View the full talk:
Speaker: Aurelien Bellet (INRIA, France)
With the advent of connected devices with computation and storage capabilities, it becomes possible to run machine learning on-device to provide personalized services. However, the currently dominant approach is to centralize data from all users on an external server for batch processing, sometimes without explicit consent from users and with little oversight. This centralization poses important privacy issues in applications involving sensitive data such as speech, medical records or geolocation logs.
In this talk, we will discuss an alternative setting where many agents with local datasets collaborate to learn personalized models by engaging in a fully decentralized peer-to-peer network. We introduce and analyze asynchronous algorithms that allow agents to improve upon their locally trained model by exchanging information with other agents that have similar objectives. We will then describe how to make such algorithms differentially private to avoid leaking information about the local datasets, and analyze the resulting privacy-utility trade-off. These results will be illustrated by a set of numerical experiments.
Aurélien Bellet (INRIA, France) is a tenured researcher at INRIA, where he is a member of the project-team MAGNET (Machine Learning in Information Networks). He obtained his Ph.D. from the University of Saint-Etienne (France) in 2012 and, prior to joining INRIA, he was a postdoctoral researcher at the University of Southern California (USA) and at Télécom ParisTech (France). His main line of research is statistical machine learning, with particular interests in large-scale algorithms which allow good trade-offs between computational complexity (or other "resources", such as privacy or communication) and statistical performance. His work has been published in top machine learning venues such as ICML, NIPS and AISTATS.e.
The Alan Turing Institute
1st floor of the British Library, 96 Euston Road, London, NW1 2DB