Summer School on "Online Learning with Applications to Digital Markets"
29 August 2022 - 2 September 2022Aims
The “Summer School of Mathematics for Economic and Social Sciences” aims to improve the knowledge of mathematical methods among graduate students in economics and social sciences, with a focus on those techniques which albeit widespread in use are not properly covered in typical graduate programs. The School is an interdisciplinary venue intended to foster the interaction of people coming from the too often separated communities of mathematical and social scientists.
It is funded and organized by the Mathematics Research Center “Ennio De Giorgi” and partially supported by the International Doctoral Program in Economics of the Scuola Superiore Sant'Anna.
Research Directions
- Description
The explosive growth of online markets has created complex ecosystems of algorithmic agents. To optimize their revenue, agents need to understand how the market works, and to do so they often resort to strategies that learn from past observations. In this course we introduce Online Convex Optimization, the main algorithmic framework for the study of sequential decision-making problems. The first part of the course describes the most important settings and algorithms for online learning. In the second part, we review some recent results characterizing the strengths and limitations of sequential decision-making approaches applied to various problems arising in digital markets. The analysis sheds light on how learning depends on the interplay between the form of the revenue function and the feedback provided during the learning process.
- Venue:
Fondazione Conservatorio Santa Chiara, San Miniato, Italy
- Lecturer:
Nicolò Cesa-Bianchi, Università degli Studi di Milano
Pierre Laforgue, Università degli Studi di Milano
- Topics:
- The online learning framework: introduction and motivation
- Sequential risk and regret
- Follow the Leader: linear lower bound
- Follow the Regularized Leader (FTRL)
- The Euclidean case: Online Gradient Descent
- The entropic case: Exponentiated Gradient (EG)
- Proving the FTRL regret bound
- Parameter-free algorithms
- Euclidean vs entropic regularization
- EG with linear losses: experts, bandits, and feedback graphs
- Multitask online learning
- Regret minimization and equilibria
- Dynamic pricing, retail pricing, and adaptive taxation
- Auctions
- Bilateral Trading
Please view Planned activities and the Syllabus for more details.
- Participation
Participation is subject to selection. At the moment the number of participants is limited to 14, but it should be increased if the pandemic situation would improve. Financial support for board and accommodation will be provided.
- Application
On-line application should be made through the Financial support section if support is requested, or through the Registration section if support is NOT requested.
All applications must include a CV that shall be sent by e-mail to crm@sns.it.
Applications without a CV will not be considered.
Deadline for application: 7th July 2022