The Data Science department of EURECOM invites applications for a postdoc position at the intersection of game theory and machine learning, supervised by Prof. Patrick Loiseau (http://www.eurecom.fr/~loiseau).
Machine learning is at the core of many online services (social networks, recommendations, online surveys, etc.) that learn from individuals’ personal data. Personal data, however, is a special kind of data because it is often revealed directly by users who can strategically obfuscate it to protect their privacy. Standard machine learning algorithms are not adapted to such strategic data and this raises fundamental questions such as ‘how to best learn from strategic personal data?’, which gained tremendous interest in recent times.
Our core idea to approach the problem is to use game theory to model the system of strategic agents and use the game solutions to derive optimal learning algorithms. Recently, we proposed a new game-theoretic model, where users choose the precision of the data they disclose and learning accuracy is modeled as a public good, which allows thinking of learning problems while taking into account the strategic nature of personal data . We obtained preliminary results in simple cases [1, 2, 3], and notably showed that standard learning algorithms (for linear regression) perform badly in this context, but the main questions remain open. Building on this approach, the goal of the postdoc will be to extend the game-theoretic model and analyze the solutions to answer fundamental learning questions such as what is the optimal learning algorithm, or what is the rate of convergence (and conditions for consistency), etc. We will notably work on handling incomplete information using Bayesian games and global games.
Candidates must have (or be close to have) a PhD in CS, EE, mathematics or economics and a track record of research in game theory or in statistical learning. Familiarity with both topics will be a plus. Qualified applicants should send a complete resume including a list of publications and the names and addresses of three references to Patrick Loiseau (email@example.com). Interested applicants can also contact directly Patrick Loiseau if they have any question.
The postdoc will be hired for up to 18 months by EURECOM with a full-time contract supported by the CONNECTED project funded by ANR under the Tremplin-ERC program. The start date is negotiable depending on the constraints of the selected candidate. A start date early 2017 is preferred but the selection will be based on the candidate’s capabilities. Interested candidates are encouraged to apply as soon as possible to ensure full consideration of their application.
EURECOM is a graduate school and a research center in digital sciences, located in Sophia Antipolis, a vibrant science park on the French Riviera. It offers excellent working conditions and a competitive package of compensation and social benefits combined with the attractive living conditions of the French Riviera.
 S. Ioannidis, P. Loiseau. Linear Regression as a Non-Cooperative Game. WINE 2013.
 M. Chessa, J. Grossklags, P. Loiseau. A Game-Theoretic Study on Non-Monetary Incentives in Data Analytics Projects with Privacy Implications. CSF 2015.
 M. Chessa, J. Grossklags, P. Loiseau. A short paper on the incentives to share private information for population estimates, FC 2015.
(c) GdR 720 ISIS - CNRS - 2011-2015.