Program

Conference Program

12:30pm-1:00pm Registration

Come pick up your name tag before the conference begins!

1:00pm-1:20pm Room-TEP3801	Opening Remarks
	Laurence Ales, Professor of Economics at Tepper School of Business, Carnegie Mellon University

1:20pm-2:00pm Room-TEP3801	Topology-Based Resilient Logistics Network Design
	Mathieu Dahan, Assistant Professor in the H. Milton Stewart School of Industrial and Systems Engineering, Georgia Tech
	We consider the problem of designing large-scale resilient relay logistics networks by locating logistics hubs to optimize resilience metrics based on demand-weighted k shortest (edge-disjoint) path lengths between each origin-destination pair. We derive mixed-integer programming formulations using path and edge decisions, and leverage their structures to devise two tailored scalable solution methodologies based on Benders decomposition and branch-and-price. We showcase the applicability of our methodology in designing large-scale resilient relay networks for a China-based parcel delivery partner. Our results demonstrate that the designed networks significantly outperform traditional lean network designs when confronted with stochastic hub disruptions, with only marginal impact on operating costs in nominal situations. This work is joint with Onkar Kulkarni and Benoit Montreuil.

2:00pm-2:40pm Room-TEP3801	Balancing Optimality and Diversity: Enhancing Human Decision-Making through Generative Curation
	Woody Zhu, Assistant Professor in Heinz College of Information Systems and Public Policy, Carnegie Mellon University
	The rapid increase in data availability has overwhelmed decision-makers with an abundance of choices and information. In response, there has been considerable work in creating optimal decision rules for a quantifiable objective. However, in many practical settings, human decision-makers must consider both explicit quantitative and implicit qualitative factors to make the final call. We introduce a general framework, termed generative curation, which generates optimal recommendations that account for both quantitative and qualitative objectives. We show that a consideration of implicit qualitative factors naturally leads to a metric that measures the diversity of the generated solutions, transforming the problem into balancing between quantitative optimality and qualitative diversity. Our proposed algorithm efficiently solves this optimization problem by generating a finite number of diverse and near-optimal solutions. We validate our approach with real-world datasets, showcasing its potential to enhance decision-making processes in complex decision-making settings.

2:40pm-3:05pm

Break

3:05pm-3:45pm Room-TEP3801	Strongly Tail-Optimal Scheduling in the Light-Tailed M/G/1
	Ziv Scully, Assistant Professor at Cornell ORIE (Operations Research and Information Engineering)
	We study the problem of scheduling jobs in a queueing system, specifically an M/G/1 with light-tailed job sizes, to asymptotically optimize the response time tail. For some time, the best known policy was First-Come First-Served (FCFS), which has an asymptotically exponential tail. FCFS achieves the optimal exponential decay rate, but its leading constant is suboptimal. Designing a policy that minimizes this leading constant is a long-standing open problem. We solve this open problem with a new scheduling policy called 𝛾-Boost. Roughly speaking, 𝛾-Boost operates similarly to FCFS, but it pretends that small jobs arrive earlier than their true arrival times. This reduces the response time of small jobs without unduly delaying large jobs. We prove 𝛾-Boost's asymptotic tail optimality, and we show via simulation that 𝛾-Boost has excellent practical performance. The 𝛾-Boost policy as described above requires knowledge of job sizes. In preliminary work, we generalize 𝛾-Boost to work with unknown job sizes, proving an analogous asymptotic optimality result in the unknown-size setting. Our generalization reveals that 𝛾-Boost is a type of Gittins index policy, but with an unusual feature: it uses a negative discount rate. Joint work with George Yu (Cornell) and Amit Harlev (Cornell). Paper links: [1] [2].

3:45pm-4:25pm Room-TEP3801	Online matching with graph neural networks
	Ellen Vitercik, Assistant Professor at Management Science and Engineering and Computer Science, Stanford University
	Online Bayesian bipartite matching is a central problem in digital marketplaces and exchanges, including advertising, crowdsourcing, ridesharing, and kidney exchange. We introduce a graph neural network (GNN) approach that emulates the problem's combinatorially complex optimal online algorithm, which selects actions (e.g., which nodes to match) by computing each action's value-to-go (VTG)—the expected weight of the final matching if the algorithm takes that action, then acts optimally in the future. We train a GNN to estimate VTG and show empirically that this GNN returns high-weight matchings across a variety of tasks. Moreover, we identify a common family of graph distributions in spatial crowdsourcing applications, such as rideshare, under which VTG can be efficiently approximated by aggregating information within local neighborhoods in the graphs. This structure matches the local behavior of GNNs, providing theoretical justification for our approach. This is joint work with Alexandre Hayderi, Amin Saberi, and Anders Wikum.

4:25pm-4:50pm

Break

4:50pm-6:30pm Room-TEP3801	10-Minute Flash Talk Competition

	Selected students will present 10-Minute Flash Talks in a competition judged by the attendees with cash prizes!

6:30pm-9:00pm Room-TEP3808	Dinner

	Everyone is welcome to join us for a free dinner catered by a local Pittsburgh restaurant. We might go out for drinks afterward as well.

Saturday, August 24th

9:00am-10:00am Breakfast
Room-TEP3801

All are welcome to a free breakfast and coffee to start the day

10:00am-10:40am Room-TEP3801	Strategically-Robust Bidding Algorithms for First-Price Auctions
	Rachitesh Kumar, Incoming Assistant Professor at Tepper School of Business, Carnegie Mellon University
	Learning to bid in repeated first-price auctions is a fundamental problem at the interface of game theory and machine learning, which has seen a recent surge in interest due to the transition of display advertising to first-price auctions. Past work has predominantly treated it as an online learning problem and focused on developing regret-minimizing algorithms. However, such an approach ignores the strategic nature of the market in which such algorithms are deployed---sellers and buyers can manipulate these algorithms for their own benefit. In this work, we go beyond regret minimization and develop bidding algorithms that are robust to manipulations by strategic market participants. We propose a novel concave formulation for pure-strategy bidding in first-price auctions and use it to analyze Gradient Ascent. Unlike previously-proposed algorithms, which attain optimal regret but are manipulable, we show that Gradient Ascent attains optimal regret while being robust to strategic manipulations by the buyer and the seller. Based on https://arxiv.org/abs/2402.07363.

10:40am-11:20am Room-TEP3801	When A/B Testing Platforms Meet Reinforcement Learning
	Tianyi Peng, Incoming Assistant Professor at Columbia University
	In the internet era, A/B testing has been regarded as the gold standard for evaluating algorithm effectiveness. However, the problem of interference—where different experimental units affect each other—remains a fundamental challenge in A/B testing. To overcome this issue, we propose an innovative solution based on the reinforcement learning framework to reevaluate A/B testing. This method estimates the treatment effect by solving the difference in Q-values in reinforcement learning, hence named the "Difference-in-Q" (DQ) estimator. Theoretically, we find that DQ performs excellently in the bias-variance trade-off: on the one hand, DQ’s bias is second-order compared to traditional estimators; on the other hand, DQ's variance can be exponentially reduced compared to any unbiased estimator. We collaborated with Douyin to apply DQ in a large-scale commercial setting, where it reduced the mean squared error by over 99% in our tests. Additionally, DQ also demonstrated outstanding performance in a commercial-grade shared vehicle simulator.

11:20am-12:00pm Room-TEP3801	Generative Social Choice
	Paul Gölz, Incoming Assistant Professor at Cornell ORIE (Operations Research and Information Engineering)
	Traditionally, social choice theory has only been applicable to choices among a few predetermined alternatives but not to more complex decisions such as collectively selecting a textual statement. We introduce generative social choice, a framework that combines the mathematical rigor of social choice theory with the capability of large language models to generate text and extrapolate preferences. This framework divides the design of AI-augmented democratic processes into two components: first, proving that the process satisfies rigorous representation guarantees when given access to oracle queries; second, empirically validating that these queries can be approximately implemented using a large language model. We apply this framework to the problem of generating a slate of statements that is representative of opinions expressed as free-form text; specifically, we develop a democratic process with representation guarantees and use this process to represent the opinions of participants in a survey about chatbot personalization. We find that 93 out of 100 participants feel ‘‘mostly’’ or ‘‘perfectly’’ represented by the slate of five statements we extracted.

12:00pm-1:30pm	Lunch

	Lunch on your own! We welcome you to explore restaurants around the area. Please ask us if you would like any recommendations!

1:30pm-2:10pm Room-TEP3801	Hexaly, a new kind of global optimization solver
	Fred Gardi, Founder & CEO at Hexaly
	Hexaly Optimizer is a new kind of global optimization solver. Its modeling interface is nonlinear and set-oriented. It also supports user-coded functions, thus enabling black-box optimization and, more particularly, simulation optimization. In a sense, Hexaly APIs unify modeling concepts from mixed-linear programming, nonlinear programming, and constraint programming. Under the hood, Hexaly combines various exact and heuristic optimization methods: spatial branch-and-bound, simplex methods, interior-point methods, augmented Lagrangian methods, automatic Dantzig-Wolfe reformulation, column and row generation, propagation methods, local search, direct search, population-based methods, and surrogate modeling techniques for black-box optimization. Regarding performance benchmarks, Hexaly distinguishes itself against the leading solvers in the market, like Gurobi, IBM Cplex, and Google OR Tools, by delivering fast and scalable solutions to problems in the spaces of Supply Chain and Workforce Management like Routing, Scheduling, Packing, Clustering, and Location. For example, on notoriously hard problems like the Pickup and Delivery Problem with Time Windows or Flexible Job Shop Scheduling with Setup Times, Hexaly delivers solutions with a gap to the best solutions known in the literature smaller than 1% in a few minutes of running times on a basic computer. In addition to the Optimizer, we provide an innovative development platform called Hexaly Studio to model and solve rich Vehicle Routing and Job Shop Scheduling problems in a no-code fashion. The user can define its problem and data, run the Optimizer, visualize the solutions and key metrics through dashboards, and deploy the resulting app in the cloud – without coding. This web-based platform is particularly interesting for educational purposes; it is free for faculty and students.

2:10pm-3:40pm Room-TEP2001, TEP2002, TEP2003	Poster Session

	Students will setup posters in an open space. Winners will be determined by both formal judges and attendees popular votes!

3:40pm-4:20pm Room-TEP3801	Layout Design for Large-Scale Multi-Robot Coordination
	Jiaoyang Li, Assistant Professor at Robotics Institute Carnegie Mellon University
	Today, thousands of robots are navigating autonomously in warehouses, transporting goods from one location to another. While numerous planning algorithms are developed to coordinate robots more efficiently and robustly, warehouse layouts remain largely unchanged – they still adhere to the traditional pattern designed for human workers rather than robots. In this talk, I will share our recent progress in exploring layout design and optimization to enhance large-scale multi-robot coordination. First, I will introduce a direct layout optimization method based on stochastic optimization. Then, I will discuss a method to optimize layout generators instead of layouts. Last, I will extend these ideas to virtual layout design, which does not require changes to the physical world that robots navigate and thus has the potential for applications beyond automated warehouses.

4:20pm-5:00pm Room-TEP3801	Machine learning for public health decision making
	Bryan Wilder, Assistant Professor in the Machine Learning Department at Carnegie Mellon University
	Many policy settings, such as public health, require allocating limited resources under uncertainty about which individuals need those resources the most. Machine learning plays an increasing role in such allocations, most commonly to provide predictions of individual-level need. In this talk, I will discuss two recent works that take a more expansive view of the role of ML in resource allocation. First, I will discuss methods that allow a decision maker to optimally balance the immediate goal of targeting high-need individuals and the long-term goal of estimating treatment effects (i.e., does the program help people?). Our proposed method provides a policy for resource allocation which minimizes the variance with which the average treatment effect is estimated, subject to a constraint that at least some fraction of in-need individuals receive the intervention. Second, I will introduce ML as a lens to study the fairness of discretionary allocations by human decision makers, such as a clinician's decision to prescribe a resource-limited drug. Here the challenge is that human decisions are likely to be confounded (based on unobserved variables); we provide nontrivial bounds on allocative equity even in this challenging setting. In both works, we give both strong guarantees for the statistical and computational properties of our proposed methods as well as applications to real-world data from health and social service settings.

5:00pm-5:30pm Room-TEP3801	Closing Remarks
	Fatma Kilinc-Karzan, Associate Professor of Operations Research and CMU INFORMS Faculty Advisor

5:30pm-7:30pm Room-TEP2001, TEP2002, TEP2003	Happy Hour
	Free refreshments to celebrate a wonderful conference. The party may continue past 7:30pm outside of the building.