equiprice_lunch-seminars
EQUIPRICE lunch seminars
Equiprice lunch seminars are hybrid (in person and / or virtual, depending on the COVID context) lunch seminars designed to serve as technical tutorials or presentation of work in progress in relation to the scientific agenda of the ERC-sponsored project EQUIPRICE. Unless noted ‘internals,’ seminars are public and open to all, but registration to the in person event is required 48 hours before.
Note: All times indicated on this page refer to the Paris time zone.
Upcoming talks:
- May 19 2022, 12pm-1pm (online), Dr. Adrian Vladu (IRIF) will present TBA.
Abstract: TBA.
Past talks:
- February 10 2022, 12pm-1pm (online), Mert Unsal (Ecole Polytechnique) will present ‘running C++ on jupyter notebook: the xeus cling kernel’.
Abstract: TBA.
- February 3 2022, 12pm-1pm (online), Professor Claire Mathieu (CNRS, Irif) will present “Stable Matching in Practice”.Abstract: Stable matching methods, based on the algorithm designed by Gale and Shapley, are used around the world in many applications such as college admissions. Several criteria measure the quality of the result: number of students assigned; rank of the college assigned to the applicant in their preference list; robustness; running time; etc.
After reviewing properties of the algorithm in the pure, ideal setting, we present issues arising in practice. The input data is uncertain and evolves with time, so a one-shot algorithm does not suffice. It is not feasible for admission committees to meet continuously, so the process cannot be fully dynamic. To reconcile those competing constraints, a hybrid implementation proceeding partly online on the student side was recently proposed for college admissions in France.
Finally, after remarking that the men-optimal stable matching and the women-optimal stable matching are almost identical, we propose a probabilistic model for highly correlated preferences to provide a theoretical explanation.
This is joint work with Hugo Gimbert and Simon Mauras. - ***Wednesday*** January 12 2022, *** 2pm-3pm *** (online), Dr. Antoine Jeanjean (OPT2A) will present “Operations research: a love story between mathematics and computer science.”
Abstract: From the optimization of highway planning to the minimization of formwork stocks on construction sites, through the optimization of the planning of advertising screens, Operations Research is a discipline that deals with the development of advanced analytical methods to improve decision-making. We will illustrate the discipline through concrete examples from real projects and we will also present some methods and tools used to solve them, in particular local search.
- December 16 2021, 12pm (online), Dr. Denis Merigoux (INRIA) will present “Turning law into micro-simulation code with the Catala programming language”.Abstract: Software called legal expert systems are used around the world by private and public organizations to compute taxes. A bug in such programs can lead to tax miscalculations and heavy legal and democratic consequences. Yet, increasing evidence suggests that some legal expert systems may not meet satisfying criteria to be in compliance with the law. Moreover, they are difficult to adapt to the continuous flow of new legislation just by using traditional software development processes. To prevent further software decay and reconcile these systems with the growing demand for algorithmic transparency and economic micro-simulation of reform impact, we present a solution built by lawyers and computer scientists : a new programming language, Catala, coupled with a pair programming development process.
- December 9 2021, 12pm (online), Anatole Gallouet (Grenoble INP) will present “Numerical resolution of semi-discrete Generated Jacobian equations”.Abstract: In non-imaging optics, we try to optimize the trajectory of the light from a source to a target without trying to form an image of the source on the target. Some non-imaging optic problems can be translated into optimal transport problems in a semi-discrete setting, meaning that the source is a continuous domain and the target is a finite set of points. Other problems of non-imaging optics can sometimes be rewritten into a slightly more global form than optimal transport, which we call Generated Jacobian Equations. During this presentation we will focus on these Generated Jacobian Equations, also in a semi-discrete setting. We will begin by introducing them using a non-imaging optic problem, and make the link with optimal transport. We will then present an algorithm to solve Generated Jacobian Equations, which was adapted from an existing algorithm to solve optimal transport problems. And finally we will detail the main lines of the proof of convergence of this algorithm.
- November 25 2021, 12pm (hybrid), Professor Benoit Rottembourg (INRIA) will present “Dynamic pricing: constraints, elasticity and algorithmic manipulation”. Abstract: Dynamic pricing was popularized in the 1980s by the U.S. airline industry, in the midst of deregulation. It consisted above all in filling the aircraft, by optimizing the fare mix under capacity constraints. However, the customer’s vision was very crude; this was before the Internet era and it was the travel agent who was in charge of the search and booking. The same techniques, based on customer segmentation, demand forecasting and (stochastic) mix optimization, have been successfully propagated to other market sectors such as rail transport, cruises, hotels or even very recently container transport.
In this presentation, we will aim at showing how dynamic pricing as applied by large digital platforms (market places, online travel agencies, etc.) offers a breakthrough in the way of anticipating and influencing customer behavior through price. We will insist on the fact that this power of influence, boosted by new families of algorithms, is powerful but can generate abuses. These abuses may concern consumers’ law, suppliers’ rights and potentially competition law. We will try to show the difficulties to audit this kind of practices and the challenges for the regulatory authorities. We willargue for new supervisory frameworks, as outlined in the Digital Market Act proposals at the European level. - November 18 2021, 12pm (hybrid). Professor César Ducruet (Paris-Nanterre University) will present “Inter-city networks: a maritime perspective”.Abstract: Although more than 80% of world trade volumes occur by sea, and about 60% of world urban population is coastal, no previous research has been done on how cities are linked by maritime flows. Based on untapped shipping records and population data, we construct a global network with cities as nodes and inter-port vessel voyages as links. The first part of the analysis looks at the evolution of traffic concentration and port-city correlation since the late nineteenth century. While the largest cities kept concentrating traffic in a stable manner, the correlation gradually lost in significance. The latter trend, however, varied according to the type of location and the level of observation.
The second part on the maritime connectivity of cities confirms the overwhelming importance of city size, but sheds more light on the shift of hub functions towards pivotal ports and from Atlantic to Asia-Pacific. A spatial interaction model shows that the network obeys gravitational properties, as larger cities connect more with each other,
but less at distance. - October 7, 2021. Reading of Roth, Rothblum and Vande Vate’s Stable Matchings, Optimal Assignments, and Linear Programming.
- September 12 2021, 12:30pm. Pauline Corblet’s job market paper’s practice talk.
- July 15 2021, 1pm. Loan Tricot on Bertsekas’ auction algorithm.
- ***WEDNESDAY*** June 23 2021, ***12.30pm-2pm***. Pauline Corblet. Job market practice talk.
- June 17, 1pm. Loan Tricot on “Reconstructing the order flow with optimal transport”.Abstract: Exchanges receive three types of orders:
- Limit orders, making liquidity
- Market orders, taking liquidity
- Cancel orders, taking liquidity
Limit orders are standing offers to buy (sell) a defined amount at a defined price. They can be cancelled using Cancel orders. Market orders are orders to buy a defined amount at market price: the best price which can be achieved using standing offers. Limit orders are accumulated and aggregated into a tick level order book, which contains the size of aggregated standing offers at each price. Market orders trigger one or more trades when they are matched with one or more Limit orders. Exchanges stream the order book updates and the trades flow to market participants, but never stream the order flow. Order book updates and trades are usually not aligned (w.r.t. time) and sometimes presented in aggregate form. The large number of, and lack of standardization between, (crypto) exchanges means there are no guarantees concerning the quality of the data. The order flow is valuable data which can be used to study agents’ behavior on the market, and possibly design algorithmic trading strategies. We attempt to reconstruct the order flow from the order book updates and trades flow.
- June 3 2021, 1pm. Bastien Patras on BLP and MPEC.
- *** Tuesday*** May 25 2021, 1pm. Reading of The Contraction Mapping Approach to the Perron-Frobenius Theory: Why Hilbert’s Metric?.
- May 20 2021, 1pm (public). Professor Guillaume Carlier (Paris-Dauphine University), on “the linear convergence of the multi-marginal Sinkhorn algorithm.”
- April 29 2021, 1pm. Reading of Bubeck and Cesa-Bianchi’s Regret Analysis of Stochastic and Nonstochastic Multi-armed Bandit Problems chapters 1-3.
- April 15 2021, 1pm. Flavien Léger on Bregman divergence and gradient descent (part 2).
- April 8 2021, 1pm. Flavien Léger on Bregman divergence and gradient descent (part 1).
- March 25 2021, 1pm. Pauline Corblet on the link between Gale and Shapley and some iterative methods in applied mathematics.
- March 18 2021, 1pm. Jules Baudet on course allocation mechanisms (continued).
- March 4 2021, 1pm. Pauline Corblet on many-to-one matching problems.Abstract: TBC.
- February 18 2021, 1pm (public). Jules Baudet on course allocation mechanisms (continued).Abstract: Continued from Jan 7.
- February 11 2021, *** 5pm *** (public). Professor Xin Chen (University of Illinois) on S-convexity.Abstract: TBC.
- January 14 2021, 1pm (public). Flavien Léger on Chambolle-Pock.Abstract: TBC.
- January 7 2021, 1pm (public). Jules Baudet on “the multi-unit assignment problem: matching students to course schedules”. Abstract: In this talk, we will review a series of papers by Eric Budish and his co-authors on the multi-assignment problem, which consists in answering: how to optimally assign bundles of items to agents? In particular, we will study the allocation of course schedules to students.
First, we will see that traditional optimality criteria used for single-item matching are difficult to extend to combinatorial settings by looking at the flaws of the Harvard Course Allocation mechanism. Then, we will introduce the A-CEEI mechanism which offers better fairness, strategyproofness and efficiency guarantees.
The reference papers for this talk are:
– Budish, E. (2011). The combinatorial assignment problem: Approximate competitive equilibrium from equal incomes. Journal of Political Economy, 119(6), 1061-1103.
– Budish, E., & Cantillon, E. (2012). The multi-unit assignment problem: Theory and evidence from course allocation at Harvard. American Economic Review, 102(5), 2237-71.
– Budish, E., Cachon, G. P., Kessler, J. B., & Othman, A. (2017). Course match: A large-scale implementation of approximate competitive equilibrium from equal incomes for combinatorial allocation. Operations Research, 65(2), 314-336. - December 10 2020, *** 5pm *** (public). Professor Pierre-Olivier Weill (UCLA) on optimal transport in asset pricing problems.The talk will be based on the paper “Incentive Constrained Risk Sharing, Segmentation, and Asset Pricing” with Bruno Biais and Johan Hombert.
Abstract: Incentive problems make securities’ payoffs imperfectly pledgeable, limiting agents’ ability to issue liabilities. We analyze the equilibrium consequences of such endogenous incompleteness in a dynamic exchange economy. Because markets are endogenously incomplete, agents have different intertemporal marginal rates of substitution, so that they value assets differently. Consequently, agents hold different portfolios. This leads to endogenous markets segmentation, which we characterize with Optimal Transport methods. Moreover, there is a basis going always in the same direction: the price of a security is lower than that of replicating portfolios of long positions. Finally, equilibrium expected returns are concave in factor loadings. - November 26 2020, 1pm (public). Flavien Léger on computing large-scale regularized optimal transport problems.Abstract: We shall describe some aspects of handling large scale problems, in Python.
- November 12 2020, 1pm. Alfred Galichon on ‘EQUIPRICE: research program and challenges’.Abstract: In this talk, I will outline the broader research program of EQUIPRICE, and I will describe some scientific challenges and some projects for EQUIPRICE in 2020-2021.
- October 29 2020, 1pm (public). Jules Baudet will be presenting on ‘Cloud Computing and Containers part 3: Understanding Docker containers, continued’.
- October 22 2020, 1pm (public). Jules Baudet will be presenting on ‘Cloud Computing and Containers part 2: Understanding Docker containers’ Abstract: In this second session (continued from part 1), we will demonstrate how to create a Virtual Machine on Google Cloud, on which we will deploy a container containing Jupyter notebooks. We will also show how to access and run these notebooks. We will conclude by comparing Google Colab to Google Cloud for running notebooks.
- October 15 2020, 1pm (public). Jules Baudet on kidney exchanges. Abstract: Re-organizing the French Kidney Exchange Program
Jules BAUDET
Market Design is an area of Economics that leverages Game Theory, Experimental Economics and Algorithms to fix market failures by proposing implementable solutions. In 2012, Alvin Roth received the Nobel Prize for his work, in which he studied and helped fix essential matching markets (doctors residency match, school seats allocation…). One of his most famous work is the design of a Kidney Paired Donation Program, allowing to overcome Patient-Donor incompatibilities by exchanging donors between pairs. In this talk, we will analyze and propose solutions to the operational challenges faced by the French Agency for Biomedicine in creating a successful Kidney Paired Donation (KPD) Program. We will study the impact of parameters such as patient priority criteria, size constraints on exchanges and frequency of match runs on the total number of grafts. This talk will underline the dynamic tradeoffs at stake in designing a KPD program, and the importance of conducting regular simulations to adapt the organization of the program to the structure of its pool of participants.
This talk will be based on the following paper. - October 8 2020, 1pm (public). Flavien Léger on ‘The back-and-forth method in optimal transport’Abstract: For this first lunch seminar we will follow the ‘‘math+econ+code’’ design. We will present the back-and-forth method [1], a recent state-of-the-art algorithm to solve optimal transport problems.
We will then offer a brief introduction to KeOps [2], an easy-to-use library with a python interface to perform large-scale kernel operations on GPUs. We will implement the back-and-forth method on optimal transport, hedonic equilibrium and equilibrium transport problems.
[1] Matt Jacobs and Flavien Léger. A fast approach to optimal transport: The back-and-forth method. Numerische Mathematik, 2020. To Appear.
[2] https://www.kernel-operations.io/keops/index.html - September 24 2020, 1pm (public). Jules Baudet on ‘Cloud Computing and Containers part 1: Understanding Docker containers’Abstract: Abstract: This series of talks aims at introducing and showcasing the basic functionalities of Virtual Machines and Docker Containers. Through live demonstrations, we will see how to bundle an application and all of its dependencies in a container before deploying it in the cloud, on a Virtual Machine. In particular, we will show you how to run Jupyter notebooks located inside containers on Virtual Machines. Note: If you would like to follow along during the tutorials, you will need to create a free account on Google Cloud Platform: https://cloud.google.com/?hl=en and to install docker on your computer: https://docs.docker.com/desktop/. In a first session, we will start by introducing Docker containers and comparing them to Virtual Machines. Containers are a technology that revolutionized the computing industry by allowing developers to bundle an application and all its dependencies in a single structure. We will then show how to create, build, and deploy a Docker container. The demonstration will end by sending our Docker Container to Google Cloud Container Registry, in order to be accessible for our following tutorial.
Sponsored by the European Research Council grant EQUIPRICE
