# Plenaries & Tutorials

Plenary Speakers:

Asu Ozdaglar (MIT)

Title:Learning from Reviews

Abstract:

Many online platforms present summaries of reviews by previous users. Even though such reviews could be useful, previous users leaving reviews are typically a selected sample of those who have purchased the good in question, and may consequently have a biased assessment. In this paper, we construct a simple model of dynamic Bayesian learning and profit-maximizing behavior of online platforms to investigate whether such review systems can successfully aggregate past information and the incentives of the online platform to choose the relevant features of the review system.

On the consumer side, we assume that each individual cares about the underlying quality of the good in question, but in addition has heterogeneous ex ante and ex post preferences (meaning that she has a different strength of preference for the good in question than other users, and her enjoyment conditional on purchase is also a random variable). After purchasing a good, depending on how much they have enjoyed it, users can decide to leave a positive or a negative review (or leave no review if they do not have strong preferences). New users observe a summary statistic of past reviews (such as fraction of all reviews that are positive or fraction of all users that have left positive review etc.). Our first major result shows that, even though reviews come from a selected sample of users, Bayesian learning ensures that as the number of potential users grows, the assessment of the underlying state converges almost surely to the true quality of the good. More importantly, we provide a tight characterization of the speed of learning (which is a contribution relative to most of the works in this area that focus on whether there is learning or not). Under the assumption that the online platform receives a constant revenue from every user that purchases (because of commissions from sellers or from advertising revenues), we then show that, in any Bayesian equilibrium, the profits of the online platform are a function of the speed of learning of users. Using this result, we study the design of the review system by the online platform.

This is joint work with Daron Acemoglu, Ali Makhdoumi, and Azarakhsh Malekian.

Bio:

Asu Ozdaglar received the B.S. degree in electrical engineering from the Middle East Technical University, Ankara, Turkey, in 1996, and the S.M. and the Ph.D. degrees in electrical engineering and computer science from the Massachusetts Institute of Technology, Cambridge, in 1998 and 2003, respectively.

She is the Joseph F. and Nancy P. Keithley Professor of Electrical Engineering and Computer Science (EECS) Department at the Massachusetts Institute of Technology. She is also the associate head of EECS. Her research expertise includes optimization theory, with emphasis on nonlinear programming and convex analysis, game theory, with applications in communication, social, and economic networks, distributed optimization and control, and network analysis with special emphasis on contagious processes, systemic risk and dynamic control.

Professor Ozdaglar is the recipient of a Microsoft fellowship, the MIT Graduate Student Council Teaching award, the NSF Career award, the 2008 Donald P. Eckman award of the American Automatic Control Council, the Class of 1943 Career Development Chair, the inaugural Steven and Renee Innovation Fellowship, and the 2014 Spira teaching award. She served on the Board of Governors of the Control System Society in 2010 and was an associate editor for IEEE Transactions on Automatic Control. She is currently the area co-editor for a new area for the journal Operations Research, entitled "Games, Information and Networks. She is the co-author of the book entitled “Convex Analysis and Optimization” (Athena Scientific, 2003).

Talk Details:

Monday, July 10th, 2017

9:00am - 10:00am

White Auditorium, 2nd Floor

Yuliy Sannikov (Stanford)

Title:Dynamic Contracts

Abstract:Dynamic incentive problems occupy an important place in economics. They appear in many fields. In macroeconomics incentives pose constraints that lead to inequality. In corporate finance, incentives problems lead financial frictions and impose limits on optimal capital allocation. This talk will address the problem of dynamic incentives through the lens of a continuous-time principal agent model. The agent puts in effort, which is observable only imperfectly, and the principal wishes to design the best contract to motivate the agent. The analysis of this problem involves double dynamic optimization: the principal designs the optimal dynamic contract recognizing that the agent will optimize with respect to an effort strategy given the contract. The principal's optimization problem has to use an endogenous state space, with variables sufficient to summarize the agent's incentives. The optimal contract exhibits inefficiencies: the agent has to face risk, inefficient termination may occur with positive probability, and various distortions may need to be imposed to control the agent's value of private information.

Bio:

Yuliy Sannikov is a theorist who has developed new methods for analyzing continuous time dynamic games using stochastic calculus methods. His work has not only broken new ground in methodology, it has had a substantial influence on applied theory. He has significantly altered the toolbox available for studying dynamic games, and as a result of his contributions, new areas of economic inquiry have become tractable for rigorous theoretical analysis. The areas of application include the design of securities, contract theory, macroeconomics with financial frictions, market microstructure, and collusion. Sannikov’s work is impressive. It is elegant, powerful, and it paves the way for further analysis on lots of problems. The early successes highlighted how even simple and well-studied models could yield new insight. His most recent work has tackled more complex models in finance and macroeconomics. Previous models abstracted from crucial economic forces in the name of tractability, but Sannikov’s methods allow models to include the most important forces and thus deliver results that are much more relevant. He is one of the few theorists in many years to have introduced a truly novel tool that changed the way theory is done.

Talk Details:

Tuesday, July 11th, 20179:00am - 10:00amWhite Auditorium, 2nd Floor

R. Srikant (UIUC)

Title:Approximate Graph Matching on Random Graphs

Abstract:

We consider an abstraction of the network deanonymization problem, where the goal is to infer the node identities in an anonymized graph by observing the node identities and the topology of a correlated graph. More precisely, the goal is to label the nodes of the anonymized graph so that the adjacency matrix of the anonymized graph matches the adjacency matrix of the correlated graph as closely as possible. We will review prior results on this problem for the case of ErdÅ‘s–Rényi graphs, and present some new results for stochastic block models. Joint work with Joseph Lubars.

Bio:

R. Srikant is the Fredric G. and Elizabeth H. Nearing Endowed Professor of Electrical and Computer Engineering and a Professor in the Coordinated Science Lab, both at the University of Illinois at Urbana-Champaign. His research interests include communication networks, machine learning, and applied probability. He served as the Editor-in-Chief of the IEEE/ACM Transactions on Networking from 2013-2017. He is the winner of several Best Paper awards, and is a recipient of the IEEE INFOCOM Achievement Award.

Talk Details:

Wednesday, July 12th, 2017

12:15pm - 1:15pm

White Auditorium, 2nd Floor

Marcel Neuts Lecture:

Colm O’Cinneide (QS Investors)

Title:Phase-type Distributions and Invariant Polytopes.

Abstract:

A phase-type distribution is the distribution of a hitting time in a finite-state Markov chain. Phase-type distributions are a fundamental building block of matrix-analytic methods, the framework for stochastic modeling that Marcel Neuts pioneered. An invariant polytope is defined as a bounded convex set with a finite number of extreme points that is mapped to itself by a given linear transformation. Phase-type distributions and invariant polytopes are closely linked, and the simple geometry of the latter gives insights into the former. Exploring this link leads to a characterization of all phase-type distributions and to insights into how the properties of a particular phase-type distribution may place restrictions on a Markov chain representation of that distribution. Marcel introduced phase-type distributions into stochastic modeling over four decades ago, and yet there are some interesting questions still awaiting answers.

Bio:

Colm O'Cinneide has worked in the investment management industry since he joined Deutsche Asset Management in 2000. He was a partner at QS Investors when it spun off in 2010, and currently is head of portfolio construction. He held faculty positions at the University of Arkansas and Purdue University from 1983 to 2000. He is currently an adjunct professor in the mathematical finance program at Columbia University, where he co-teaches a course on portfolio management. From 2009 to 2010 he served as the president of the Society of Quantitative Analysts. He holds a PhD in Statistics from the University of Kentucky.

Talk Details:

Tuesday, July 11th, 201712:15pm - 1:15pmWhite Auditorium, 2nd Floor

Tutorials:Jose Blanchet (Stanford)

Title:Optimal Transport Methods in Stochastic Operations Research and Statistics

Abstract:

In this tutorial we will review recent results at the intersection of optimal transport, stochastic OR and statistics. After reviewing basic notions of optimal transport costs and Wasserstein distances, we will discuss distributionally robust performance analysis and optimization results. For example, we will show how the theory of diffusion approximations can be harnessed in this setting to provide model-robust sample path estimates for general stochastic systems. In addition, using the same mathematical principles, we will show how many machine learning algorithms such as sqrt- Lasso

, regularized logistic regression, group Lasso, adaptive Lasso, support vector machines, among many others admit distrivutionally robust representations based on optimal transport costs. Finally, we also introduce model statistical methodology which can be used to optimally choose the uncertainty size in distrivutionally robust formulations.**This tutorial is based on work with Yang Kang and Karthyek Murray.

Bio:

Jose Blanchet is a faculty member in the departments of IEOR and Statistics at Columbia University. Jose holds a Ph.D. in Management Science and Engineering from Stanford University. Prior to joining Columbia he was a faculty member in the Statistics Department at Harvard University. Jose is a recipient of the 2009 Best Publication Award given by the INFORMS Applied Probability Society and of the 2010 Erlang Prize. He also received a PECASE award given by NSF in 2010. He worked as an analyst in Prote go Financial Advisors, a leading investment bank in Mexico. He has research interests in applied probability and Monte Carlo methods. He serves in the editorial board of Advances in Applied Probability, Journal of Applied Probability, Mathematics of Operations Research, QUESTA, Stochastic Models, and Stochastic Systems.

Talk Details:

Monday, July 10th, 2017

1:30pm - 3:00pm

Room L - 130

Bert Zwart (CWI)

Title:Selected Topics on the Interface of Power Systems and Applied robability

Abstract:The green revolution is irreversible, and various academic studies and policies point towards a society where all energy, or at least the vast majority of all electricity, will be generated by renewable energy sources in 2050.

My vision is that applied probabilists can contribute to such a society: many outstanding problems require the development of fundamental rather than incremental research, and uncertainty will play a key role.

After giving a short introduction into the physics, economics and control of the power grid, I will give an overview of several recent research results and opportunitiies that are of interest to applied probabilists. A tentative overview is as follows:

- the physics and economics of the power grid

- traffic models for solar and wind power

- performance and control of storage devices using renewable input

- interacting energy markets and nodal pricing

- stochastic control for demand response

- electrical vehicle charging

- reliability

*No background knowledge on power systems is assumed.

Bert Zwart is a researcher at CWI, where he leads the Stochastics group. He also holds secondary positions at Eindhoven University of Technology (Professor), and Georgia Tech (Adjunct Professor). Previously, he held a Coca-Cola Chair at the H. Milton Stewart School of Industrial and Systems Engineering at the Georgia Institute of Technology. His research is in applied probability and stochastic operations research, inspired by problems in computer, communication, energy and service networks. Zwart is the 2008 recipient of the Erlang prize for outstanding contributions to applied probability by a researcher not older than 35 years old, an IBM faculty award, VENI, VIDI and VICI awards from the Dutch Science Foundation (NWO), the 2015 Van Dantzig award, and numerous best papers awards. He has co-authored more than 100 refereed publications, has been area editor of Stochastic Models for the journal

Bio:

Operations Researchsince 2009, and serves on several additional journal boards and TPCs.

Talk Details:

Wednesday, July 12th, 2017

10:30pm - 12:00pm

Room L - 130

Jose Blanchet (Columbia)

Title:Optimal Transport Methods in Stochastic Operations Research and Statistics

Abstract:

In this tutorial we will review recent results at the intersection of optimal transport, stochastic OR and statistics. After reviewing basic notions of optimal transport costs and Wasserstein distances, we will discuss distributionally robust performance analysis and optimization results. For example, we will show how the theory of diffusion approximations can be harnessed in this setting to provide model-robust sample path estimates for general stochastic systems. In addition, using the same mathematical principles, we will show how many machine learning algorithms such as dart-Lasso, regularized logistic regression, group Lasso, adaptive Lasso, support vector machines, among many others admit distrivutionally robust representations based on optimal transport costs. Finally, we also introduce model statistical methodology which can be used to optimally choose the uncertainty size in distrivutionally robust formulations.*

*This tutorial is based on work with Yang Kang and Karthyek Murray.

Bio:

Jose Blanchet is a faculty member in the departments of IEOR and Statistics at Columbia University. Jose holds a Ph.D. in Management Science and Engineering from Stanford University. Prior to joining Columbia he was a faculty member in the Statistics Department at Harvard University. Jose is a recipient of the 2009 Best Publication Award given by the INFORMS Applied Probability Society and of the 2010 Erlang Prize. He also received a PECASE award given by NSF in 2010. He worked as an analyst in Prote go Financial Advisors, a leading investment bank in Mexico. He has research interests in applied probability and Monte Carlo methods. He serves in the editorial board of Advances in Applied Probability, Journal of Applied Probability, Mathematics of Operations Research, QUESTA, Stochastic Models, and Stochastic Systems.

Talk Details:

Monday, July 10th, 2017

1:30pm - 3:00pm

Room L - 130

Jose Blanchet (Columbia)

Title:Optimal Transport Methods in Stochastic Operations Research and Statistics

Abstract:

In this tutorial we will review recent results at the intersection of optimal transport, stochastic OR and statistics. After reviewing basic notions of optimal transport costs and Wasserstein distances, we will discuss distributionally robust performance analysis and optimization results. For example, we will show how the theory of diffusion approximations can be harnessed in this setting to provide model-robust sample path estimates for general stochastic systems. In addition, using the same mathematical principles, we will show how many machine learning algorithms such as dart-Lasso, regularized logistic regression, group Lasso, adaptive Lasso, support vector machines, among many others admit distrivutionally robust representations based on optimal transport costs. Finally, we also introduce model statistical methodology which can be used to optimally choose the uncertainty size in distrivutionally robust formulations.*

*This tutorial is based on work with Yang Kang and Karthyek Murray.

Bio:

Jose Blanchet is a faculty member in the departments of IEOR and Statistics at Columbia University. Jose holds a Ph.D. in Management Science and Engineering from Stanford University. Prior to joining Columbia he was a faculty member in the Statistics Department at Harvard University. Jose is a recipient of the 2009 Best Publication Award given by the INFORMS Applied Probability Society and of the 2010 Erlang Prize. He also received a PECASE award given by NSF in 2010. He worked as an analyst in Prote go Financial Advisors, a leading investment bank in Mexico. He has research interests in applied probability and Monte Carlo methods. He serves in the editorial board of Advances in Applied Probability, Journal of Applied Probability, Mathematics of Operations Research, QUESTA, Stochastic Models, and Stochastic Systems.

Talk Details:

Monday, July 10th, 2017

1:30pm - 3:00pm

Room L - 130

Jose Blanchet (Columbia)

Title:Optimal Transport Methods in Stochastic Operations Research and Statistics

Abstract:

In this tutorial we will review recent results at the intersection of optimal transport, stochastic OR and statistics. After reviewing basic notions of optimal transport costs and Wasserstein distances, we will discuss distributionally robust performance analysis and optimization results. For example, we will show how the theory of diffusion approximations can be harnessed in this setting to provide model-robust sample path estimates for general stochastic systems. In addition, using the same mathematical principles, we will show how many machine learning algorithms such as dart-Lasso, regularized logistic regression, group Lasso, adaptive Lasso, support vector machines, among many others admit distrivutionally robust representations based on optimal transport costs. Finally, we also introduce model statistical methodology which can be used to optimally choose the uncertainty size in distrivutionally robust formulations.*

*This tutorial is based on work with Yang Kang and Karthyek Murray.

Bio:

Jose Blanchet is a faculty member in the departments of IEOR and Statistics at Columbia University. Jose holds a Ph.D. in Management Science and Engineering from Stanford University. Prior to joining Columbia he was a faculty member in the Statistics Department at Harvard University. Jose is a recipient of the 2009 Best Publication Award given by the INFORMS Applied Probability Society and of the 2010 Erlang Prize. He also received a PECASE award given by NSF in 2010. He worked as an analyst in Prote go Financial Advisors, a leading investment bank in Mexico. He has research interests in applied probability and Monte Carlo methods. He serves in the editorial board of Advances in Applied Probability, Journal of Applied Probability, Mathematics of Operations Research, QUESTA, Stochastic Models, and Stochastic Systems.

Talk Details:

Monday, July 10th, 2017

1:30pm - 3:00pm

Room L - 130

Title:Optimal Transport Methods in Stochastic Operations Research and Statistics

Abstract:

In this tutorial we will review recent results at the intersection of optimal transport, stochastic OR and statistics. After reviewing basic notions of optimal transport costs and Wasserstein distances, we will discuss distributionally robust performance analysis and optimization results. For example, we will show how the theory of diffusion approximations can be harnessed in this setting to provide model-robust sample path estimates for general stochastic systems. In addition, using the same mathematical principles, we will show how many machine learning algorithms such as dart-Lasso, regularized logistic regression, group Lasso, adaptive Lasso, support vector machines, among many others admit distrivutionally robust representations based on optimal transport costs. Finally, we also introduce model statistical methodology which can be used to optimally choose the uncertainty size in distrivutionally robust formulations.*

*This tutorial is based on work with Yang Kang and Karthyek Murray.

Bio:

Jose Blanchet is a faculty member in the departments of IEOR and Statistics at Columbia University. Jose holds a Ph.D. in Management Science and Engineering from Stanford University. Prior to joining Columbia he was a faculty member in the Statistics Department at Harvard University. Jose is a recipient of the 2009 Best Publication Award given by the INFORMS Applied Probability Society and of the 2010 Erlang Prize. He also received a PECASE award given by NSF in 2010. He worked as an analyst in Prote go Financial Advisors, a leading investment bank in Mexico. He has research interests in applied probability and Monte Carlo methods. He serves in the editorial board of Advances in Applied Probability, Journal of Applied Probability, Mathematics of Operations Research, QUESTA, Stochastic Models, and Stochastic Systems.

Talk Details:

Monday, July 10th, 2017

1:30pm - 3:00pm

Room L - 130