Warren B. Powell

Warren B Powell, Saeed Ghadimi

The most common approaches for solving multistage stochastic programming problems in the research literature have been to either use value functions ("dynamic programming") or scenario trees ("stochastic programming") to approximate the impact of a decision now on the future. By contrast, common industry practice is to use a deterministic approximation of the future which is easier to understand and solve, but which is criticized for ignoring uncertainty. We show that a parameterized version of a deterministic optimization model can be an effective way of handling uncertainty without the complexity of either stochastic programming or dynamic programming. We present the idea of a parameterized deterministic optimization model, and in particular a deterministic lookahead model, as a powerful strategy for many complex stochastic decision problems. This approach can handle complex, high-dimensional state variables, and avoids the usual approximations associated with scenario trees or value function approximations. Instead, it introduces the offline challenge of designing and tuning the parameterization. We illustrate the idea by using a series of application settings, and demonstrate its use in a nonstationary energy storage problem with rolling forecasts.

Lawrence Thul, Warren Powell

The pandemic caused by the SARS-CoV-2 virus has exposed many flaws in the decision-making strategies used to distribute resources to combat global health crises. In this paper, we leverage reinforcement learning and optimization to improve upon the allocation strategies for various resources. In particular, we consider a problem where a central controller must decide where to send testing kits to learn about the uncertain states of the world (active learning); then, use the new information to construct beliefs about the states and decide where to allocate resources. We propose a general model coupled with a tunable lookahead policy for making vaccine allocation decisions without perfect knowledge about the state of the world. The lookahead policy is compared to a population-based myopic policy which is more likely to be similar to the present strategies in practice. Each vaccine allocation policy works in conjunction with a testing kit allocation policy to perform active learning. Our simulation results demonstrate that an optimization-based lookahead decision making strategy will outperform the presented myopic policy.

Kristopher Reyes, Warren B Powell

The process of discovery in the physical, biological and medical sciences can be painstakingly slow. Most experiments fail, and the time from initiation of research until a new advance reaches commercial production can span 20 years. This tutorial is aimed to provide experimental scientists with a foundation in the science of making decisions. Using numerical examples drawn from the experiences of the authors, the article describes the fundamental elements of any experimental learning problem. It emphasizes the important role of belief models, which include not only the best estimate of relationships provided by prior research, previous experiments and scientific expertise, but also the uncertainty in these relationships. We introduce the concept of a learning policy, and review the major categories of policies. We then introduce a policy, known as the knowledge gradient, that maximizes the value of information from each experiment. We bring out the importance of reducing uncertainty, and illustrate this process for different belief models.

Warren B Powell

State variables are easily the most subtle dimension of sequential decision problems. This is especially true in the context of active learning problems (bandit problems") where decisions affect what we observe and learn. We describe our canonical framework that models {\it any} sequential decision problem, and present our definition of state variables that allows us to claim: Any properly modeled sequential decision problem is Markovian. We then present a novel two-agent perspective of partially observable Markov decision problems (POMDPs) that allows us to then claim: Any model of a real decision problem is (possibly) non-Markovian. We illustrate these perspectives using the context of observing and treating flu in a population, and provide examples of all four classes of policies in this setting. We close with an indication of how to extend this thinking to multiagent problems.

Dionysios S. Kalogerias, Warren B. Powell

We present $\textit{Free-MESSAGE}^{p}$, the first zeroth-order algorithm for (weakly-)convex mean-semideviation-based risk-aware learning, which is also the first three-level zeroth-order compositional stochastic optimization algorithm whatsoever. Using a non-trivial extension of Nesterov's classical results on Gaussian smoothing, we develop the $\textit{Free-MESSAGE}^{p}$ algorithm from first principles, and show that it essentially solves a smoothed surrogate to the original problem, the former being a uniform approximation of the latter, in a useful, convenient sense. We then present a complete analysis of the $\textit{Free-MESSAGE}^{p}$ algorithm, which establishes convergence in a user-tunable neighborhood of the optimal solutions of the original problem for convex costs, as well as explicit convergence rates for convex, weakly convex, and strongly convex costs, and in a unified way. Orderwise, and for fixed problem parameters, our results demonstrate no sacrifice in convergence speed as compared to existing first-order methods, while striking a certain balance among the condition of the problem, its dimensionality, as well as the accuracy of the obtained results, naturally extending previous results in zeroth-order risk-neutral learning.

Warren B Powell

There are over 15 distinct communities that work in the general area of sequential decisions and information, often referred to as decisions under uncertainty or stochastic optimization. We focus on two of the most important fields: stochastic optimal control, with its roots in deterministic optimal control, and reinforcement learning, with its roots in Markov decision processes. Building on prior work, we describe a unified framework that covers all 15 different communities, and note the strong parallels with the modeling framework of stochastic optimal control. By contrast, we make the case that the modeling framework of reinforcement learning, inherited from discrete Markov decision processes, is quite limited. Our framework (and that of stochastic control) is based on the core problem of optimizing over policies. We describe four classes of policies that we claim are universal, and show that each of these two fields have, in their own way, evolved to include examples of each of these four classes.

Lina Al-Kanj, Juliana Nascimento, Warren B. Powell

Within a decade, almost every major auto company, along with fleet operators such as Uber, have announced plans to put autonomous vehicles on the road. At the same time, electric vehicles are quickly emerging as a next-generation technology that is cost effective, in addition to offering the benefits of reducing the carbon footprint. The combination of a centrally managed fleet of driverless vehicles, along with the operating characteristics of electric vehicles, is creating a transformative new technology that offers significant cost savings with high service levels. This problem involves a dispatch problem for assigning riders to cars, a surge pricing problem for deciding on the price per trip and a planning problem for deciding on the fleet size. We use approximate dynamic programming to develop high-quality operational dispatch strategies to determine which car is best for a particular trip, when a car should be recharged, and when it should be re-positioned to a different zone which offers a higher density of trips. We prove that the value functions are monotone in the battery and time dimensions and use hierarchical aggregation to get better estimates of the value functions with a small number of observations. Then, surge pricing is discussed using an adaptive learning approach to decide on the price for each trip. Finally, we discuss the fleet size problem which depends on the previous two problems.

Dionysios S. Kalogerias, Warren B. Powell

We develop recursive, data-driven, stochastic subgradient methods for optimizing a new, versatile, and application-driven class of convex risk measures, termed here as mean-semideviations, strictly generalizing the well-known and popular mean-upper-semideviation. We introduce the MESSAGEp algorithm, which is an efficient compositional subgradient procedure for iteratively solving convex mean-semideviation risk-averse problems to optimality. We analyze the asymptotic behavior of the MESSAGEp algorithm under a flexible and structure-exploiting set of problem assumptions. In particular: 1) Under appropriate stepsize rules, we establish pathwise convergence of the MESSAGEp algorithm in a strong technical sense, confirming its asymptotic consistency. 2) Assuming a strongly convex cost, we show that, for fixed semideviation order $p>1$ and for $ε\in\left[0,1\right)$, the MESSAGEp algorithm achieves a squared-${\cal L}_{2}$ solution suboptimality rate of the order of ${\cal O}(n^{-\left(1-ε\right)/2})$ iterations, where, for $ε>0$, pathwise convergence is simultaneously guaranteed. This result establishes a rate of order arbitrarily close to ${\cal O}(n^{-1/2})$, while ensuring strongly stable pathwise operation. For $p\equiv1$, the rate order improves to ${\cal O}(n^{-2/3})$, which also suffices for pathwise convergence, and matches previous results. 3) Likewise, in the general case of a convex cost, we show that, for any $ε\in\left[0,1\right)$, the MESSAGEp algorithm with iterate smoothing achieves an ${\cal L}_{1}$ objective suboptimality rate of the order of ${\cal O}(n^{-\left(1-ε\right)/\left(4\bf{1}_{\left\{ p>1\right\} }+4\right)})$ iterations. This result provides maximal rates of ${\cal O}(n^{-1/4})$, if $p\equiv1$, and ${\cal O}(n^{-1/8})$, if $p>1$, matching the state of the art, as well.

Yingfei Wang, Juliana Martins Do Nascimento, Warren Powell

Truckload brokerages, a $100 billion/year industry in the U.S., plays the critical role of matching shippers with carriers, often to move loads several days into the future. Brokerages not only have to find companies that will agree to move a load, the brokerage often has to find a price that both the shipper and carrier will agree to. The price not only varies by shipper and carrier, but also by the traffic lanes and other variables such as commodity type. Brokerages have to learn about shipper and carrier response functions by offering a price and observing whether each accepts the quote. We propose a knowledge gradient policy with bootstrap aggregation for high-dimensional contextual settings to guide price experimentation by maximizing the value of information. The learning policy is tested using a carefully calibrated fleet simulator that includes a stochastic lookahead policy that simulates fleet movements, as well as the stochastic modeling of driver assignments and the carrier's load commitment policies with advance booking.

Yingfei Wang, Warren Powell

We address the relative paucity of empirical testing of learning algorithms (of any type) by introducing a new public-domain, Modular, Optimal Learning Testing Environment (MOLTE) for Bayesian ranking and selection problem, stochastic bandits or sequential experimental design problems. The Matlab-based simulator allows the comparison of a number of learning policies (represented as a series of .m modules) in the context of a wide range of problems (each represented in its own .m module) which makes it easy to add new algorithms and new test problems. State-of-the-art policies and various problem classes are provided in the package. The choice of problems and policies is guided through a spreadsheet-based interface. Different graphical metrics are included. MOLTE is designed to be compatible with parallel computing to scale up from local desktop to clusters and clouds. We offer MOLTE as an easy-to-use tool for the research community that will make it possible to perform much more comprehensive testing, spanning a broader selection of algorithms and test problems. We demonstrate the capabilities of MOLTE through a series of comparisons of policies on a starter library of test problems. We also address the problem of tuning and constructing priors that have been largely overlooked in optimal learning literature. We envision MOLTE as a modest spur to provide researchers an easy environment to study interesting questions involved in optimal learning.

Yingfei Wang, Chu Wang, Warren Powell

We consider the problem of sequentially making decisions that are rewarded by "successes" and "failures" which can be predicted through an unknown relationship that depends on a partially controllable vector of attributes for each instance. The learner takes an active role in selecting samples from the instance pool. The goal is to maximize the probability of success in either offline (training) or online (testing) phases. Our problem is motivated by real-world applications where observations are time-consuming and/or expensive. We develop a knowledge gradient policy using an online Bayesian linear classifier to guide the experiment by maximizing the expected value of information of labeling each alternative. We provide a finite-time analysis of the estimated error and show that the maximum likelihood estimator based produced by the KG policy is consistent and asymptotically normal. We also show that the knowledge gradient policy is asymptotically optimal in an offline setting. This work further extends the knowledge gradient to the setting of contextual bandits. We report the results of a series of experiments that demonstrate its efficiency.

Daniel R. Jiang, Lina Al-Kanj, Warren B. Powell

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 towards 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 technique called Primal-Dual MCTS that utilizes sampled information relaxation upper bounds on potential actions, creating the possibility of "ignoring" parts of the tree that stem from highly suboptimal choices. This allows us 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 significant 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 dataset of 7,000 trips in New Jersey suggest that Primal-Dual MCTS improves upon standard MCTS by producing deeper decision trees and exhibits a reduced sensitivity to the size of the action space.

Xinyu He, Warren B. Powell

We consider the problem of estimating the expected value of information (the knowledge gradient) for Bayesian learning problems where the belief model is nonlinear in the parameters. Our goal is to maximize some metric, while simultaneously learning the unknown parameters of the nonlinear belief model, by guiding a sequential experimentation process which is expensive. We overcome the problem of computing the expected value of an experiment, which is computationally intractable, by using a sampled approximation, which helps to guide experiments but does not provide an accurate estimate of the unknown parameters. We then introduce a resampling process which allows the sampled model to adapt to new information, exploiting past experiments. We show theoretically that the method converges asymptotically to the true parameters, while simultaneously maximizing our metric. We show empirically that the process exhibits rapid convergence, yielding good results with a very small number of experiments.

Yingfei Wang, Warren Powell

A treatment regime is a function that maps individual patient information to a recommended treatment, hence explicitly incorporating the heterogeneity in need for treatment across individuals. Patient responses are dichotomous and can be predicted through an unknown relationship that depends on the patient information and the selected treatment. The goal is to find the treatments that lead to the best patient responses on average. Each experiment is expensive, forcing us to learn the most from each experiment. We adopt a Bayesian approach both to incorporate possible prior information and to update our treatment regime continuously as information accrues, with the potential to allow smaller yet more informative trials and for patients to receive better treatment. By formulating the problem as contextual bandits, we introduce a knowledge gradient policy to guide the treatment assignment by maximizing the expected value of information, for which an approximation method is used to overcome computational challenges. We provide a detailed study on how to make sequential medical decisions under uncertainty to reduce health care costs on a real world knee replacement dataset. We use clustering and LASSO to deal with the intrinsic sparsity in health datasets. We show experimentally that even though the problem is sparse, through careful selection of physicians (versus picking them at random), we can significantly improve the success rates.

Yingfei Wang, Warren Powell

We consider sequential decision problems in which we adaptively choose one of finitely many alternatives and observe a stochastic reward. We offer a new perspective of interpreting Bayesian ranking and selection problems as adaptive stochastic multi-set maximization problems and derive the first finite-time bound of the knowledge-gradient policy for adaptive submodular objective functions. In addition, we introduce the concept of prior-optimality and provide another insight into the performance of the knowledge gradient policy based on the submodular assumption on the value of information. We demonstrate submodularity for the two-alternative case and provide other conditions for more general problems, bringing out the issue and importance of submodularity in learning problems. Empirical experiments are conducted to further illustrate the finite time behavior of the knowledge gradient policy.

Lina Al-Kanj, Warren B. Powell, Belgacem Bouzaiene-Ayari

Utilities face the challenge of responding to power outages due to storms and ice damage, but most power grids are not equipped with sensors to pinpoint the precise location of the faults causing the outage. Instead, utilities have to depend primarily on phone calls (trouble calls) from customers who have lost power to guide the dispatching of utility trucks. In this paper, we develop a policy that routes a utility truck to restore outages in the power grid as quickly as possible, using phone calls to create beliefs about outages, but also using utility trucks as a mechanism for collecting additional information. This means that routing decisions change not only the physical state of the truck (as it moves from one location to another) and the grid (as the truck performs repairs), but also our belief about the network, creating the first stochastic vehicle routing problem that explicitly models information collection and belief modeling. We address the problem of managing a single utility truck, which we start by formulating as a sequential stochastic optimization model which captures our belief about the state of the grid. We propose a stochastic lookahead policy, and use Monte Carlo tree search (MCTS) to produce a practical policy that is asymptotically optimal. Simulation results show that the developed policy restores the power grid much faster compared to standard industry heuristics.

Yingfei Wang, Chu Wang, Warren Powell

We consider sequential decision making problems for binary classification scenario in which the learner takes an active role in repeatedly selecting samples from the action pool and receives the binary label of the selected alternatives. Our problem is motivated by applications where observations are time consuming and/or expensive, resulting in small samples. The goal is to identify the best alternative with the highest response. We use Bayesian logistic regression to predict the response of each alternative. By formulating the problem as a Markov decision process, we develop a knowledge-gradient type policy to guide the experiment by maximizing the expected value of information of labeling each alternative and provide a finite-time analysis on the estimated error. Experiments on benchmark UCI datasets demonstrate the effectiveness of the proposed method.

Daniel R. Jiang, Warren B. Powell

In this paper, we consider a finite-horizon Markov decision process (MDP) for which the objective at each stage is to minimize a quantile-based risk measure (QBRM) of the sequence of future costs; we call the overall objective a dynamic quantile-based risk measure (DQBRM). In particular, we consider optimizing dynamic risk measures where the one-step risk measures are QBRMs, a class of risk measures that includes the popular value at risk (VaR) and the conditional value at risk (CVaR). Although there is considerable theoretical development of risk-averse MDPs in the literature, the computational challenges have not been explored as thoroughly. We propose data-driven and simulation-based approximate dynamic programming (ADP) algorithms to solve the risk-averse sequential decision problem. We address the issue of inefficient sampling for risk applications in simulated settings and present a procedure, based on importance sampling, to direct samples toward the "risky region" as the ADP algorithm progresses. Finally, we show numerical results of our algorithms in the context of an application involving risk-averse bidding for energy storage.

Yan Li, Kristofer G. Reyes, Jorge Vazquez-Anderson et al.

We present a sparse knowledge gradient (SpKG) algorithm for adaptively selecting the targeted regions within a large RNA molecule to identify which regions are most amenable to interactions with other molecules. Experimentally, such regions can be inferred from fluorescence measurements obtained by binding a complementary probe with fluorescence markers to the targeted regions. We use a biophysical model which shows that the fluorescence ratio under the log scale has a sparse linear relationship with the coefficients describing the accessibility of each nucleotide, since not all sites are accessible (due to the folding of the molecule). The SpKG algorithm uniquely combines the Bayesian ranking and selection problem with the frequentist $\ell_1$ regularized regression approach Lasso. We use this algorithm to identify the sparsity pattern of the linear model as well as sequentially decide the best regions to test before experimental budget is exhausted. Besides, we also develop two other new algorithms: batch SpKG algorithm, which generates more suggestions sequentially to run parallel experiments; and batch SpKG with a procedure which we call length mutagenesis. It dynamically adds in new alternatives, in the form of types of probes, are created by inserting, deleting or mutating nucleotides within existing probes. In simulation, we demonstrate these algorithms on the Group I intron (a mid-size RNA molecule), showing that they efficiently learn the correct sparsity pattern, identify the most accessible region, and outperform several other policies.

Yan Li, Han Liu, Warren Powell

We propose a sequential learning policy for noisy discrete global optimization and ranking and selection (R\&S) problems with high dimensional sparse belief functions, where there are hundreds or even thousands of features, but only a small portion of these features contain explanatory power. We aim to identify the sparsity pattern and select the best alternative before the finite budget is exhausted. We derive a knowledge gradient policy for sparse linear models (KGSpLin) with group Lasso penalty. This policy is a unique and novel hybrid of Bayesian R\&S with frequentist learning. Particularly, our method naturally combines B-spline basis expansion and generalizes to the nonparametric additive model (KGSpAM) and functional ANOVA model. Theoretically, we provide the estimation error bounds of the posterior mean estimate and the functional estimate. Controlled experiments show that the algorithm efficiently learns the correct set of nonzero parameters even when the model is imbedded with hundreds of dummy parameters. Also it outperforms the knowledge gradient for a linear model.

Ilya O. Ryzhov, Peter I. Frazier, Warren B. Powell

Approximate dynamic programming (ADP) has proven itself in a wide range of applications spanning large-scale transportation problems, health care, revenue management, and energy systems. The design of effective ADP algorithms has many dimensions, but one crucial factor is the stepsize rule used to update a value function approximation. Many operations research applications are computationally intensive, and it is important to obtain good results quickly. Furthermore, the most popular stepsize formulas use tunable parameters and can produce very poor results if tuned improperly. We derive a new stepsize rule that optimizes the prediction error in order to improve the short-term performance of an ADP algorithm. With only one, relatively insensitive tunable parameter, the new rule adapts to the level of noise in the problem and produces faster convergence in numerical experiments.

Warren R. Scott, Warren B. Powell, Somayeh Moazehi

This paper studies approximate policy iteration (API) methods which use least-squares Bellman error minimization for policy evaluation. We address several of its enhancements, namely, Bellman error minimization using instrumental variables, least-squares projected Bellman error minimization, and projected Bellman error minimization using instrumental variables. We prove that for a general discrete-time stochastic control problem, Bellman error minimization using instrumental variables is equivalent to both variants of projected Bellman error minimization. An alternative to these API methods is direct policy search based on knowledge gradient. The practical performance of these three approximate dynamic programming methods are then investigated in the context of an application in energy storage, integrated with an intermittent wind energy supply to fully serve a stochastic time-varying electricity demand. We create a library of test problems using real-world data and apply value iteration to find their optimal policies. These benchmarks are then used to compare the developed policies. Our analysis indicates that API with instrumental variables Bellman error minimization prominently outperforms API with least-squares Bellman error minimization. However, these approaches underperform our direct policy search implementation.