讲演摘要

Data shuffling of training data among different computing nodes (or workers) has been identified as a core element to improve the statistical performance of modern large scale machine learning algorithms. Data shuffling is often considered to be a significant bottleneck in such systems due to the heavy communication load it imposes on the system. Under a master-worker architecture (where a master node has access to the entire dataset, and only communications between the master and workers is allowed) coding has been recently proved to considerably reduce the communication load. In this talk, we consider a different communication paradigm referred to as ``distributed data shuffling,'' where workers are allowed to communicate with one another while no communication between the master and workers is allowed. Under the constraint of uncoded cache placement, we propose a converse bound and several achievable schemes, and overall optimality to within a factor of 3/2. The proposed schemes are also exactly optimal under the constraint of uncoded cache placement for either large memory size or for systems with at most four workers. As a by-product, we also propose a novel distributed clique covering scheme, for distributed broadcast problem with side information, which can be applied to distributed index coding problems, device-to-device caching problems, etc.

讲演者简介

Daniela Tuninetti is currently a Professor within the Department of Electrical and Computer Engineering (ECE) at the University of Illinois at Chicago (UIC), which she joined in 2005. Dr. Tuninetti got her Ph.D. in Electrical Engineering in 2002 from ENST/Telecom ParisTech (Paris, France, with work done at the Eurecom Institute in Sophia Antipolis, France), and she was a postdoctoral research associate at the School of Communication and Computer Science at the Swiss Federal Institute of Technology in Lausanne (EPFL, Lausanne, Switzerland) from 2002 to 2004. Dr. Tuninetti is a recipient of a best paper award at the European Wireless Conference in 2002, of an NSF CAREER award in 2007, and named UIC University Scholar in 2015. Dr. Tuninetti was the editor-in-chief of the IEEE Information Theory Society Newsletter from 2006 to 2008; she was an editor for IEEE COMMUNICATION LETTERS from 2006 to 2009, for IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS from 2011 to 2014, and for IEEE TRANSACTIONS ON INFORMATION THEORY from 2015 to 2017. Dr. Tuninetti's research interests are in the ultimate performance limits of wireless interference networks (with special emphasis on cognition and user cooperation), coexistence between radar and communication systems, multi-relay networks, content-type coding, and caching systems.