTRACT FROM AUTHOR]

Temporal difference learning (TD) is a simple iterative algorithm widely used for policy evaluation in Markov reward processes. Bhandari et al. prove finite time convergence rates for TD learning with linear function approximation. The analysis follows using a key insight that establishes rigorous connections between TD updates and those of online gradient descent. In a model where observations are corrupted by i.i.d. noise, convergence results for TD follow by essentially mirroring the analysis for online gradient descent. Using an information-theoretic technique, the authors also provide results for the case when TD is applied to a single Markovian data stream where the algorithm's updates can be severely biased. Their analysis seamlessly extends to the study of TD learning with eligibility traces and Q-learning for high-dimensional optimal stopping problems. Temporal difference learning (TD) is a simple iterative algorithm used to estimate the value function corresponding to a given policy in a Markov decision process. Although TD is one of the most widely used algorithms in reinforcement learning, its theoretical analysis has proved challenging and few guarantees on its statistical efficiency are available. In this work, we provide a simple and explicit finite time analysis of temporal difference learning with linear function approximation. Except for a few key insights, our analysis mirrors standard techniques for analyzing stochastic gradient descent algorithms and therefore inherits the simplicity and elegance of that literature. Final sections of the paper show how all of our main results extend to the study of TD learning with eligibility traces, known as TD(λ), and to Q-learning applied in high-dimensional optimal stopping problems. [ABSTRACT FROM AUTHOR]

Real-Time Resource Allocation: Beyond Stochastic Demand Modeling Motivated by the practical concern that algorithm designs based on stochastic modeling of demand may perform poorly because of unpredictable patterns, "Online Resource Allocation Under Partially Predictable Demand" proposes a new arrival model with both unpredictable and predictable components. Under the proposed model—which requires no demand forecasting—the authors design novel online algorithms that take advantage of the limited information the data reveal, despite the presence of an unpredictable component. Their algorithms are adjustable to the relative size of the stochastic component; their analysis reveals that as the portion of the stochastic component grows, the loss due to making online decisions decreases. This highlights the value of (even partial) predictability in online resource allocation. This paper serves as a first step in bridging the long-standing gap between the two well-studied approaches to the design and analysis of online algorithms based on (1) adversarial models and (2) stochastic ones. For online resource allocation problems, we propose a new demand arrival model where the sequence of arrivals contains both an adversarial component and a stochastic one. Our model requires no demand forecasting; however, because of the presence of the stochastic component, we can partially predict future demand as the sequence of arrivals unfolds. Under the proposed model, we study the problem of the online allocation of a single resource to two types of customers and design online algorithms that outperform existing ones. Our algorithms are adjustable to the relative size of the stochastic component; our analysis reveals that as the portion of the stochastic component grows, the loss due to making online decisions decreases. This highlights the value of (even partial) predictability in online resource allocation. We impose no conditions on how the resource capacity scales with the maximum number of customers. However, we show that using an adaptive algorithm—which makes online decisions based on observed data—is particularly beneficial when capacity scales linearly with the number of customers. Our work serves as a first step in bridging the long-standing gap between the two well-studied approaches to the design and analysis of online algorithms based on (1) adversarial models and (2) stochastic ones. Using novel algorithm design, we demonstrate that even if the arrival sequence contains an adversarial component, we can take advantage of the limited information that the data reveal to improve allocation decisions. We also study the classical secretary problem under our proposed arrival model, and we show that randomizing over multiple stopping rules may increase the probability of success. [ABSTRACT FROM AUTHOR]

New algorithms and models enable randomized network interdiction strategies Traditional network interdiction problems limit the leader to use a deterministic interdiction strategy, but a stochastic interdiction strategy can present a greater dilemma to the follower who seeks a shortest path on the network. Holzmann and Smith examine a shortest-path interdiction problem in which the leader randomizes its attacks, and the follower only knows the probability of where attacks will occur. The cost of each arc increases as a function of the number of times the arc is interdicted, and the leader seeks to maximize the follower's shortest expected path. When the arc cost function is linear in the number of attacks that occur on the arcs, this problem is solvable in polynomial time by linear programming. However, both the special cases in which the arcs are all convex or all concave turn out to be NP-hard. Holzmann and Smith give general-purpose algorithms for this interdiction problem, along with a polynomial-time approximation algorithm for the case of concave cost functions. Shortest-path interdiction problems involve a leader and a follower playing a zero-sum game over a directed network. The leader interdicts a set of arcs, and arc costs increase as a function of the number of times they are interdicted. The follower observes the leader's actions and selects a shortest path in response. The leader's optimal interdiction strategy maximizes the follower's minimum-cost path. In classic formulations of these problems, the leader's interdiction actions are deterministic. In this paper the leader selects a policy of randomized interdiction actions, and the follower only knows the probability of where interdictions are deployed on the network. The follower identifies a path having the minimum expected cost, whereas the leader seeks to maximize the follower's objective. When the arc costs are affine functions of the number of times they are interdicted, this problem is solvable by linear programming. However, when the cost functions are convex or when they are concave, we show that the expected costs are Schur concave or convex, respectively. This property allows us to prove that the problem becomes strongly NP-hard in the nonaffine case. We also propose several algorithms for this problem. Some are for special cases, and one is a general algorithm. We examine the efficacy of our algorithms on a test bed of randomly generated instances by comparing our algorithms to a standard solver. [ABSTRACT FROM AUTHOR]

This paper lays the foundation for employing Gaussian Markov random fields (GMRFs) for discrete decision–variable optimization via simulation; that is, optimizing the performance of a simulated system. Gaussian processes have gained popularity for inferential optimization, which iteratively updates a model of the simulated solutions and selects the next solution to simulate by relying on statistical inference from that model. We show that, for a discrete problem, GMRFs, a type of Gaussian process defined on a graph, provides better inference on the remaining optimality gap than the typical choice of continuous Gaussian process and thereby enables the algorithm to search efficiently and stop correctly when the remaining optimality gap is below a predefined threshold. We also introduce the concept of multiresolution GMRFs for large-scale problems, with which GMRFs of different resolutions interact to efficiently focus the search on promising regions of solutions. We consider optimizing the expected value of some performance measure of a dynamic stochastic simulation with a statistical guarantee for optimality when the decision variables are discrete, in particular, integer-ordered; the number of feasible solutions is large; and the model execution is too slow to simulate even a substantial fraction of them. Our goal is to create algorithms that stop searching when they can provide inference about the remaining optimality gap similar to the correct-selection guarantee of ranking and selection when it simulates all solutions. Further, our algorithm remains competitive with fixed-budget algorithms that search efficiently but do not provide such inference. To accomplish this we learn and exploit spatial relationships among the decision variables and objective function values using a Gaussian Markov random field (GMRF). Gaussian random fields on continuous domains are already used in deterministic and stochastic optimization because they facilitate the computation of measures, such as expected improvement, that balance exploration and exploitation. We show that GMRFs are particularly well suited to the discrete decision–variable problem, from both a modeling and a computational perspective. Specifically, GMRFs permit the definition of a sensible neighborhood structure, and they are defined by their precision matrices, which can be constructed to be sparse. Using this framework, we create both single and multiresolution algorithms, prove the asymptotic convergence of both, and evaluate their finite-time performance empirically. The e-companion is available at https://doi.org/10.1287/opre.2018.1778. [ABSTRACT FROM AUTHOR]

With the emergence of ride-sharing companies that offer transportation on demand at a large scale and the increasing availability of corresponding demand data sets, new challenges arise to develop routing optimization algorithms that can solve massive problems in real time. In "Online Vehicle Routing: The Edge of Optimization in Large-Scale Applications," D. Bertsimas, P. Jaillet, and S. Martin present a novel and generalizable backbone algorithm that uses integer optimization to find high-quality solutions to large routing optimization problems. The algorithm, together with the real-time routing optimization software framework developed and shared by the authors, can dispatch thousands of taxis serving more than 25,000 customers per hour. An extensive study with historical simulations of Yellow Cabs in New York City is used to both show that the algorithm improves on the performance of existing heuristics and to provide insights on the optimization opportunities of a large ride-sharing fleet. With the emergence of ride-sharing companies that offer transportation on demand at a large scale and the increasing availability of corresponding demand data sets, new challenges arise to develop routing optimization algorithms that can solve massive problems in real time. In this paper, we develop an optimization framework, coupled with a novel and generalizable backbone algorithm, that allows us to dispatch in real time thousands of taxis serving more than 25,000 customers per hour. We provide evidence from historical simulations using New York City routing network and yellow cab data to show that our algorithms improve upon the performance of existing heuristics in such real-world settings. The online supplement is available at https://doi.org/10.1287/opre.2018.1763. [ABSTRACT FROM AUTHOR]

Maximum-entropy sampling is a difficult nonlinear discrete optimization problem that arises in spatial statistics, for example, in the design of weather-monitoring networks. An exact algorithm for maximum-entropy sampling was first described in 1995, and subsequent papers have devised a variety of methods that obtain bounds and, in some cases, exact solutions. In "An Efficient Algorithm for Maximum-Entropy Sampling," Anstreicher describes a new bound for the maximum-entropy sampling problem that is superior to all previously known bounds and is also efficiently computable. A branch-and-bound algorithm that incorporates the new bound solves challenging benchmark instances to optimality for the first time. We consider a new approach for the maximum-entropy sampling problem (MESP) that is based on bounds obtained by maximizing a function of the form ldet M (x) over linear constraints, where M (x) is linear in the n -vector x. These bounds can be computed very efficiently and are superior to all previously known bounds for MESP on most benchmark test problems. A branch-and-bound algorithm using the new bounds solves challenging instances of MESP to optimality for the first time. [ABSTRACT FROM AUTHOR]

Air traffic management measures comprise tactical operating procedures to minimize delay costs and strategic scheduling interventions to control overcapacity scheduling. Although interdependent, these problems have been treated in isolation. This paper proposes an integrated model of scheduling and operations in airport networks that jointly optimizes scheduling interventions and ground-holding operations across airports networks under operating uncertainty. It is formulated as a two-stage stochastic program with integer recourse. To solve it, we develop an original decomposition algorithm with provable solution quality guarantees. The algorithm relies on new optimality cuts—dual integer cuts—that leverage the reduced costs of the dual linear programming relaxation of the second-stage problem. The algorithm also incorporates neighborhood constraints, which shift from exploration to exploitation at later stages. We also use a scenario generation approach to construct representative scenarios from historical records of operations—using integer programming. Computational experiments show that our algorithm yields near-optimal solutions for the entire U.S. National Airspace System network. Ultimately, the proposed approach enhances airport demand management models through scale integration (by capturing network-wide interdependencies) and scope integration (by capturing interdependencies between scheduling and operations). [ABSTRACT FROM AUTHOR]

Adaptive Matching Under Uncertainties. Two-sided online markets often propose matches based on imperfect knowledge of the characteristics of the two parties to be matched. Such uncertainty may result in inferior matches and may in turn incur negative externalities: a matched resource becomes unavailable to a more suitable party, at least for a while. For example, in labour platforms, if an expert is matched to a task which does not meet his or her expertise, then the other tasks which meet the expert's expertise may suffer. Similarly, in hospitality platforms, an economical accommodation becomes unavailable to a financially constrained customer if it is matched to a flexible customer. Thus, to optimize the long-term efficiency, a platform needs to optimize exploration-exploitation trade-off while accounting for the resource usage. To address this challenge, the paper "Adaptive Matching for Expert Systems with Uncertain Task Types" develops a new model and optimal algorithm for a task-expert matching system, where a task is matched to an expert using not only the prior information about the task but also the feedback obtained from the past matches. A matching in a two-sided market often incurs an externality: a matched resource may become unavailable to the other side of the market, at least for a while. This is especially an issue in online platforms involving human experts, as the expert resources are often scarce. The efficient utilization of experts in these platforms is made challenging by the fact that the information available about the parties involved is usually limited. To address this challenge, we develop a model of a task-expert matching system where a task is matched to an expert using not only the prior information about the task but also the feedback obtained from the past matches. In our model, the tasks arrive online while the experts are fixed and constrained by a finite service capacity. For this model, we characterize the maximum task resolution throughput a platform can achieve. We show that the natural greedy approach where each expert is assigned a task most suitable to his or her skill is suboptimal, as it does not internalize the aforementioned externality. We develop a throughput-optimal backpressure algorithm which does so by accounting for the "congestion" among different task types. Finally, we validate our model and confirm our theoretical findings with data-driven simulations via logs of Math.StackExchange.com, a Stack Overflow forum dedicated to mathematics. [ABSTRACT FROM AUTHOR]

A New Algorithm for Static Bin Packing Static bin packing is the problem of partitioning a set of items (with scalar sizes) into identical bins of a given capacity, under the constraint that the total size of no partition exceeds the bin capacity. In the online variant, the list of items is revealed one at a time—an item must be assigned irrevocably to a partition at the time it is revealed. In "Interior-Point Based Online Stochastic Bin Packing," V. Gupta and A. Radovanović look at the online variant under the assumption that the item sizes are independent samples from a common unknown distribution. The authors show that a greedy heuristic that assigns items so as to minimize a carefully chosen penalized Lagrangian of the offline math program is asymptotically optimal. Since the proposed heuristic is greedy in nature, it extends readily to nonstationary item size sequences for which the authors prove a novel family of performance bounds. Bin packing is an algorithmic problem that arises in diverse applications such as remnant inventory systems, shipping logistics, and appointment scheduling. In its simplest variant, a sequence of T items (e.g., orders for raw material, packages for delivery) is revealed one at a time, and each item must be packed on arrival in an available bin (e.g., remnant pieces of raw material in inventory, shipping containers). The sizes of items are independent and identically distributed (i.i.d.) samples from an unknown distribution, but the sizes are known when the items arrive. The goal is to minimize the number of nonempty bins (equivalently waste, defined to be the total unused space in nonempty bins). This problem has been extensively studied in the operations research and theoretical computer science communities, yet all existing heuristics either rely on learning the distribution or exhibit o(T) additive suboptimality compared with the optimal offline algorithm only for certain classes of distributions (those with sublinear optimal expected waste). In this paper, we propose a family of algorithms that are the first truly distribution-oblivious algorithms for stochastic bin packing, and achieve O (T) additive suboptimality for all item size distributions. Our algorithms are inspired by approximate interior-point algorithms for convex optimization. In addition to regret guarantees for discrete i.i.d. sequences, we extend our results to continuous item size distribution with bounded density, and prove a family of novel regret bounds for non-i.i.d. input sequences. To the best of our knowledge, these are the first such results for non-i.i.d. and non-random-permutation input sequences for online stochastic packing. [ABSTRACT FROM AUTHOR]

Vehicle scheduling is a central problem in transportation. As railways and the airspace become increasingly congested, it is crucial to develop mathematical tools to better schedule trains and airplanes off-line (timetabling) and online (dispatching). These traffic-management problems can be modeled as job-shop scheduling problems, which, in turn, can be solved using mixed-integer linear programming (MILP). The two most successful and widespread formulations in the past decades are the big-M and the time-indexed methods, each with its strengths and weaknesses. However, there have been few advancements in either approach in recent years. In this paper, we introduce an alternative formulation, obtained by strengthening and lifting the Benders' reformulation of a natural initial formulation. This new approach improves performances on our real-life test bed and opens up interesting research directions. A central problem in traffic management is that of scheduling the movements of vehicles so as to minimize the cost of the schedule. It arises in important applications such as train timetabling, rescheduling, delay and disruption management, airplane surface routing, runway scheduling, air-traffic control, and more. This problem can be modeled as a job-shop scheduling problem. We introduce a new mixed-integer linear program (MILP) formulation for job-shop scheduling, which is an alternative to classical approaches, namely, big-M and time-indexed formulations. It does not make use of artificially large coefficients, and its constraints correspond to basic graph structures, such as paths, cycles, and trees. The new formulation can be obtained by strengthening and lifting the constraints of a classical Benders' reformulation. Tests on a large set of real-life instances from train rescheduling show that the new approach performs on average better than our previous approaches based on big-M formulations and particularly better on a class of instances with nonconvex costs very common in the practice. [ABSTRACT FROM AUTHOR]

The question of aggregating pairwise comparisons to obtain a global ranking over a collection of objects has been of interest for a very long time: be it ranking of online gamers (e.g., MSR's TrueSkill system) and chess players, aggregating social opinions, or deciding which product to sell based on transactions. In most settings, in addition to obtaining a ranking, finding 'scores' for each object (e.g., player's rating) is of interest for understanding the intensity of the preferences. In this paper, we propose Rank Centrality, an iterative rank aggregation algorithm for discovering scores for objects (or items) from pairwise comparisons. The algorithm has a natural random walk interpretation over the graph of objects with an edge present between a pair of objects if they are compared; the score, which we call Rank Centrality, of an object turns out to be its stationary probability under this random walk. To study the efficacy of the algorithm, we consider the popular Bradley-Terry-Luce (BTL) model (equivalent to the Multinomial Logit (MNL) for pairwise comparisons) in which each object has an associated score that determines the probabilistic outcomes of pairwise comparisons between objects. In terms of the pairwise marginal probabilities, which is the main subject of this paper, the MNL model and the BTL model are identical. We bound the finite sample error rates between the scores assumed by the BTL model and those estimated by our algorithm. In particular, the number of samples required to learn the score well with high probability depends on the structure of the comparison graph. When the Laplacian of the comparison graph has a strictly positive spectral gap, e.g., each item is compared to a subset of randomly chosen items, this leads to dependence on the number of samples that is nearly order optimal. Experimental evaluations on synthetic data sets generated according to the BTL model show that our algorithm performs as well as the maximum likelihood estimator for that model and outperforms other popular ranking algorithms. [ABSTRACT FROM AUTHOR]

We study a class of problems with both binary selection decisions and associated continuous choices that result in stochastic rewards and costs. The rewards are received based on the decision maker's selection, and the costs depend both on the decisions and realizations of the stochastic variables. We consider a family of risk-based objective functions that contains the traditional risk-neutral expected-value objective as a special case. A combination of rounding and sample average approximation is used to produce solutions that are guaranteed to be close to the optimal solution with high probability. We also provide an empirical comparison of the performance of the algorithms on a set of randomly generated instances of a supply chain example problem. The computational results illustrate the theoretical claims in the paper that, for this problem, high-quality solutions can be found with small computational effort. [ABSTRACT FROM AUTHOR]

This paper proposes a general framework for a large class of multiperiod principal-agent problems. In this framework, a principal has a primary stake in the performance of a system but delegates its control to an agent. The underlying system is a Markov decision process, where the state of the system can only be observed by the agent but the agent's action is observed by both parties. This paper develops a dynamic programming algorithm to derive optimal long-term contracts for the principal. The principal indirectly controls the underlying system by offering the agent a menu of continuation utility vectors along public information paths; the agent's best response, expressed in his choice of continuation utilities, induces truthful state revelation and results in actions that maximize the principal's expected payoff. This problem is meaningful to the operations research community because it can be framed as the problem of optimally designing the reward structure of a Markov decision process with hidden states and has many applications of interest as discussed in this paper [ABSTRACT FROM AUTHOR]

This paper focuses on revealed preference tests of the collective model of household consumption. We start by showing that the decision problems corresponding to testing collective rationality are NP-complete. This makes the application of these tests problematic for (increasingly available) large(r)-scale data sets. We then present two approaches to overcome this negative result. First, we introduce exact algorithms based on mixed-integer programming (MIP) formulations of the collective rationality tests, which can be usefully applied to medium-sized data sets. Next, we propose simulated annealing heuristics, which allow for efficient testing of the collective model in the case of large data sets. We illustrate our methods by a number of computational experiments based on Dutch labor supply data. [ABSTRACT FROM AUTHOR]

Effective route planning for battery electric commercial vehicle (ECV) fleets has to take into account their limited autonomy and the possibility of visiting recharging stations during the course of a route. In this paper, we consider four variants of the electric vehicle-routing problem with time windows: (i) at most a single recharge per route is allowed, and batteries are fully recharged on visit of a recharging station; (ii) multiple recharges per route, full recharges only; (iii) at most a single recharge per route, and partial battery recharges are possible; and (iv) multiple, partial recharges. For each variant, we present exact branch-price-and-cut algorithms that rely on customized monodirectional and bidirectional labeling algorithms for generating feasible vehicle routes. In computational studies, we find that all four variants are solvable for instances with up to 100 customers and 21 recharging stations. This success can be attributed to the tailored resource extension functions (REFs) that enable efficient labeling with constant time feasibility checking and strong dominance rules, even if these REFs are intricate and rather elaborate to derive. The studies also highlight the superiority of the bidirectional labeling algorithms compared to the monodirectional ones. Finally, we find that allowing multiple as well as partial recharges both help to reduce routing costs and the number of employed vehicles in comparison to the variants with single and with full recharges. [ABSTRACT FROM AUTHOR]

Game-theoretic tools are becoming a popular design choice for distributed resource allocation algorithms. A central component of this design choice is the assignment of utility functions to the individual agents. The goal is to assign each agent an admissible utility function such that the resulting game possesses a host of desirable properties, including scalability, tractability, and existence and efficiency of pure Nash equilibria. In this paper we formally study this question of utility design on a class of games termed distributed welfare games. We identify several utility design methodologies that guarantee desirable game properties irrespective of the specific application domain. Lastly, we illustrate the results in this paper on two commonly studied classes of resource allocation problems: "coverage" problems and "coloring" problems. [ABSTRACT FROM AUTHOR]

Lately, the problem of cell formation (CF) has gained a lot of attention in the industrial engineering literature. Since it was formulated (more than 50 years ago), the problem has incorporated additional industrial factors and constraints while its solution methods have been constantly improving in terms of the solution quality and CPU times. However, despite all the efforts made, the available solution methods (including those for a popular model based on the p-median problem, PMP) are prone to two major types of errors. The first error (the modeling one) occurs when the intended objective function of the CF (as a rule, verbally formulated) is substituted by the objective function of the PMP. The second error (the algorithmic one) occurs as a direct result of applying a heuristic for solving the PMP. In this paper we show that for instances that make sense in practice, the modeling error induced by the PMP is negligible. We exclude the algorithmic error completely by solving the adjusted pseudo-Boolean formulation of the PMP exactly, which takes less than one second on a general-purpose PC and software. Our experimental study shows that the PMP-based model produces high-quality cells and in most cases outperforms several contemporary approaches. [ABSTRACT FROM AUTHOR]

Zhao showed that the log barrier associated with the recourse function of two-stage stochastic linear programs behaves as a strongly self-concordant barrier and forms a self-concordant family on the first-stage solutions. In this paper, we show that the recourse function is also strongly self-concordant and forms a self-concordant family for the two-stage stochastic convex quadratic programs with recourse. This allows us to develop Bender's decomposition based linearly convergent interior point algorithms. An analysis of such an algorithm is given in this paper. [ABSTRACT FROM AUTHOR]

In this paper, we investigate the properties of the sampled version of the fictitious play algorithm, familiar from game theory, for games with identical payoffs, and propose a heuristic based on fictitious play as a solution procedure for discrete optimization problems of the form max {u(y) y = (y1,…yn ∈ y1 x … x yn} i.e., in which the feasible region is a Cartesian product of finite sets y1 i ∈ N = (1, …, n). The contributions of this paper are twofold. In the first part of the paper, we broaden the existing results on convergence properties of the fictitious play algorithm on games with identical payoffs to include an approximate fictitious play algorithm that allows for errors in players 'best replies. Moreover, we introduce sampling-based approximate fictitious play that possesses the above convergence properties, and at the same time provides a computationally efficient method for implementing fictitious play. In the second part of the paper, we motivate the use of algorithms based on sampled fictitious play to solve optimization problems in the above form with particular focus on the problems in which the objective function u(·) comes from a "black box," such as a simulation model, where significant computational effort is required for each function evaluation. [ABSTRACT FROM AUTHOR]

We develop a new method for pricing American options. The main practical contribution of this paper is a general algorithm for constructing upper and lower bounds on the true price of the option using any approximation to the option price. We show that our bounds are tight, so that if the initial approximation is close to the true price of the option, the bounds are also guaranteed to be close. We also explicitly characterize the worst-case performance of the pricing bounds. The computation of the lower bound is straightforward and relies on simulating the suboptimal exercise strategy implied by the approximate option price. The upper bound is also computed using Monte Carlo simulation. This is made feasible by the representation of the American option price as a solution of a properly defined dual minimization problem, which is the main theoretical result of this paper. Our algorithm proves to be accurate on a set of sample problems where we price call options on the maximum and the geometric mean of a collection of stocks. These numerical results suggest that our pricing method can be successfully applied to problems of practical interest. [ABSTRACT FROM AUTHOR]

We study the strip packing problem, in which a set of two-dimensional rectangular items has to be packed in a rectangular strip of fixed width and infinite height, with the aim of minimizing the height used. The problem is important because it models a large number of real-world applications, including cutting operations where stocks of materials such as paper or wood come in large rolls and have to be cut with minimum waste, scheduling problems in which tasks require a contiguous subset of identical resources, and container loading problems arising in the transportation of items that cannot be stacked one over the other. The strip packing problem has been attacked in the literature with several heuristic and exact algorithms, nevertheless, benchmark instances of small size remain unsolved to proven optimality. In this paper we propose a new exact method that solves a large number of the open benchmark instances within a limited computational effort. Our method is based on a Benders' decomposition, in which in the master we cut items into unit-width slices and pack them contiguously in the strip, and in the slave we attempt to reconstruct the rectangular items by fixing the vertical positions of their unit-width slices. If the slave proves that the reconstruction of the items is not possible, then a cut is added to the master, and the algorithm is reiterated. We show that both the master and the slave are strongly NP-hard problems and solve them with tailored preprocessing, lower and upper bounding techniques, and exact algorithms. We also propose several new techniques to improve the standard Benders' cuts, using the so-called combinatorial Benders' cuts, and an additional lifting procedure. Extensive computational tests show that the proposed algorithm provides a substantial breakthrough with respect to previously published algorithms. [ABSTRACT FROM AUTHOR]

The article presents a zero-one linear program and a dynamic programming algorithm for finding the shortest route containing exactly q arcs from node 1 to node n in a network (N, A) with distances c(i, j). This note shows that the linear programming formulation and his extension based on it are defective, and that the dynamic programming algorithm can lead to suboptimal solutions, but a minor change in the dynamic programming formulation relieves the difficulty. A special feature of this linear program is that there can never be a loop in the basis, because the vectors corresponding to the variables of a loop are linearly dependent. The last comment is related to the dynamic programming algorithm. One is allowed to pass through a node more than once in the shortest route . The proposed method for solving this program consists of converting it into a shortest route problem containing exactly q arcs. But for some arbitrary reason, the dynamic program is formulated in such a way that one can never have starting node 1 more than once in the final solution.

We develop the first nonparametric learning algorithm for periodic-review perishable inventory systems. In contrast to the classical perishable inventory literature, we assume that the firm does not know the demand distribution a priori and makes replenishment decisions in each period based only on the past sales (censored demand) data. It is well known that even with complete information about the demand distribution a priori, the optimal policy for this problem does not possess a simple structure. Motivated by the studies in the literature showing that base-stock policies perform near optimal in these systems, we focus on finding the best base-stock policy. We first establish a convexity result, showing that the total holding, lost sales and outdating cost is convex in the base-stock level. Then, we develop a nonparametric learning algorithm that generates a sequence of order-up-to levels whose running average cost converges to the cost of the optimal base-stock policy. We establish a square-root convergence rate of the proposed algorithm, which is the best possible. Our algorithm and analyses require a novel method for computing a valid cycle subgradient and the construction of a bridging problem, which significantly departs from previous studies. [ABSTRACT FROM AUTHOR]

This paper is concerned with a linear fractional problem of the form: max[sub x] min[sub y] F(X, Y) = (cX + dY + α)/(fX + gY + β), subject to AX + BY less than or equal to b; X, Y greater than or equal to O. This problem represents a generalization of a problem considered in the literature in which F(X, Y) is assumed to be linear. A number of results for the linear case are extended; and, in particular, it is shown that this fractional max-min problem is equivalent to a quasi-convex programming problem whose optimal solution lies at a vertex of the feasible region. Using these results, we develop an algorithm for solving this problem. The paper concludes with a numerical example. [ABSTRACT FROM AUTHOR]

In an earlier paper [Opns. Res. 20, 1153-1161 (1972)] we proved that any feasible integer solution to the linear program associated with the equality-constrained set-covering problem can be obtained from any other feasible integer solution by a sequence of less than m pivots (where m is the number of equations), such that each solution generated in the sequence is integer. However, degeneracy makes it difficult to find a sequence of pivots leading to an integer optimum. In this paper we give a constructive characterization of adjacency relations between integer vertices of the feasible set that enables us to generate edges (all, if necessary) connecting a given integer vertex to adjacent integer vertices. This helps overcome the difficulties caused by degeneracy and leads to a class of algorithms, of which we discuss two. [ABSTRACT FROM AUTHOR]

This paper studies the linear programming problem in which all coefficients (even those of the stipulations matrix) are rational functions of a single parameter t called ‘time,’ and provides an algorithm that can serve problems of the following two types: (1) Steady-state behavior [the algorithm can be used to determine the functional form x(t) of the optimal solution as a function of t, this form being valid for all ‘sufficiently large’ values of t], and (2) sensitivity analysis [if a value t0 of ‘time’ is given, the algorithm can be used to determine the two possible functional forms of the optimal solution for all values of t ‘sufficiently dose’ to t0 (the first functional form valid for t«t0, the second for t»t0)]. In addition, the paper gives certain qualitative information regarding steady-state behavior, including the following result: If for some one of the properties of consistency, boundedness, or bounded constraint set, there exists a sequence tn↗+∞ such that the linear program at tn has this property for all n, then the program has this property for all ‘sufficiently large’ values of t. [ABSTRACT FROM AUTHOR]

This letter calls attention to the problems of validating the claim made for algorithms in published papers, and proposes a system for review that will validate such claims thoroughly before publication. [ABSTRACT FROM AUTHOR]

This paper examines properties and interrelations of several concepts of generalized concavity. It shows that the class of functions that are both ‘generalized concave’ and ‘generalized convex’ is closely related to the class of monotone functions of a single variable. After excluding a certain small class of exceptions, the paper shows that, for arbitrary (perhaps not differentiable) functions, concave implies pseudoconcave, pseudoconcave implies strictly quasiconcave, and strictly quasiconcave implies quasiconcave. Several results characterizing the extreme values of generalized concave functions are given. [ABSTRACT FROM AUTHOR]

This paper considers the mathematical properties of chance constrained programming problems where the restriction is on the joint probability of a multivariate random event. One model that is considered arises when the right-hand side constants of the linear constraints are random. Another model treated here occurs when the coefficients of the linear programming variables are described by a multinormal distribution. It is shown that under certain restrictions both situations can be viewed as a deterministic nonlinear programming problem. Since most computational methods for solving nonlinear programming models require the constraints be concave, this paper explores whether the resultant problem meets the concavity assumption. For many probability laws of practical importance, the constraint in the first type of model is shown to violate concavity. However, a simple logarithmic transformation does produce a concave restriction for an important class of problems. The paper also surveys the `generalized linear programming' method for solving such problems when the logarithmic transformation is justified. For the second type model, the constraint is demonstrated to be nonconcave. [ABSTRACT FROM AUTHOR]

This paper deals with a nonstationary discrete-time Markov process whose transition probabilities vary periodically in time. Each transition results in a reward that varies within the same cycle as the transition matrix. For infinite processes a policy-iteration algorithm is developed that effectively determines an optimal policy maximizing the total discounted reward. The paper represents an extension of R. A. HOWARD's policy-iteration technique for stationary Markov processes. A numerical example is given in which the developed iteration algorithm is demonstrated. [ABSTRACT FROM AUTHOR]