BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:LASER: Linear Compression in Wireless Distributed Optimization
DTSTART:20231204T161500
DTEND:20231204T171500
DTSTAMP:20260404T071519Z
UID:111011ce46ee1f90c77812482bb478120d7cbfbf259a9032a42bd943
CATEGORIES:Conferences - Seminars
DESCRIPTION:Dr. Ashok Makkuva (LINX)\nAbstract: Data-parallel SGD is the 
 de facto algorithm for distributed optimization\, especially for large sca
 le machine learning. Despite its merits\, communication bottleneck is one 
 of its persistent issues. Most compression schemes to alleviate this eithe
 r assume noiseless communication links\, or fail to achieve good performan
 ce on practical tasks. In this work\, we close this gap and introduce LASE
 R: LineAr CompreSsion in WirEless DistRibuted Optimization. LASER capital
 izes on the inherent low-rank structure of gradients and transmits them ef
 ficiently over the noisy channels. Whilst enjoying theoretical guarantees 
 similar to those of the classical SGD\, LASER shows consistent gains over
  baselines on a variety of practical benchmarks. In particular\, it outper
 forms the state-of-the-art compression schemes on challenging computer vis
 ion and GPT language modeling tasks. On the latter\, we obtain 50-64% impr
 ovement in perplexity over our baselines for noisy channels. (Joint work w
 ith Marco Bondaschi\, Thijs Vogels\, Martin Jaggi\, Hyeji Kim\, and Michae
 l Gastpar.) \nBio: Ashok is a postdoctoral researcher at EPFL with Mic
 hael Gastpar. He obtained his PhD in ECE from the University of Illinois a
 t Urbana-Champaign in August 2022\, with Pramod Viswanath and Sewoong Oh
 .  He is a recipient of Best Paper Award from ACM MobiHoc 2019.  For mor
 e details about him\, please visit https://ashokvardhan.github.io/
LOCATION:BC 129 https://plan.epfl.ch/?room==BC%20129
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR