Bat. Alan Turing
1 rue Honoré d'Estienne d'Orves
Campus de l'École Polytechnique
91128 Palaiseau, FRANCE
+33(0)1 69 35 69 75
One generic question is how to enforce desirable global properties by tweaking the local interaction rules between system components? Here, interaction rules are dictated by communication protocols or distributed algorithms, and can thus be changed. This contrasts with the classical context of statistical physics, where one cannot change the Laws of Nature.
For Internet congestion control and information propagation in peer-to-peer systems, "desirable global properties" could mean maximally efficient use of communication resources. Related work can be found in my "HDR" (Habilitation à Diriger des Recherches) thesis "Information flows through networks: models and algorithms" (slides available here). Models of networks based on random graphs and epidemic propagation can be found in our book "Epidemics and Rumours in Complex Networks" co-authored with Moez Draief at Cambridge University Press in the London Mathematical Society Lecture Note Series.
Content placement conditions how efficiently bandwidth resources can be used to absorb demands for content. Recent work on performance of various content placement schemes can be found here and there.
There is a trade-off between how efficiently one can learn from individuals' private data and how protected their privacy is. An attempt at understanding this trade-off for learning to perform recommendation and when privacy protection is formalized by "differential privacy" can be found here.
"Search through comparison" is an alternative to keyword-based content search which, in addition to being of practical interest, also highlights links between information retrieval and information theory. Some recent work on its performance limitations and related algorithms can be found here.