IEOR-DRO Seminar

Title: Optimal Match Recommendations in Two-sided Marketplaces with Endogenous Prices

Abstract: Many two-sided marketplaces rely on match recommendations to help customers find suitable service providers at suitable prices. This paper develops a tractable methodology that a platform can use to optimize its match recommendation policy so as to maximize the total value generated by the platform while accounting for the endogeneity of transaction prices, which are set by the providers based on supply and demand and can depend on the platform's match recommendation policy. Despite the complications of price endogeneity, an optimal match recommendation policy has a simple structure and can be computed efficiently. In particular, an optimal policy always recommends the providers who deliver the highest conversion rates. Moreover, an optimal policy can be encoded simply in terms of the frequency of recommending each provider to each customer segment, without the need to encode which subsets of providers are to be recommended together. On the other hand, if the platform were to optimize its match recommendations without accounting for price endogeneity, then the resultant policy would be more complex, and the market is likely to get stuck at a strictly suboptimal outcome, even if the platform were to continually re-optimize its match recommendations after prices re-equilibrate.

Bio: Peng Shi is an Assistant Professor of Data Sciences and Operations in the USC Marshall School of Business. His current research focuses on optimization in matching markets, with applications in school choice, public housing, organ allocation, and online marketplaces. His research has won multiple awards, including the MSOM Responsible Research in OM Award, the MSOM Service Management SIG Best Paper Award, the ACM SIGecom Doctoral Dissertation Award, the INFORMS Public Sector Operations Best Paper Competition, and the INFORMS Doing Good with Good OR Student Paper Competition. Prior to joining USC, he completed a PhD in operations research at MIT, and was a post-doctoral researcher at Microsoft Research.

Title: Completing large low rank matrices with only few observed entries: A one-line algorithm with provable guarantees.

Abstract: Suppose you observe very few entries from a large matrix. Can we predict the missing entries, say assuming the matrix is (approximately) low rank ? We describe a very simple method to solve this matrix completion problem. We show our method is able to recover matrices from very few entries and/or with ill conditioned matrices, where many other popular methods fail. Furthermore, due to its simplicity, it is easy to extend our method to incorporate additional knowledge on the underlying matrix, for example to solve the inductive matrix completion problem. On the theoretical front, we prove that our method enjoys some of the strongest available theoretical recovery guarantees. Finally, for inductive matrix completion, we prove that under suitable conditions the problem has a benign optimization landscape with no bad local minima. Joint work with Pini Zilber.

Bio: Professor Nadler received his Ph.D. at Tel Aviv University. After 3 years as a Gibbs instructor/assistant professor at Yale University, he joined the faculty at the department of computer science and applied mathematics at the Weizmann Institute of Science. His research interests span mathematical statistics, machine learning and various applications, including optics and signal processing.

Zoom Recording

Title: The Physics (and Ethics) of Sequential Fair Allocation

Abstract: In many settings, resources are allocated among people over time, without using monetary transfers: cloud resources among employees, food among food-banks, medical supplies between hospitals, funding between non-profit projects, etc. The underlying aim is to try and be ‘fair’ in these allocations…but what exactly do we mean?

Understanding fairness in decision-making is one of the most socially urgent (and also intellectually beautiful) topics today, with deep connections to economics, optimization, and normative philosophy. I will describe a foundational result of Varian’s that serves as my lodestar, and build on it to give my opinions on what any theory of fairness in sequential decision-making should and should not do. My core thesis will be that, as academic researchers, our main (and only!) task should be to characterize Pareto trade-offs between fairness and efficiency in different problem settings, as a guide to policy-makers to help choose which tradeoffs is socially relevant. I will illustrate this approach with a series of settings (inspired by work we have been doing with our local food-bank), where we have been able to characterize information-theoretic "uncertainty principles" between efficiency and fairness loss. Time permitting, I will outline questions relating fairness, efficiency and information trade-offs in more complex settings.

Bio: Sid Banerjee is an associate professor in the School of Operations Research at Cornell, working on topics at the intersection of data-driven decision-making, network algorithms and market design. His research is supported by grants from the NSF (including an NSF CAREER award), the ARL Network Sciences division, and Engaged Cornell. He received his PhD from the ECE Department at UT Austin, and was a postdoctoral researcher in the Social Algorithms Lab at Stanford. He also served as a technical consultant with the research science group at Lyft from 2014-18. 

Zoom Recording

Title: Learning the Scheduling Policy in Time-Varying Multiclass Many Server Queues with Abandonment

Abstract: We consider a learning variant of a canonical scheduling problem in a multiclass many server queue with abandonment (specifically, the M_t/M/N+M and the GI/GI/N+GI queues). The objective is to minimize the long-run average class-dependent expected linear holding and abandonment costs when the class-dependent model parameters (arrival rates, service rates and abandonment rates) are a priori unknown. The difficulty is that even when parameters are known, characterizing an optimal scheduling policy appears intractable. Fortunately, the simple cμ/θ rule, that prioritizes classes in accordance with a static ranking that depends on the costs, the service rates, and the abandonment rates, is asymptotically optimal as the arrival rates and number of servers become large, under certain conditions. Then, our task is to learn the service and abandonment rates well enough to determine an optimal static priority ranking for the classes, and we can benchmark our performance by defining the regret relative to the cμ/θ rule.

We propose a Learn-Then-Schedule algorithm, which is composed of a learning phase during which point estimates of the mean service and patience times are formed, and an exploitation phase during which the cμ/θ rule with empirical mean estimates as a surrogate for actual parameters is followed. It is shown that the smallest achievable regret for static priority scheduling policies in T periods is Ω(logT), and we prove that our proposed algorithm achieves a regret upper bound of O(logT), which matches the lower bound. 

Bio: Amy is the Rothman Family Professor of Operations Management at the University of Chicago Booth School of Business.  She received her Ph.D. from Stanford University in 2001.  She is the Editor-in-Chief of Operations Research Letters (since April 2021), and was the Area Editor for the Stochastic Models Area of Operations Research (2018-2021).  She was the chair of the Applied Probability Society (APS) from 2016-2018, during which she started the APS best student paper competition.  Her main interest is in developing queueing theory methodology to support efficient service operations.

Title: Equilibrium Complexity and Deep Learning

Abstract: Deep Learning has recently made significant progress in learning challenges such as speech and image recognition, automatic translation, and text generation, much of that progress being fueled by the success of gradient descent-based optimization methods in computing local optima of non-convex objectives. From robustifying machine learning models against adversarial attacks to causal inference, training generative models, multi-robot interactions, and learning in strategic environments, many outstanding challenges in Machine Learning lie at its interface with Game Theory. On this front, however, Deep Learning has been less successful. Here, the role of single-objective optimization is played by equilibrium computation, but gradient-descent based methods fail to find equilibria, and even computing local equilibria -- the analog of computing local optima in single-agent settings -- has remained elusive. 

We shed light on these challenges through a combination of learning-theoretic, complexity-theoretic, and game-theoretic techniques, presenting obstacles and opportunities for Machine Learning and Game Theory going forward, including recent progress on multi-agent reinforcement learning.

(I will present results from joint works with Noah Golowich, Stratis Skoulakis, Manolis Zampetakis, and Kaiqing Zhang.)

Bio: Constantinos "Costis" Daskalakis is the Avanessians Professor of Electrical Engineering and Computer Science at MIT. He holds a Diploma in Electrical and Computer Engineering from the National Technical University of Athens, and a PhD in Electrical Engineering and Computer Science from UC Berkeley. He works on Computation Theory and its interface with Game Theory, Economics, Probability Theory, Machine Learning and Statistics. He has resolved long-standing open problems about the computational complexity of Nash equilibrium, and the mathematical structure and computational complexity of multi-item auctions. His current work focuses on multi-agent learning, high-dimensional statistics, learning from biased and dependent data, causal inference and econometrics. He has been honored with the ACM Doctoral Dissertation Award, the Kalai Prize from the Game Theory Society, the Sloan Fellowship in Computer Science, the SIAM Outstanding Paper Prize, the Microsoft Research Faculty Fellowship, the Simons Investigator Award, the Rolf Nevanlinna Prize from the International Mathematical Union, the ACM Grace Murray Hopper Award, the Bodossaki Foundation Distinguished Young Scientists Award, and the ACM SIGECOM Test of Time Award.

Zoom Recording

Title: Continuity of Care Increases Physician Productivity in Primary Care  

Abstract: Continuity of care, defined as an ongoing therapeutic relationship between a patient and a physician, is a defining characteristic of primary care. However, arranging a consultation with one’s regular doctor is increasingly difficult as practices face physician shortages. We study the effect of declining care continuity on the productivity of physicians by analyzing data of over 10 million consultations in 381 English primary care practices over a period of 11 years. Specifically, we examine whether a consultation with the patient’s regular doctor is more productive than with another doctor in the practice. Using statistical models that account for confounding and selection bias and restricting the sample to consultations with patients who had at least three consultations over the past two years, we find that the time to a patient’s next visit is on average 18.1% (95% CI: [16.9%, 19.2%]) longer when the patient sees the doctor they have seen most frequently over the past two years, while there is no operationally meaningful difference in consultation duration. The data shows that the productivity benefit of care continuity is larger for older patients, patients with multiple chronic conditions, and patients with mental health conditions. We estimate that the total consultation demand in our sample could have fallen by up to 5.2% had all practices offered continuity of care at the level of the top decile of practices while prioritizing patients expected to yield the largest productivity benefits. We discuss operational and strategic implications of these findings for primary care practices and for third-party payers.

 Bio: Michael Freeman is an Assistant Professor of Technology and Operations Management at INSEAD. Before joining INSEAD, he received his MPhil and Ph.D. in Management Science and Operations from the University of Cambridge. His research is primarily empirical in approach and focuses on topics in healthcare management and empirical operations, applying advanced empirical methods to large multi-provider data sets to study the organizational determinants of decision quality in multi-stage flow systems. Working closely with executives and practitioners in hospitals and primary care practices, his research provides insights into how health providers can evolve their business models and adopt new technologies to meet the dual challenge of improving clinical outcomes while controlling costs. He has several papers published in the premier operations management journals Management Science and M&SOM, and his research has also been recognized with various awards, including winning the 2016 MSOM Student Paper Competition. Michael is also an award-winning teacher at INSEAD. He teaches in INSEAD’s executive education programs, the EMBA program, the MBA program, and the PhD program. In the classroom, his expertise lies in helping firms to realize new growth opportunities by unlocking existing capabilities, implementing cutting-edge technologies, and identifying untapped business opportunities.

Speaker: Renato Paes Leme, Google Research 11/1

Title: Interactive Communication in Bilateral Trade

Abstract: The common wisdom in bilateral trade is that it is impossible to achieve efficiency while preserving incentives. Here we will allow the agents to have (possibly long) conversation before they trade and will show that communication leads to efficiency. 

(Joint work with Jieming Mao and Kangning Wang) 

Arxiv link: https://arxiv.org/abs/2106.02150

Bio: Renato Paes Leme is a Research Scientist at Google NYC. He manages a team whose goal is to apply ideas from mechanism design to different Google products. His current research interests are auction design, machine learning in economic environments (e.g. learning to price) and topics in applied probability. Before Google, he was a postdoc at Microsoft Research and obtained his PhD from Cornell University.

Zoom Recording
 

More details to follow. 

More details to follow. 

More details to follow. 

More details to follow. 

More details to follow.