Andrew A. Li
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Andrew A. Li has a total of 11 co-authors. Here are the top co-authors by number of shared publications:
- Sridhar Tayur
Quantum Technologies Group, Tepper School of Business, Carnegie Mellon University, Pittsburgh, Pennsylvania 15213
2 shared publications - Ward Whitt
Department of Industrial Engineering and Operations Research, Columbia University, New York, NY 10027-6699, USA
2 shared publications - Alan Scheller-Wolf
Tepper School of Business, Carnegie Mellon University, Pittsburgh, Pennsylvania 15213
1 shared publication - Benjamin Moseley
Carnegie Mellon University, Pittsburgh, PA, USA
1 shared publication - Jingtong Zhao
School of Economics, Renmin University of China, Beijing, China
1 shared publication - Kyra Gan
Operations Research and Information Engineering, Cornell University, New York City, USA
1 shared publication
Latest Publications
- 1 April 2026
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Management Science
- Vol. 72 (4) , 3291-3308
Toward a Liquid Biopsy: Greedy Approximation Algorithms for Active Sequential Hypothesis Testing
This paper addresses challenges in developing liquid biopsies for early-stage cancer detection through active sequential hypothesis testing (ASHT). In the problem of ASHT, a learner seeks to identify the true hypothesis (true cancer type) from a known set of hypotheses (candidate cancer types). The learner is given a set of actions (sequencing genetic intervals) and knows the distribution of the random outcome (whether a mutation is detected) of any action under any true hypothesis. Given a target error [ ], the goal is to sequentially select the fewest number of actions to identify the true hypothesis with probability at least [ ]. Motivated by applications in which the number of hypotheses or actions is massive (e.g., genomics-based cancer detection), we propose efficient greedy algorithms and provide the first approximation guarantees for ASHT, under two types of adaptivity. Our guarantees are independent of the number of actions and logarithmic in the number of hypotheses. Numerical tests on synthetic and real DNA mutation data show that our algorithms significantly outperform previous heuristic policies. This paper was accepted by David Simchi-Levi, healthcare management. Supplemental Material: The online appendices and data files are available at https://doi.org/10.1287/mnsc.2023.00829 .
- 1 April 2026
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Management Science
- Vol. 72 (4) , 3291-3308
Toward a Liquid Biopsy: Greedy Approximation Algorithms for Active Sequential Hypothesis Testing
This paper addresses challenges in developing liquid biopsies for early-stage cancer detection through active sequential hypothesis testing (ASHT). In the problem of ASHT, a learner seeks to identify the true hypothesis (true cancer type) from a known set of hypotheses (candidate cancer types). The learner is given a set of actions (sequencing genetic intervals) and knows the distribution of the random outcome (whether a mutation is detected) of any action under any true hypothesis. Given a target error [ ], the goal is to sequentially select the fewest number of actions to identify the true hypothesis with probability at least [ ]. Motivated by applications in which the number of hypotheses or actions is massive (e.g., genomics-based cancer detection), we propose efficient greedy algorithms and provide the first approximation guarantees for ASHT, under two types of adaptivity. Our guarantees are independent of the number of actions and logarithmic in the number of hypotheses. Numerical tests on synthetic and real DNA mutation data show that our algorithms significantly outperform previous heuristic policies. This paper was accepted by David Simchi-Levi, healthcare management. Supplemental Material: The online appendices and data files are available at https://doi.org/10.1287/mnsc.2023.00829 .
- 1 May 2025
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Manufacturing & Service Operations Management
- Vol. 27 (3) , 881-896
The Nonstationary Newsvendor with (and Without) Predictions
Problem definition: The classic newsvendor model yields an optimal decision for a “newsvendor” selecting a quantity of inventory under the assumption that the demand is drawn from a known distribution. Motivated by applications such as cloud provisioning and staffing, we consider a setting in which newsvendor-type decisions must be made sequentially in the face of demand drawn from a stochastic process that is both unknown and nonstationary. All prior work on this problem either (a) assumes that the level of nonstationarity is known or (b) imposes additional statistical assumptions that enable accurate predictions of the unknown demand. Our research tackles the Nonstationary Newsvendor without these assumptions both with and without predictions. Methodology/results: In the setting without predictions, we first design a policy that we prove (via matching upper and lower bounds) achieves order-optimal regret; ours is the first policy to accomplish this without being given the level of nonstationarity of the underlying demand. We then, for the first time, introduce a model for generic (i.e., with no statistical assumptions) predictions with arbitrary accuracy and propose a policy that incorporates these predictions without being given their accuracy. We upper bound the regret of this policy and show that it matches the best achievable regret had the accuracy of the predictions been known. Managerial implications: Our findings provide valuable insights on inventory management. Managers can make more informed and effective decisions in dynamic environments, reducing costs and enhancing service levels despite uncertain demand patterns. This study advances understanding of sequential decision-making under uncertainty, offering robust methodologies for practical applications with nonstationary demand. We empirically validate our new policy with experiments based on three real-world data sets containing thousands of time-series, showing that it succeeds in closing approximately 74% of the gap between the best approaches based on nonstationarity and predictions alone. History: This paper was selected as part of the 1RR initiative between the M&SOM Journal and the MSOM Society. This particular paper was part of the 2024 MSOM Service Operations SIG Conference. Funding: L. An and B. Moseley were supported in part by a Google Research Award, an Infor Research Award, a Carnegie Bosch Junior Faculty Chair, NSF [Grants CCF-2121744 and CCF-1845146] and ONR [Grant N000142212702]. Supplemental Material: The online appendix is available at https://doi.org/10.1287/msom.2024.1168 .
- 1 May 2025
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Manufacturing & Service Operations Management
- Vol. 27 (3) , 881-896
The Nonstationary Newsvendor with (and Without) Predictions
Problem definition: The classic newsvendor model yields an optimal decision for a “newsvendor” selecting a quantity of inventory under the assumption that the demand is drawn from a known distribution. Motivated by applications such as cloud provisioning and staffing, we consider a setting in which newsvendor-type decisions must be made sequentially in the face of demand drawn from a stochastic process that is both unknown and nonstationary. All prior work on this problem either (a) assumes that the level of nonstationarity is known or (b) imposes additional statistical assumptions that enable accurate predictions of the unknown demand. Our research tackles the Nonstationary Newsvendor without these assumptions both with and without predictions. Methodology/results: In the setting without predictions, we first design a policy that we prove (via matching upper and lower bounds) achieves order-optimal regret; ours is the first policy to accomplish this without being given the level of nonstationarity of the underlying demand. We then, for the first time, introduce a model for generic (i.e., with no statistical assumptions) predictions with arbitrary accuracy and propose a policy that incorporates these predictions without being given their accuracy. We upper bound the regret of this policy and show that it matches the best achievable regret had the accuracy of the predictions been known. Managerial implications: Our findings provide valuable insights on inventory management. Managers can make more informed and effective decisions in dynamic environments, reducing costs and enhancing service levels despite uncertain demand patterns. This study advances understanding of sequential decision-making under uncertainty, offering robust methodologies for practical applications with nonstationary demand. We empirically validate our new policy with experiments based on three real-world data sets containing thousands of time-series, showing that it succeeds in closing approximately 74% of the gap between the best approaches based on nonstationarity and predictions alone. History: This paper was selected as part of the 1RR initiative between the M&SOM Journal and the MSOM Society. This particular paper was part of the 2024 MSOM Service Operations SIG Conference. Funding: L. An and B. Moseley were supported in part by a Google Research Award, an Infor Research Award, a Carnegie Bosch Junior Faculty Chair, NSF [Grants CCF-2121744 and CCF-1845146] and ONR [Grant N000142212702]. Supplemental Material: The online appendix is available at https://doi.org/10.1287/msom.2024.1168 .
- 1 March 2025
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Manufacturing & Service Operations Management
- Vol. 27 (2) , 640-658
Multi-Armed Bandits with Endogenous Learning Curves: An Application to Split Liver Transplantation
Problem Definition: Proficiency in many sophisticated tasks is attained through experience-based learning, in other words, learning by doing. For example, transplant centers’ surgical teams need to practice difficult surgeries to master the skills required. Meanwhile, this experience-based learning may affect other stakeholders, such as patients eligible for transplant surgeries, and require resources, including scarce organs and continual efforts. To ensure that patients have excellent outcomes and equitable access to organs, the organ allocation authority needs to quickly identify and develop medical teams with high aptitudes. This entails striking a balance between exploring surgical combinations with initially unknown full potential and exploiting existing knowledge based on observed outcomes. Methodology/results: We formulate a multi-armed bandit (MAB) model in which parametric learning curves are embedded in the reward functions to capture endogenous experience-based learning. In addition, our model includes provisions ensuring that the choices of arms are subject to fairness constraints to guarantee equity. To solve our MAB problem, we propose the L-UCB and FL-UCB algorithms, variants of the upper confidence bound (UCB) algorithm that attain the optimal [ ] regret on problems enhanced with experience-based learning and fairness concerns. We demonstrate our model and algorithms on the split liver transplantation (SLT) allocation problem, showing that our algorithms have superior numerical performance compared with standard bandit algorithms in a setting where experience-based learning and fairness concerns exist. Managerial implications: From a methodological point of view, our proposed MAB model and algorithms are generic and have broad application prospects. From an application standpoint, our algorithms could be applied to help evaluate potential strategies to increase the proliferation of SLT and other technically difficult procedures. Funding: The authors acknowledge the support of CMU Tepper’s Health Care Initiative Funding. Supplemental Material: The electronic companion is available at https://doi.org/10.1287/msom.2022.0412 .
- 1 March 2025
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Manufacturing & Service Operations Management
- Vol. 27 (2) , 640-658
Multi-Armed Bandits with Endogenous Learning Curves: An Application to Split Liver Transplantation
Problem Definition: Proficiency in many sophisticated tasks is attained through experience-based learning, in other words, learning by doing. For example, transplant centers’ surgical teams need to practice difficult surgeries to master the skills required. Meanwhile, this experience-based learning may affect other stakeholders, such as patients eligible for transplant surgeries, and require resources, including scarce organs and continual efforts. To ensure that patients have excellent outcomes and equitable access to organs, the organ allocation authority needs to quickly identify and develop medical teams with high aptitudes. This entails striking a balance between exploring surgical combinations with initially unknown full potential and exploiting existing knowledge based on observed outcomes. Methodology/results: We formulate a multi-armed bandit (MAB) model in which parametric learning curves are embedded in the reward functions to capture endogenous experience-based learning. In addition, our model includes provisions ensuring that the choices of arms are subject to fairness constraints to guarantee equity. To solve our MAB problem, we propose the L-UCB and FL-UCB algorithms, variants of the upper confidence bound (UCB) algorithm that attain the optimal [ ] regret on problems enhanced with experience-based learning and fairness concerns. We demonstrate our model and algorithms on the split liver transplantation (SLT) allocation problem, showing that our algorithms have superior numerical performance compared with standard bandit algorithms in a setting where experience-based learning and fairness concerns exist. Managerial implications: From a methodological point of view, our proposed MAB model and algorithms are generic and have broad application prospects. From an application standpoint, our algorithms could be applied to help evaluate potential strategies to increase the proliferation of SLT and other technically difficult procedures. Funding: The authors acknowledge the support of CMU Tepper’s Health Care Initiative Funding. Supplemental Material: The electronic companion is available at https://doi.org/10.1287/msom.2022.0412 .
- 1 July 2019
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Management Science
- Vol. 65 (7) , 3131-3149
Learning Preferences with Side Information
Product and content personalization is now ubiquitous in e-commerce. There are typically not enough available transactional data for this task. As such, companies today seek to use a variety of information on the interactions between a product and a customer to drive personalization decisions. We formalize this problem as one of recovering a large-scale matrix with side information in the form of additional matrices of conforming dimension. Viewing the matrix we seek to recover and the side information we have as slices of a tensor, we consider the problem of slice recovery, which is to recover specific slices of “simple” tensors from noisy observations of the entire tensor. We propose a definition of simplicity that on the one hand elegantly generalizes a standard generative model for our motivating problem and on the other hand subsumes low-rank tensors for a variety of existing definitions of tensor rank. We provide an efficient algorithm for slice recovery that is practical for massive data sets and provides a significant performance improvement over state-of-the-art incumbent approaches to tensor recovery. Furthermore, we establish near-optimal recovery guarantees that, in an important regime, represent an order improvement over the best available results for this problem. Experiments on data from a music streaming service demonstrate the performance and scalability of our algorithm. The e-companion is available at https://doi.org/10.1287/mnsc.2018.3092 . This paper was accepted by Noah Gans, stochastic models and simulation.
- 1 July 2019
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in Management Science
- Vol. 65 (7) , 3131-3149
Learning Preferences with Side Information
Product and content personalization is now ubiquitous in e-commerce. There are typically not enough available transactional data for this task. As such, companies today seek to use a variety of information on the interactions between a product and a customer to drive personalization decisions. We formalize this problem as one of recovering a large-scale matrix with side information in the form of additional matrices of conforming dimension. Viewing the matrix we seek to recover and the side information we have as slices of a tensor, we consider the problem of slice recovery, which is to recover specific slices of “simple” tensors from noisy observations of the entire tensor. We propose a definition of simplicity that on the one hand elegantly generalizes a standard generative model for our motivating problem and on the other hand subsumes low-rank tensors for a variety of existing definitions of tensor rank. We provide an efficient algorithm for slice recovery that is practical for massive data sets and provides a significant performance improvement over state-of-the-art incumbent approaches to tensor recovery. Furthermore, we establish near-optimal recovery guarantees that, in an important regime, represent an order improvement over the best available results for this problem. Experiments on data from a music streaming service demonstrate the performance and scalability of our algorithm. The e-companion is available at https://doi.org/10.1287/mnsc.2018.3092 . This paper was accepted by Noah Gans, stochastic models and simulation.
- 9 December 2015
- Published by Cambridge University Press (CUP) in Probability in the Engineering and Informational Sciences
- Vol. 30 (2) , 185-211
STAFFING TO STABILIZE BLOCKING IN LOSS MODELS WITH TIME-VARYING ARRIVAL RATES
The modified-offered-load approximation can be used to choose a staffing function (the time-varying number of servers) to stabilize delay probabilities at target levels in multi-server delay models with time-varying arrival rates, with or without customer abandonment. In contrast, as we confirm with simulations, it is not possible to stabilize blocking probabilities to the same extent in corresponding loss models, without extra waiting space, because these probabilities necessarily change dramatically after each staffing change. Nevertheless, blocking probabilities can be stabilized provided that we either randomize the times of staffing changes or average the blocking probabilities over a suitably small time interval. We develop systematic procedures and study how to choose the averaging parameters.
- 9 December 2015
- Published by Cambridge University Press (CUP) in Probability in the Engineering and Informational Sciences
- Vol. 30 (2) , 185-211
STAFFING TO STABILIZE BLOCKING IN LOSS MODELS WITH TIME-VARYING ARRIVAL RATES
The modified-offered-load approximation can be used to choose a staffing function (the time-varying number of servers) to stabilize delay probabilities at target levels in multi-server delay models with time-varying arrival rates, with or without customer abandonment. In contrast, as we confirm with simulations, it is not possible to stabilize blocking probabilities to the same extent in corresponding loss models, without extra waiting space, because these probabilities necessarily change dramatically after each staffing change. Nevertheless, blocking probabilities can be stabilized provided that we either randomize the times of staffing changes or average the blocking probabilities over a suitably small time interval. We develop systematic procedures and study how to choose the averaging parameters.
- 31 October 2014
- Published by Elsevier in Performance Evaluation
- Vol. 80, 82-101
Approximate blocking probabilities in loss models with independence and distribution assumptions relaxed
- 31 October 2014
- Published by Elsevier in Performance Evaluation
- Vol. 80, 82-101
Approximate blocking probabilities in loss models with independence and distribution assumptions relaxed