Yannick parle

Organizing committee : Yannick

Parle: parallelizing stochastic gradient descent

We propose a new algorithm called Parle for parallel training of deep networks that converges 2-4× faster than a data-parallel implementation of SGD, while achieving significantly improved error rates that are nearly state-of-the-art on several benchmarks including CIFAR-10 and CIFAR-100, without introducing any additional hyper-parameters. We exploit the phenomenon of flat minima that has been...

* , Joachim Kurtz 2 , Yannick Moret

One con *Autho Zoological Institute, University of Kiel, Am Botanischen Garten, 24098 Kiel, Germany Institute for Evolution and Biodiversity, University of Münster, Hüfferstrasse 1, 48149 Münster, Germany Université de Bourgogne, UMR5561 Biogéosciences, Équipe Écologie Évolutive, 6 Boulevard Gabriel, 21000 Dijon, France Department of Animal and Plant Science, University of Sheffield, Sheffield ...

Publication IV Mika Karjalainen , Juha Hyyppä , and Yannick Devillairs

This paper describes the potential of Radarsat-1 fine beam intensity SAR images in urban mapping and change detection. The main objective was to determine the possibility of building detection e.g. as a function of height, orientation, roof material and roof type in the Helsinki metropolitan region. Altogether 22235 buildings were used to estimate the effect of height, and then 14270 buildings ...

J ul 2 00 8 Properties and applications of dual reduction Yannick

The dual reduction process, introduced by Myerson, allows to reduce a finite game into a smaller dimensional game such that any equilibrium of the reduced game is an equilibrium of the original game. This holds both for Nash equilibrium and correlated equilibrium. We present examples of applications of dual reduction and argue that this is a useful tool to study Nash equilibria and correlated e...