BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Optimal Transport and Deep Generative Models - Gabriel Peyre (CNRS
  - Ecole Normale Superieure Paris)
DTSTART:20171214T100000Z
DTEND:20171214T110000Z
UID:TALK96628@talks.cam.ac.uk
CONTACT:INI IT
DESCRIPTION:<span>Co-authors: Marco Cuturi		(ENSAE)\, Aude Genevay		(ENS) 
        <br></span><br>In this talk\, I will review some recent advances on
  deep generative models through the prism of Optimal Transport (OT). OT pr
 ovides a way to define robust loss functions to perform high dimensional d
 ensity fitting using generative models. This defines so called Minimum Kan
 torovitch Estimators (MKE) [1]. This approach is especially useful to reca
 st several unsupervised deep learning methods in a unifying framework. Mos
 t notably\, as shown respectively in [2\,3] (and reviewed in [4]) Variatio
 nal Autoencoders (VAE) and Generative Adversarial Networks (GAN) can be in
 terpreted as (respectively primal and and dual) approximate MKE. This is a
  joint work with Aude Genevay and Marco Cuturi. <br><br>References: [1] Fe
 derico Bassetti\, Antonella Bodini\, and Eugenio Regazzini. On minimum Kan
 torovich distance estimators. Statistics & probability letters\, 76(12):12
 98&ndash\;1302\, 2006. [2] Olivier Bousquet\, Sylvain Gelly\, Ilya Tolstik
 hin\, Carl-Johann Simon-Gabriel\, and Bernhard Schoelkopf. From optimal tr
 ansport to generative modeling: the VEGAN cookbook. Arxiv:1705.07642\, 201
 7. [3] Martin Arjovsky\, Soumith Chintala\, and L&eacute\;on Bottou. Wasse
 rstein GAN. Arxiv:1701.07875\, 2017. [4] Aude Genevay\, Gabriel Peyr&eacut
 e\;\, Marco Cuturi\, GAN and VAE from an Optimal Transport Point of View\,
  Arxiv:1706.01807\, 2017<br><br>Related Links<ul><li><a target="_blank" re
 l="nofollow" href="http://www-old.newton.ac.uk/cgi/https%3A%2F%2Farxiv.org
 %2Fabs%2F1706.01807">https://arxiv.org/abs/1706.01807</a> - GAN and VAE fr
 om an Optimal Transport Point of View</li></ul>
LOCATION:Seminar Room 1\, Newton Institute
END:VEVENT
END:VCALENDAR