About

My research focuses on the intersection of data driven robust optimization and mechanism design for systems of interacting, self-interested, strategic agents. I am motivated by practical problems at the intersection of learning, optimization, computation, and economics, specifically the design of markets (or mechanisms). As the world moves towards increasing automation, there are, more and more, opportunities to combine the wealth of data with a principled, provably optimal approach to mechanism design in order to make intractable problems tractable and impossible problems possible.

Research topics

• Computer Science
• Computer Security
• Advertising
• Psychology
• Mathematical optimization
• Operations research
• Finance
• Business
• Engineering
• Transport engineering
• Microeconomics
• Economics
• Mathematics

Selected publications

• Is Learning Effective in Dynamic Strategic Interactions? Evidence from Stackelberg Games

  ArXiv.org · 2025-04-22

  preprintOpen access1st authorCorresponding

  In many settings of interest, a policy is set by one party, the leader, in order to influence the action of another party, the follower, where the follower's response is determined by some private information. A natural question to ask is, can the leader improve their strategy by learning about the unknown follower through repeated interactions? A well known folk theorem from dynamic pricing, a special case of this leader-follower setting, would suggest that the leader cannot learn effectively from the follower when the follower is fully strategic, leading to a large literature on learning in strategic settings that relies on limiting the strategic space of the follower in order to provide positive results. In this paper, we study dynamic Bayesian Stackelberg games, where a leader and a \emph{fully strategic} follower interact repeatedly, with the follower's type unknown. Contrary to existing results, we show that the leader can improve their utility through learning in repeated play. Using a novel average-case analysis, we demonstrate that learning is effective in these settings, without needing to weaken the follower's strategic space. Importantly, this improvement is not solely due to the leader's ability to commit, nor does learning simply substitute for communication between the parties. We provide an algorithm, based on a mixed-integer linear program, to compute the optimal leader policy in these games and develop heuristic algorithms to approximate the optimal dynamic policy more efficiently. Through simulations, we compare the efficiency and runtime of these algorithms against static policies.

  PublisherOA PDFDOI

• Learning in Online Principal-Agent Interactions: The Power of Menus

  arXiv (Cornell University) · 2023-12-15 · 1 citations

  preprintOpen access

  We study a ubiquitous learning challenge in online principal-agent problems during which the principal learns the agent's private information from the agent's revealed preferences in historical interactions. This paradigm includes important special cases such as pricing and contract design, which have been widely studied in recent literature. However, existing work considers the case where the principal can only choose a single strategy at every round to interact with the agent and then observe the agent's revealed preference through their actions. In this paper, we extend this line of study to allow the principal to offer a menu of strategies to the agent and learn additionally from observing the agent's selection from the menu. We provide a thorough investigation of several online principal-agent problem settings and characterize their sample complexities, accompanied by the corresponding algorithms we have developed. We instantiate this paradigm to several important design problems $-$ including Stackelberg (security) games, contract design, and information design. Finally, we also explore the connection between our findings and existing results about online learning in Stackelberg games, and we offer a solution that can overcome a key hard instance of Peng et al. (2019).

  PublisherOA PDFDOI

• Operationalizing Machine Learning Models for Strategic Planning

  2023-06-08 · 1 citations

  article

  View Video Presentation: https://doi.org/10.2514/6.2023-4213.vid The Air Traffic Organization within the Federal Aviation Administration is focused on improving the pre-tactical planning phase of air traffic flow management. The Advanced Planning team at the Air Traffic Control System Command Center is responsible for planning operations for the following day. The purpose of this research is to analyze the feasibility of operationalizing machine learning to identify potential demand-capacity imbalances for use by the Advanced Planning Team. Random Forest models are trained for each of the CONUS Core 30 airports to predict both runway configurations and rates. The accuracy of these models is assessed with the time-series walk-forward validation method. The preliminary results are encouraging. The speed at which these models can be retrained and their associated accuracies are sufficient for phase-based trial implementation at specific airports to determine operational feasibility. However, the efficacy of the machine learning models varies by airport and more research is needed to improve the low performing models.

  PublisherDOI

• Predicting Runway Configurations and Arrival and Departure Rates at Airports: Comparing the Accuracy of Multiple Machine Learning Models

  2015 IEEE/AIAA 34th Digital Avionics Systems Conference (DASC) · 2021 · 4 citations

  Senior authorCorresponding
  • Computer Science
  • Computer Science
  • Operations research

  In response to the needs of various stakeholders, the FAA has developed an automated tool that provides values for expected next day airport runway configurations and their respective arrival and departure rates. The arrival and departure rates at an airport define the capacity of an airport and as such are a critical piece in determining whether not a capacity/demand imbalance is expected. A forecast of these imbalances facilitates next-day planning and provides to stakeholders transparency regarding potential traffic constraints. The use case that motivated the work described in this paper is to develop various machine-learning (ML) models and compare their performance against the current automated tool, which finds operational periods with similar facility (airport) weather and uses the most frequently occurring runway configuration and rates as the capacity forecast for the next day. Extensive model development across five airports and four years of operational data and forecasted weather showed that the ML models consistently had higher accuracy than the automated tool. Among the ML models, superior options were found, but a single best model across all airports and predictions was not. In a follow-on effort, the authors hope to explore opportunities for working with the user community, incorporate their feedback, and determine if one or more ML models can be used in practice.

  PublisherDOI

• Mechanism Design for Correlated Valuations: Efficient Methods for Revenue Maximization

  Operations Research · 2021 · 7 citations

  1st authorCorresponding
  • Computer Science
  • Computer Science
  • Mathematical optimization

  Traditionally, much of the focus of the mechanism/auction design community has been on revenue optimal mechanisms for settings where bidders’ private valuations over outcomes can be reasonably thought of as independent of each other. This has been the case even though there is good reason to believe that valuations are often correlated and there are theoretical results suggesting that mechanisms designed with this correlation in mind can generate much higher revenue. In “Mechanism Design for Correlated Valuations: Efficient Methods for Revenue Maximization,” we look at the setting where there is correlation, but the exact distribution is unknown and must be estimated from samples. We show that in this setting, the previous extremely strong theoretical results around the usefulness of correlation are now very sensitive to the degree of correlation in the underlying distribution and the number of samples that the mechanism designer has access to. However, we also show that if correlation is sufficient, we can construct mechanisms, using a computationally efficient procedure, that significantly outperform traditional mechanism design paradigms.

  PublisherDOI

• Walker Advertising: Los Defensores and 1-800-THE-LAW2

  SSRN Electronic Journal · 2020

  1st authorCorresponding
  • Advertising
  • Psychology
  • Business
  PublisherDOI

• Walker Advertising: Los Defensores and 1-800-THE–LAW2

  SSRN Electronic Journal · 2020-01-01

  articleOpen access1st authorCorresponding
  PublisherDOI

• AAAI/ACM SIGAI job fair 2020

  AI Matters · 2020-05-01

  article1st authorCorresponding

  For the sixth year running, AAAI and ACM SIGAI jointly ran the popular AAAI/ACM SIGAI Job Fair. In lockstep with the growth of AAAI and the growth of the greater artificial intelligence and machine learning (AI/ML) community, our once-small job fair also grew. This year, thirty-eight companies and universities formally attended---typically with a booth, team of recruiters, swag, and other representatives---increasing from twenty-six companies during the job fair's previous run in 2019, and twenty-one companies in the year prior to that. Last year, we purchased a dedicated domain---https://aaaijobfair.com/---for the job fair. This year, we provided a link on that site through which job-seekers---students, postdocs, practitioners, and maybe even a few faculty---could upload their resumes or CVs. We then shared that data and contact information for slightly under four hundred job-seekers with participants on the other side: prospective employers.

  PublisherDOI

• Complexity of Scheduling Charging in the Smart Grid : Extended Abstract

  2018-01-01

  article

  The problem of optimally scheduling the charging demand of electric vehicles within the constraints of the electricity infrastructure is called the charge scheduling problem. The models of the charging speed, horizon, and charging demand determine the computational complexity of the charge scheduling problem. For about 20 variants the problem is either in P or weakly NP-hard and dynamic programs exist to compute optimal solutions. About 10 other variants of the problem are strongly NP-hard, presenting a potentially significant obstacle to their use in practical situations of scale.

  Publisher

• Traffic Optimization for a Mixture of Self-Interested and Compliant Agents

  Proceedings of the AAAI Conference on Artificial Intelligence · 2018-04-25 · 5 citations

  preprintOpen access

  This paper focuses on two commonly used path assignment policies for agents traversing a congested network: self-interested routing, and system-optimum routing. In the self-interested routing policy each agent selects a path that optimizes its own utility, while in the system-optimum routing, agents are assigned paths with the goal of maximizing system performance. This paper considers a scenario where a centralized network manager wishes to optimize utilities over all agents, i.e., implement a system-optimum routing policy. In many real-life scenarios, however, the system manager is unable to influence the route assignment of all agents due to limited influence on route choice decisions. Motivated by such scenarios, a computationally tractable method is presented that computes the minimal amount of agents that the system manager needs to influence (compliant agents) in order to achieve system optimal performance. Moreover, this methodology can also determine whether a given set of compliant agents is sufficient to achieve system optimum and compute the optimal route assignment for the compliant agents to do so. Experimental results are presented showing that in several large-scale, realistic traffic networks optimal flow can be achieved with as low as 13% of the agent being compliant and up to 54%.

  PublisherOA PDFDOI

Frequent coauthors

• Peter Stone

  11 shared

• Vincent Conitzer

  Carnegie Mellon University

  8 shared

• Stephen D. Boyles

  7 shared

• Guni Sharon

  Texas A&M University

  7 shared

• Josiah P. Hanna

  5 shared

• Tarun Rambha

  Indian Institute of Science Bangalore

  5 shared

• Michael W. Levin

  3 shared

• Giuseppe Lopomo

  Duke University

  3 shared