International Conference on Machine Learning for Networking (MLN'2018)
Invited speaker : Pierre Gaillard
Title: Distributed averaging of observations in a graph: the gossip problem
Abstract
Consider a network of agents connected by communication links, where each agent holds a real value. The gossip problem consists in estimating the average of the values diffused in the network in a distributed manner.
In this talk, I will present the basic Gossip and accelerated second-order methods that achieve better rates of convergence depending on the spectral measure of the network graph. I will also present message passing algorithms (usually better when the network structure is a tree) and extensions when the observations are i.i.d. samples from the same distribution.
Biography
Pierre Gaillard is a researcher at INRIA Paris, member of the SIERRA project-team since January 2017. Until 2015, he carried out an industrial PhD at EDF R&D on forecasting the electricity consumption by aggregating expert advice. His work mainly focus on theoretical online learning.