Machine learning

At Technicolor, we designed and evaluated a distributed inference algorithm to provide recommendations for media such as movies or TV shows. Thanks to the Netflix Prize data set, we had ample material to evaluate our algorithm with. As opposed to the objective of the Netflix Prize, however, our goal was to offer provable privacy guarantees to all users. To accomplish this, we relied on two core tools. First, belief propagation on Bayesian networks allowed us to distribute computations. Second, differential privacy provided the provable privacy guarantees we were looking for. Differential privacy was originally developed at Microsoft Research; more recently, Apple has picked up differential privacy as a research focus.

Selected publications

Simon Heimlicher, Marc Lelarge, Laurent Massoulié:
Community Detection in the Labelled Stochastic Block Model
NIPS 2012 Workshop: Algorithmic and Statistical Approaches for Large Social Networks, Lake Tahoe, Nevada.
 Paper as PDF

Please refer to the complete list of publications for further information.

Photo by Robynne Hu on Unsplash