Publications

Papers are grouped by research area so they are easier to find. A few belong to two areas and are listed under both. For everything in date order, including patents and older conference papers, use the chronological view.

By topicBy year

Recent · AI alignment and oversight · Language models and agents · Experimentation and causal inference · Bandits and online decisions · Health care · Message passing and probability

Recent

Everything from 2025 onward, newest first. These also appear under their research areas below.

  1. Annealed Softmax Greedy in Many-Armed Bayesian Bandits W. Overman, M. Bayati · Reinforcement Learning Conference (RLC) 2026
    An uncertainty-agnostic annealed softmax rule that is near optimal in many-armed Bayesian bandits, with a connection to reinforcement learning from verifiable rewards.
  2. A Few Teacher Steps Go a Long Way: Cost-Efficient On-Policy Data Augmentation for Agent Post-Training J. Ye, J. Cheng, M. Lu, M. Mankowski, J. Blanchet, M. Bayati · Preprint
    Where to spend a limited budget of teacher supervision when post-training agents.
  3. Calibrating Conservatism for Scalable Oversight W. Overman, M. Bayati · ICML 2026
    Aggregating several scoring functions into a single conservatism penalty, then using conformal decision theory to hold undesirable outcomes below a chosen threshold.
  4. The Oversight Game: Learning to Cooperatively Balance an AI Agent's Safety and Autonomy W. Overman, M. Bayati · ICML 2026
    A tuning procedure that trains an AI agent and its human overseer to work together, built by setting their interaction up as a game.
  5. Speed Up the Cold-Start Learning in Two-Sided Bandits with Many Arms M. Bayati, J. Cao, W. Chen · Management Science
    Cutting the cold-start cost of matching markets by exploiting the structure of a two-sided arm pool.
  6. Text-to-Distribution Prediction with Quantile Tokens and Neighbor Context Y. Zhu, Y. Zhuang, N. Vedula, D. Dhyani, S. Xu, M. Li, M. Bayati, B. Wang, S. Malmasi · Association for Computational Linguistics (ACL) 2026
    Quantile tokens plus retrieved neighbor context, so a language model returns a calibrated distribution rather than a point estimate.
  7. Causal Effects with Unobserved Unit Types in Interacting Human-AI Systems W. Overman, S. Shirani, M. Bayati · Preprint
    Estimating treatment effects when a population mixes humans and AI agents whose types are not observed.
  8. Validating Causal Message Passing Against Network-Aware Methods on Real Experiments A. Tan, S. Shirani, J. Nordlund, M. Bayati · Preprint
    A head-to-head evaluation of network-agnostic and network-aware interference estimators on real platform experiments.
  9. Conformal Arbitrage: Risk-Controlled Balancing of Competing Objectives in Language Models W. Overman, M. Bayati · NeurIPS 2025
    Routing between a cheap and an expensive model so that a stated risk budget is respected with finite-sample guarantees.
  10. Scalable Stewardship of an LLM-Assisted Clinical Benchmark with Physician Oversight J. Ye, D. Tawfik, A. J. Goodell, N. V. Kotha, M. K. Buyyounouski, M. Bayati · Preprint
    Keeping a clinical benchmark trustworthy by routing a small physician-review budget to the labels most likely to be wrong. The label errors this surfaced led to a correction on Anthropic's healthcare page.
    Finalist, INFORMS Health Applications Society Student Paper Competition, 2026Code and data
  11. Estimating Total Effects in Bipartite Experiments with Spillovers and Partial Eligibility A. Tan, M. Bayati, J. Nordlund, R. Istomin · Preprint
    Marketplace experiments where treatment is applied to one side and only some units on the other side are eligible.
    Oral, CODE 2025
  12. On Evolution-Based Models for Experimentation Under Interference S. Shirani, M. Bayati · Preprint
    What class of outcome dynamics makes population-level interference estimators provably correct.
  13. Simulating and Experimenting with Social Media Mobilization Using LLM Agents S. Shirani, M. Bayati · Preprint
    Rebuilding a 61-million-person Facebook mobilization experiment with LLM agents, then using it as a testbed for interference estimators.
    Code and data
  14. On Aligning Prediction Models with Clinical Experiential Learning: A Prostate Cancer Case Study J. J. Vallon, W. Overman, W. Xu, N. Panjwani, X. Ling, S. Vij, H. P. Bagshaw, J. T. Leppert, S. Shah, G. Sonn, S. Srinivas, E. Pollom, M. K. Buyyounouski, M. Bayati · Preprint
    Making a recurrence-risk model obey the monotonicity relationships oncologists learn from experience, without giving up accuracy.
  15. Scaling Clinician-Grade Feature Generation from Clinical Notes with Multi-Agent Language Models J. Wang, J. J. Vallon, N. V. Kotha, N. Panjwani, X. Ling, M. Redfield, S. Vij, S. Srinivas, J. Leppert, M. K. Buyyounouski, M. Bayati · Preprint
    A multi-agent pipeline that reads clinical notes and produces the structured variables a physician would have abstracted by hand.
    Project page
  16. Quantile Regression with Large Language Models for Price Prediction N. Vedula, D. Dhyani, L. Jalali, B. Oreshkin, M. Bayati, S. Malmasi · Association for Computational Linguistics (ACL) 2025
    Getting a calibrated predictive distribution out of a language model asked to predict a price, with the uncertainty attached.
  17. Geometry-Aware Approaches for Balancing Performance and Theoretical Guarantees in Linear Bandits Y. Luo, M. Bayati · International Conference on Learning Representations (ICLR) 2025
  18. Qualitative verification of machine learning-based burnout predictors in primary care physicians D. Tawfik, S. S. Sebok-Syer, C. Bragdon, C. Brown-Johnson, M. Winget, M. Bayati, T. Shanafelt, J. Profit · Applied Clinical Informatics 16(4), 1031–1040
  19. Can We Validate Counterfactual Estimations in the Presence of General Network Interference? S. Shirani, Y. Luo, W. Overman, R. Xiong, M. Bayati · Preprint
    Second place, INFORMS RMP Jeff McGill Student Paper Award, 2025Honorable mention, INFORMS HAS Best Student Paper Competition, 2025CodeOral presentation, CODE 2025
  20. Post-Launch Evaluation of Policies in a High-Dimensional Setting S. Nassiri, M. Bayati, J. Cooprider · Preprint
    CODE 2025
  21. Electronic health record use patterns among well-being survey responders and non-responders: Longitudinal observational study D. Tawfik, T. Shanafelt, M. Bayati, J. Profit · JMIR Medical Informatics 13, e64722

AI alignment, oversight, and safe deployment

These papers ask what it takes to make a deployed AI system satisfy the properties its users actually need, and how a person and an AI agent can be tuned to work well together when the agent is usually right and occasionally badly wrong. The methods are conformal risk control and statistical decision theory, so the guarantees hold at deployment time and do not require access to the model's internals.

  1. Calibrating Conservatism for Scalable Oversight W. Overman, M. Bayati · ICML 2026
    Aggregating several scoring functions into a single conservatism penalty, then using conformal decision theory to hold undesirable outcomes below a chosen threshold.
  2. The Oversight Game: Learning to Cooperatively Balance an AI Agent's Safety and Autonomy W. Overman, M. Bayati · ICML 2026
    A tuning procedure that trains an AI agent and its human overseer to work together, built by setting their interaction up as a game.
  3. Causal Effects with Unobserved Unit Types in Interacting Human-AI Systems W. Overman, S. Shirani, M. Bayati · Preprint
    Estimating treatment effects when a population mixes humans and AI agents whose types are not observed.
  4. Conformal Arbitrage: Risk-Controlled Balancing of Competing Objectives in Language Models W. Overman, M. Bayati · NeurIPS 2025
    Routing between a cheap and an expensive model so that a stated risk budget is respected with finite-sample guarantees.
  5. On Aligning Prediction Models with Clinical Experiential Learning: A Prostate Cancer Case Study J. J. Vallon, W. Overman, W. Xu, N. Panjwani, X. Ling, S. Vij, H. P. Bagshaw, J. T. Leppert, S. Shah, G. Sonn, S. Srinivas, E. Pollom, M. K. Buyyounouski, M. Bayati · Preprint
    Making a recurrence-risk model obey the monotonicity relationships oncologists learn from experience, without giving up accuracy.
  6. Aligning Model Properties via Conformal Risk Control W. Overman, J. Vallon, M. Bayati · NeurIPS 2024
    Post-hoc adjustment that forces a trained model to satisfy properties like monotonicity, with a risk guarantee.
  7. A Probabilistic Approach for Model Alignment with Human Comparisons J. Cao and M. Bayati · Preprint
    When is it worth spending human comparison labels, and how many are needed, in a two-stage fine-tuning pipeline.
  8. Clinically Consistent Prostate Cancer Outcome Prediction Models with Machine Learning J. J. Vallon, N. Panjwani, X. Ling, S. Vij, E. Pollom, H. P. Bagshaw, S. Srinivas, J. Leppert, M. Bayati, M. K. Buyyounouski · American Society for Radiation Oncology (ASTRO) 2022

Language models and agentic systems

Work on what large language models can do when they are asked to produce something a domain expert would sign off on: a medication direction, a clinical feature extracted from a note, a price, a benchmark label. Each project pairs an agentic pipeline with a measurement design that says how often the system is right and where it fails, because in these settings the error rate is an important part of the product.

  1. A Few Teacher Steps Go a Long Way: Cost-Efficient On-Policy Data Augmentation for Agent Post-Training J. Ye, J. Cheng, M. Lu, M. Mankowski, J. Blanchet, M. Bayati · Preprint
    Where to spend a limited budget of teacher supervision when post-training agents.
  2. Text-to-Distribution Prediction with Quantile Tokens and Neighbor Context Y. Zhu, Y. Zhuang, N. Vedula, D. Dhyani, S. Xu, M. Li, M. Bayati, B. Wang, S. Malmasi · Association for Computational Linguistics (ACL) 2026
    Quantile tokens plus retrieved neighbor context, so a language model returns a calibrated distribution rather than a point estimate.
  3. Conformal Arbitrage: Risk-Controlled Balancing of Competing Objectives in Language Models W. Overman, M. Bayati · NeurIPS 2025
    Routing between a cheap and an expensive model so that a stated risk budget is respected with finite-sample guarantees.
  4. Scalable Stewardship of an LLM-Assisted Clinical Benchmark with Physician Oversight J. Ye, D. Tawfik, A. J. Goodell, N. V. Kotha, M. K. Buyyounouski, M. Bayati · Preprint
    Keeping a clinical benchmark trustworthy by routing a small physician-review budget to the labels most likely to be wrong. The label errors this surfaced led to a correction on Anthropic's healthcare page.
    Finalist, INFORMS Health Applications Society Student Paper Competition, 2026Code and data
  5. Simulating and Experimenting with Social Media Mobilization Using LLM Agents S. Shirani, M. Bayati · Preprint
    Rebuilding a 61-million-person Facebook mobilization experiment with LLM agents, then using it as a testbed for interference estimators.
    Code and data
  6. Scaling Clinician-Grade Feature Generation from Clinical Notes with Multi-Agent Language Models J. Wang, J. J. Vallon, N. V. Kotha, N. Panjwani, X. Ling, M. Redfield, S. Vij, S. Srinivas, J. Leppert, M. K. Buyyounouski, M. Bayati · Preprint
    A multi-agent pipeline that reads clinical notes and produces the structured variables a physician would have abstracted by hand.
    Project page
  7. Quantile Regression with Large Language Models for Price Prediction N. Vedula, D. Dhyani, L. Jalali, B. Oreshkin, M. Bayati, S. Malmasi · Association for Computational Linguistics (ACL) 2025
    Getting a calibrated predictive distribution out of a language model asked to predict a price, with the uncertainty attached.
  8. Large language models for preventing medication direction errors in online pharmacies C. Pais, J. Liu, R. Voigt, V. Gupta, E. Wade, M. Bayati · Nature Medicine 30, 1574–1582
    A fine-tuned language model that catches erroneous medication directions before they reach a patient, evaluated against pharmacist judgment.
    INFORMS Health Applications Pierskalla runner-up paper award, 2024Story: An AI copilot for prescriptions

Experimentation and causal inference

Randomized experiments on platforms and in health systems break the usual assumption that units do not affect each other. This line of work develops estimators that stay valid when interference is present but its network is unknown, along with panel-data and design tools for staggered rollouts. The central idea, causal message passing, borrows from statistical physics: follow how the distribution of outcomes across the population moves over time, which works without recovering the network.

  1. Causal Effects with Unobserved Unit Types in Interacting Human-AI Systems W. Overman, S. Shirani, M. Bayati · Preprint
    Estimating treatment effects when a population mixes humans and AI agents whose types are not observed.
  2. Validating Causal Message Passing Against Network-Aware Methods on Real Experiments A. Tan, S. Shirani, J. Nordlund, M. Bayati · Preprint
    A head-to-head evaluation of network-agnostic and network-aware interference estimators on real platform experiments.
  3. Estimating Total Effects in Bipartite Experiments with Spillovers and Partial Eligibility A. Tan, M. Bayati, J. Nordlund, R. Istomin · Preprint
    Marketplace experiments where treatment is applied to one side and only some units on the other side are eligible.
    Oral, CODE 2025
  4. On Evolution-Based Models for Experimentation Under Interference S. Shirani, M. Bayati · Preprint
    What class of outcome dynamics makes population-level interference estimators provably correct.
  5. Simulating and Experimenting with Social Media Mobilization Using LLM Agents S. Shirani, M. Bayati · Preprint
    Rebuilding a 61-million-person Facebook mobilization experiment with LLM agents, then using it as a testbed for interference estimators.
    Code and data
  6. Can We Validate Counterfactual Estimations in the Presence of General Network Interference? S. Shirani, Y. Luo, W. Overman, R. Xiong, M. Bayati · Preprint
    Second place, INFORMS RMP Jeff McGill Student Paper Award, 2025Honorable mention, INFORMS HAS Best Student Paper Competition, 2025CodeOral presentation, CODE 2025
  7. Post-Launch Evaluation of Policies in a High-Dimensional Setting S. Nassiri, M. Bayati, J. Cooprider · Preprint
    CODE 2025
  8. Higher-Order Causal Message Passing for Experimentation with Complex Interference M. Bayati, Y. Luo, W. Overman, S. Shirani, R. Xiong · NeurIPS 2024
  9. Causal Message Passing for Experiments with Unknown and General Network Interference S. Shirani, M. Bayati · Proceedings of the National Academy of Sciences 121(40)
    An estimator for total treatment effects when the interference network is unknown, built by tracking population-level outcome dynamics.
    Honorable mention, 2024 Nicholson Student Paper CompetitionFinalist, 2024 MSOM Student Paper CompetitionCodeOral presentation, CODE 2024
  10. Optimal Experimental Design for Staggered Rollouts R. Xiong, S. Athey, M. Bayati, G. Imbens · Management Science 70(8), 5317–5336
    Choosing when each unit enters treatment, when the experiment runs over many periods and the design can adapt.
    Finalist, 2020 MSOM Student Paper CompetitionCode
  11. Matrix Completion Methods for Causal Panel Data Models S. Athey, M. Bayati, N. Doudchenko, G. Imbens, K. Khosravi · Journal of the American Statistical Association 116(536)
    Recasting synthetic control and difference-in-differences as one matrix completion problem, and estimating it that way.
    R package
  12. Ensemble Methods for Causal Effects in Panel Data Settings S. Athey, M. Bayati, G. Imbens, Z. Qu · AEA Papers and Proceedings 109

Bandits, online learning, and personalized decisions

How much deliberate experimentation does a decision-maker actually need? These papers give conditions under which greedy, exploration-free policies are near optimal, sharpen the regret theory for linear and Thompson sampling algorithms, and handle the high-dimensional and many-armed regimes that appear when decisions are personalized.

  1. Annealed Softmax Greedy in Many-Armed Bayesian Bandits W. Overman, M. Bayati · Reinforcement Learning Conference (RLC) 2026
    An uncertainty-agnostic annealed softmax rule that is near optimal in many-armed Bayesian bandits, with a connection to reinforcement learning from verifiable rewards.
  2. Speed Up the Cold-Start Learning in Two-Sided Bandits with Many Arms M. Bayati, J. Cao, W. Chen · Management Science
    Cutting the cold-start cost of matching markets by exploiting the structure of a two-sided arm pool.
  3. Geometry-Aware Approaches for Balancing Performance and Theoretical Guarantees in Linear Bandits Y. Luo, M. Bayati · International Conference on Learning Representations (ICLR) 2025
  4. Technical Note — The Elliptical Potential Lemma for General Distributions with an Application to Linear Thompson Sampling N. Hamidi, M. Bayati · Operations Research 71(4), 1434–1439
  5. Thompson Sampling Efficiently Learns to Control Diffusion Processes M. K. Shirani Faradonbeh, M. S. Shirani Faradonbeh, M. Bayati · NeurIPS 2022
  6. Learning to Recommend Using Non-Uniform Data W. Chen and M. Bayati · Preprint
    Recommendation when the observed ratings are selected by users rather than sampled at random.
  7. Mostly Exploration-Free Algorithms for Contextual Bandits H. Bastani, M. Bayati, K. Khosravi · Management Science 67(3)
    Natural variation in the covariates supplies enough exploration that a greedy policy is rate optimal.
    CodeStory
  8. The Unreasonable Effectiveness of Greedy Algorithms in Multi-Armed Bandit with Many Arms M. Bayati, N. Hamidi, R. Johari, K. Khosravi · NeurIPS 2020, spotlight
    With many arms, free exploration from the arm pool itself makes a purely greedy policy near optimal.
    CodeStory: When “Greedy” Is Good
  9. On Frequentist Regret of Linear Thompson Sampling N. Hamidi and M. Bayati · Preprint
    Settles an open problem stated in the Lattimore-Szepesvari bandit monograph
  10. Recommendation on a Budget: Column Space Recovery from Partially Observed Entries with Random or Active Sampling C. Kim, M. Bayati · AISTATS 2020
  11. A General Theory of the Stochastic Linear Bandit and Its Applications N. Hamidi and M. Bayati · Preprint
    A unified regret analysis covering optimism, Thompson sampling, and greedy policies in one framework.
  12. Online Decision-Making with High-Dimensional Covariates H. Bastani, M. Bayati · Operations Research 68(1)
    Personalized decisions when the number of candidate covariates far exceeds the number of observations.
    INFORMS Health Applications Pierskalla best paper award, 2016Nicholson, MSOM, and IBM Service Science best student paper, 2016
  13. Personalizing Many Decisions with High-Dimensional Covariates N. Hamidi, M. Bayati, K. Gupta · NeurIPS 2019
  14. Dynamic Pricing with Demand Covariates S. Qiang, M. Bayati · INFORMS Revenue Management and Pricing 2016

AI and analytics in health care

Work with clinicians since 2009 on problems where a better prediction changes a high-stakes decision: readmission risk, prostate cancer recurrence, emergency department wait times, drug safety surveillance, and the effect of electronic health record burden on physician burnout. Several of these models were deployed in operating hospitals and pharmacies.

  1. Scalable Stewardship of an LLM-Assisted Clinical Benchmark with Physician Oversight J. Ye, D. Tawfik, A. J. Goodell, N. V. Kotha, M. K. Buyyounouski, M. Bayati · Preprint
    Keeping a clinical benchmark trustworthy by routing a small physician-review budget to the labels most likely to be wrong. The label errors this surfaced led to a correction on Anthropic's healthcare page.
    Finalist, INFORMS Health Applications Society Student Paper Competition, 2026Code and data
  2. On Aligning Prediction Models with Clinical Experiential Learning: A Prostate Cancer Case Study J. J. Vallon, W. Overman, W. Xu, N. Panjwani, X. Ling, S. Vij, H. P. Bagshaw, J. T. Leppert, S. Shah, G. Sonn, S. Srinivas, E. Pollom, M. K. Buyyounouski, M. Bayati · Preprint
    Making a recurrence-risk model obey the monotonicity relationships oncologists learn from experience, without giving up accuracy.
  3. Scaling Clinician-Grade Feature Generation from Clinical Notes with Multi-Agent Language Models J. Wang, J. J. Vallon, N. V. Kotha, N. Panjwani, X. Ling, M. Redfield, S. Vij, S. Srinivas, J. Leppert, M. K. Buyyounouski, M. Bayati · Preprint
    A multi-agent pipeline that reads clinical notes and produces the structured variables a physician would have abstracted by hand.
    Project page
  4. Qualitative verification of machine learning-based burnout predictors in primary care physicians D. Tawfik, S. S. Sebok-Syer, C. Bragdon, C. Brown-Johnson, M. Winget, M. Bayati, T. Shanafelt, J. Profit · Applied Clinical Informatics 16(4), 1031–1040
  5. Electronic health record use patterns among well-being survey responders and non-responders: Longitudinal observational study D. Tawfik, T. Shanafelt, M. Bayati, J. Profit · JMIR Medical Informatics 13, e64722
  6. Large language models for preventing medication direction errors in online pharmacies C. Pais, J. Liu, R. Voigt, V. Gupta, E. Wade, M. Bayati · Nature Medicine 30, 1574–1582
    A fine-tuned language model that catches erroneous medication directions before they reach a patient, evaluated against pharmacist judgment.
    INFORMS Health Applications Pierskalla runner-up paper award, 2024Story: An AI copilot for prescriptions
  7. Predicting primary care physician burnout from electronic health record use measures D. Tawfik, M. Bayati, J. Liu, L. Nguyen, A. Sinha, T. Kannampallil, T. Shanafelt, J. Profit · Mayo Clinic Proceedings 99(9), 1411–1421
  8. Clinically Consistent Prostate Cancer Outcome Prediction Models with Machine Learning J. J. Vallon, N. Panjwani, X. Ling, S. Vij, E. Pollom, H. P. Bagshaw, S. Srinivas, J. Leppert, M. Bayati, M. K. Buyyounouski · American Society for Radiation Oncology (ASTRO) 2022
  9. Frustration with technology and its relation to emotional exhaustion among health care workers D. S. Tawfik, A. Sinha, M. Bayati, K. C. Adair, T. D. Shanafelt, J. B. Sexton, J. Profit · Journal of Medical Internet Research 23(7)
    Story
  10. PatientFlowNet: A Deep Learning Approach to Patient Flow Prediction in Emergency Departments A. R. Sharafat, M. Bayati · IEEE Access 9
  11. Online Decision-Making with High-Dimensional Covariates H. Bastani, M. Bayati · Operations Research 68(1)
    Personalized decisions when the number of candidate covariates far exceeds the number of observations.
    INFORMS Health Applications Pierskalla best paper award, 2016Nicholson, MSOM, and IBM Service Science best student paper, 2016
  12. Evidence of Upcoding in Pay-for-Performance Programs H. Bastani, J. Goh, M. Bayati · Management Science 65(3), 1042–1060
    INFORMS Health Applications Society best student paper award, 2015StoryStory
  13. Low-Acuity Patients Delay High-Acuity Patients in the Emergency Department D. Luo, M. Bayati, E. Plambeck, M. Aratow · Preprint
  14. Data Uncertainty in Markov Chains: Application to Cost-Effectiveness Analyses of Medical Innovations J. Goh, M. Bayati, S. Zenios, S. Singh, D. Moore · Operations Research 66(3)
    INFORMS Health Applications Pierskalla best paper award, 2014
  15. Statistical analysis of a low-cost method for multiple disease prediction M. Bayati, S. Bhaskar, A. Montanari · Statistical Methods in Medical Research 27(8)
  16. Analysis of Medicare Pay-for-Performance Contracts H. Bastani, M. Bayati, M. Braverman, R. Gummadi, R. Johari · EC Workshop on Mechanism Design for Social Good 2017
  17. Accurate Emergency Department Wait Time Prediction E. Ang, S. Kwasnick, M. Bayati, E. Plambeck, M. Aratow · Manufacturing & Service Operations Management 18(1), 141–156
    A quantile-regression wait time predictor that beat the rolling-average method hospitals were using, then went into production.
    Online supplementStoryDeployed at San Mateo Medical Center
  18. A Low-Cost Method for Multiple Disease Prediction M. Bayati, S. Bhaskar, A. Montanari · American Medical Informatics Association (AMIA) 2015
  19. Active Postmarketing Drug Surveillance for Multiple Adverse Events J. Goh, M. V. Bjarnadottir, M. Bayati, S. Zenios · Operations Research 63(6), 1528–1546
  20. Data-driven decisions for reducing readmissions for heart failure: general methodology and case study M. Bayati, M. Braverman, M. Gillam, K. Mack, G. Ruiz, M. Smith, E. Horvitz · PLOS ONE 9(10)
    CodeStoryHBRDeployed in several hospitals
  21. Identifying patients at high risk for readmission following treatment for acute myocardial infarction: a data-centric approach N. C. Baker, M. Bayati, R. Torguson, K. Mack, H. Rappaport, E. Horvitz, R. Waksman · American Heart Association Scientific Sessions 2014
  22. Health care policy development and execution M. Bayati, M. Braverman, M. Gillam, E. Horvitz · US 2012/0004925 A1, filed 2010

Message passing, random graphs, and high-dimensional statistics

The methodological foundation underneath much of the rest. Rigorous analysis of belief propagation and approximate message passing, exact risk characterizations for the LASSO, universality results for phase transitions in high-dimensional geometry, and algorithms for generating and aligning large random graphs.

  1. Higher-Order Causal Message Passing for Experimentation with Complex Interference M. Bayati, Y. Luo, W. Overman, S. Shirani, R. Xiong · NeurIPS 2024
  2. Causal Message Passing for Experiments with Unknown and General Network Interference S. Shirani, M. Bayati · Proceedings of the National Academy of Sciences 121(40)
    An estimator for total treatment effects when the interference network is unknown, built by tracking population-level outcome dynamics.
    Honorable mention, 2024 Nicholson Student Paper CompetitionFinalist, 2024 MSOM Student Paper CompetitionCodeOral presentation, CODE 2024
  3. On Low-rank Trace Regression under General Sampling Distribution N. Hamidi, M. Bayati · Journal of Machine Learning Research 23(321)
    Code
  4. Recommendation on a Budget: Column Space Recovery from Partially Observed Entries with Random or Active Sampling C. Kim, M. Bayati · AISTATS 2020
  5. Scalable Approximations for Generalized Linear Problems M. A. Erdogdu, M. Bayati, L. H. Dicker · Journal of Machine Learning Research 20:1–45
    Short version, NeurIPS 2016
  6. Generating Random Networks Without Short Cycles M. Bayati, A. Montanari, A. Saberi · Operations Research 66(5)
  7. Scaled Least Squares Estimator for GLMs in Large-Scale Problems M. A. Erdogdu, M. Bayati, L. H. Dicker · NeurIPS 2016
  8. Bargaining dynamics in exchange networks M. Bayati, C. Borgs, J. Chayes, Y. Kanoria, A. Montanari · Journal of Economic Theory 156, 417–454
  9. Universality in polytope phase transitions and message passing algorithms M. Bayati, M. Lelarge, A. Montanari · Annals of Applied Probability 25(2), 753–822
    INFORMS Applied Probability Society best paper award, 2015StorySettled a conjecture of Donoho and Tanner
  10. Combinatorial approach to the interpolation method and scaling limits in sparse random graphs M. Bayati, D. Gamarnik, P. Tetali · Annals of Probability 41(6), 4080–4115
    Settled an open problem of Aldous
  11. Estimating LASSO risk and noise level M. Bayati, M. A. Erdogdu, A. Montanari · NeurIPS 2013
  12. Message-passing algorithms for sparse network alignment M. Bayati, D. F. Gleich, A. Saberi, Y. Wang · ACM Transactions on Knowledge Discovery from Data 7(1)
    Code and data
  13. The LASSO risk for Gaussian matrices M. Bayati, A. Montanari · IEEE Transactions on Information Theory 58(4)
    An exact asymptotic characterization of LASSO estimation error, proved with approximate message passing.
  14. Belief propagation for weighted b-matchings on arbitrary graphs and its relation to linear programs with integer solutions M. Bayati, C. Borgs, J. Chayes, R. Zecchina · SIAM Journal on Discrete Mathematics 25, 989–1011
  15. Fast convergence of natural bargaining dynamics in exchange networks Y. Kanoria, M. Bayati, C. Borgs, J. Chayes, A. Montanari · ACM-SIAM Symposium on Discrete Algorithms (SODA) 2011
  16. The dynamics of message passing on dense graphs, with applications to compressed sensing M. Bayati, A. Montanari · IEEE Transactions on Information Theory 57(2)
    The state evolution proof that put approximate message passing on rigorous footing.
  17. A sequential algorithm for generating random graphs M. Bayati, J. H. Kim, A. Saberi · Algorithmica 58(4)
  18. Predicting web advertisement click success by using head-to-head ratings M. Bayati, M. Braverman, S. Kale, Y. Makarychev · US 2010/0198685 A1
  19. The LASSO risk: asymptotic results and real world examples M. Bayati, J. Bento, A. Montanari · NeurIPS 2010
  20. A rigorous analysis of the cavity equations for the minimum spanning tree M. Bayati, A. Braunstein, R. Zecchina · Journal of Mathematical Physics 49, 125206
  21. Algorithms for large, sparse network alignment problems M. Bayati, M. Gerritsen, D. F. Gleich, A. Saberi, Y. Wang · IEEE International Conference on Data Mining (ICDM) 2009
  22. Generating random graphs with large girth M. Bayati, A. Montanari, A. Saberi · ACM-SIAM Symposium on Discrete Algorithms (SODA) 2009
    Implementation details in ITW 2009
  23. Generating random Tanner-graphs with large girth M. Bayati, R. Keshavan, A. Montanari, S. Oh, A. Saberi · IEEE Information Theory Workshop (ITW) 2009
  24. Network analysis with Steiner trees M. Bayati, C. Borgs, A. Braunstein, J. Chayes, R. Zecchina · US 2009/0222782 A1, granted 2011
  25. Max-product for maximum weight matching: convergence, correctness, and LP duality M. Bayati, D. Shah, M. Sharma · IEEE Transactions on Information Theory 54(3)
  26. On the exactness of the cavity method for weighted b-matchings on arbitrary graphs and its relation to linear programs M. Bayati, C. Borgs, J. Chayes, R. Zecchina · Journal of Statistical Mechanics, L06001
  27. Statistical mechanics of Steiner trees M. Bayati, C. Borgs, A. Braunstein, J. Chayes, A. Ramezanpour, R. Zecchina · Physical Review Letters 101, 037208
  28. Iterative scheduling algorithms M. Bayati, B. Prabhakar, D. Shah, M. Sharma · IEEE INFOCOM 2007
  29. Simple deterministic approximation algorithms for counting matchings M. Bayati, D. Gamarnik, D. Katz, C. Nair, P. Tetali · ACM Symposium on Theory of Computing (STOC) 2007
  30. A rigorous proof of the cavity method for counting matchings M. Bayati, C. Nair · Allerton Conference on Communication, Control and Computing 2006
  31. A simpler max-product maximum weight matching algorithm and the auction algorithm M. Bayati, D. Shah, M. Sharma · IEEE International Symposium on Information Theory (ISIT) 2006
  32. Optimal scheduling in multi-server queueing networks M. Bayati, M. Squillante, M. Sharma · ACM SIGMETRICS/Performance 2006
  33. Achieving stability in networks of input-queued switches using a local online scheduling policy N. Kumar, S. Nabar, M. Bayati, A. Keshavarzian · IEEE GLOBECOM 2005
  34. Maximum weight matching via max-product belief propagation M. Bayati, D. Shah, M. Sharma · IEEE International Symposium on Information Theory (ISIT) 2005
  35. Stability of the maximum size matching in input-queued switches M. Bayati, N. Beheshti · Allerton Conference on Communication, Control and Computing 2004