Johan van Leeuwaarden
Title: Load balancing algorithms
Cloud systems crucially depend on how imbalanced data can be smoothened. Load balancing algorithms (LBAs) dispatch jobs to one of N servers. Designing fast LBAs with good delay-communication tradeoffs is challenging, particularly for large N. Using coupling techniques, mean-field and diffusion limits we present universality classes of LBAs that achieve (almost) full resource pooling with minimal communication. We discuss algorithms that probe d(N) servers, Join-the-Idle Queue (JIQ) algorithms, and algorithms that operate on networks. Some results and many open problems.
Johan van Leeuwaarden (1978) is professor of Mathematics at Eindhoven University of Technology. He chairs the group Stochastic Networks and investigates phenomena arising in complex networks, such as communication networks, social networks and biological networks. Johan received an ERC Starting Grant (2010) and the Erlang Prize (2012) for his contributions to applied probability. In 2014 he co-founded the 10-year multidisciplinary research program NETWORKS, funded by the Dutch Government (www.thenetworkcenter.nl). Johan is member of the Young Academy (part of The Royal Netherlands Academy of Arts and Sciences), a group of scientists with outspoken views and the ambition to popularize science. Johan promotes the role of mathematics and data in the networked society.
Title: The limits of recommendations, and the role of social ties
The internet has become one of the main platforms for buying and selling cultural items. Songs, movies, books, sport events and much more can now be acquired and enjoyed via the internet. In this context, recommender systems play a crucial role, for increasingly they are becoming the main cognitive gateway between people and products.
A fundamental and fascinating question arises naturally: to what extent can a recommender system alter a market in which it is operating?
We present some research inspired by this question. First, we discuss a mathematical model of a market in which a recommender system is operating, with the aim of predicting what will happen to such a market in the long run. Second, we present a couple of user studies of the types of user-feedback commonly deployed online, whose outcome is somewhat surprising.
Joint work with Marzia Antenore, Marco Bressan, Giovanna Leone, Stefano Leucci, Prabhakar Raghavan, and Erisa Terolli