Upcoming events

Reseaux de neurones bayesiens

Rafael présentera le papier "Renyi Differential Privacy"

P506

amphi Grenat (un étage en-dessous de l'entrée rue Barrault)

Seminar "Graph-Convolutional
Neural Networks" that will be held by Dr. Diego Valsesia (Politecnico di
Torino, Italy) on the 7th of March at 10h00, Télécom Paristech, 46 rue
Barrault (room to be announced). Please find below an abstract of the
seminar and feel free to circulate this invitation to whoever may concern.

Abstract:

Convolutional neural networks (CNNs) have enjoyed great success in tasks
such as image classification, object detection, etc thanks to weight
sharing, invariances to geometric transformations and hierarchical
decompositions induced by the convolution operations. However, many data
types of interest such as interactions in social networks, 3D point
clouds, biological data, etc. do not lie on regular grids like images do
and are better represented by graphs. A graph signal has scalar or
vector information on every node and edges describe relationships among
nodes. Extending the notion of convolution to this kind of signals
creates a new building block called graph convolution that can be used
in neural networks to solve new problems. Graph-convolutional neural
networks are a hot research topic and have been shown to be
state-of-the-art for problems such as fake news detection, link
prediction in networks, point cloud classification and generation that
are badly suited for traditional convolution. Latest research is also
showing that graph convolution can improve upon classic convolution even
for more traditional problems such as image segmentation and denoising.
This lecture will introduce concepts from graph signal processing to
present the approaches to define graph-convolutional neural networks. It
will then present their application to supervised, semi-supervised and
unsupervised problems.

A707

arxiv.org/pdf/1608.08967.pdf

Frederic Koriche
"Online combinatorial optimization with arithmetic circuits"

Abstract: In online optimization, the goal is to iteratively choose
solutions from a decision space, so as to minimize the average cost over
time. As long as this decision space is described by combinatorial
constraints, the problem is generally intractable. In this talk, we
shall discuss about two ideas: (i) using the paradigm of compilation for
encoding a set of combinatorial constraints into an arithmetic circuit,
and (ii) providing efficient characterizations of existing online
optimization algorithms, with a particular attention to Bregman
projections in mirror descent techniques. We shall conclude with open
questions about the extension of this framework to combinatorial
semi-bandits and bandits, and the potential interest of using arithmetic
circuits in structured prediction."

www.cril.univ-artois.fr/~koriche/

ENS-Data Science colloquium in Salle Jaures - March 25th @ 12.00 by Béatrice Prunel and Gregory Chatonsky: "Art and artificial imagination"

The seminar will take place at 12h00 in room Salle Jean Jaurès, 29 rue d’Ulm (sous-sol).

The colloquium will be followed by an open buffet around which participants can meet and discuss.

More information can be found on the web page of the seminar: data-ens.github.io/

These seminars and the buffet are being made possible through the support of the CFM-ENS Chair "Modèles et Sciences des Données".

The organizers:

Florent Krzakala, Stephane Mallat, Pascal Mamassian and Gabriel Peyré

———————

March 25th, 2019, 12h00-13h00, room Salle Jean Jaurès, 29 rue d’Ulm (sous-sol).
Béatrice Prunel (ENS) and Gregory Chatonsky
Title: Art and artificial imagination

Abstract: Contemporary media are fascinated by the applications of neural networks in creation. They regularly highlight moments when artificial artistic productions have "deceived" humans and "replaced" artists. All this seems to confirm that AI would have conquered up to the last ramparts of humanity: interiority and creativity. The dialogue between an art historian, invested in the digital humanities, and an artist who is himself familiar with deep learning, invites us to change our perspective. A historical and materialistic approach makes it possible to better distinguish what is new in the apparent emergence of AI in the arts and to better grasp the implicit conception of art that develops there: the change of purpose of a new technique, which generates surprising results, is also a way of thwarting the assumptions of the contemporary economic system. It suggest criticism of it as much as it opens up new possibilities.

Grégory Chatonsky est un artiste franco-canadien dont le travail pour sur Internet et l’imagination artificielle. Il a participé à de nombreuses expositions en France, au Canada et à l’étranger dont France Electronique à Toulouse, Terre/mer/signal au Rua Red de Dublin, Imprimer le monde en 2017 au Centre Pompidou, Capture : Submersion en 2016 à Arts Santa Mònica Barcelone, La condition post-photographique à Montréal, Walkers: Hollywood afterlives in art en 2015 au Museum of the Moving Image de New York, Telofossils en 2013 au Musée d’art contemporain de Taipei, Erreur d’impression en 2012 au Jeu de Paume. Il a été en résidence à Abou Dhabi (2017), en Amazonie à Taluen (2017), Colab à Auckland (2016), Hangar à Barcelone (2016), IMAL (2015), Villa Kujoyama (2014), CdA Enghein-les-Bains (2013), MOCA Taipei (2012), 3331 Arts Chiyoda (2012), Xiyitang, Shanghai, (2011), Les Inclassables à Montréal (2003), Abbaye royale de Fontevraud (2002). Il a reçu le prix Audi Talents en 2018 et est résident à la Cité Internationale des Arts de Paris en 2019-2020. Il a fondé en 1994 Incident.net, l’un des premiers collectifs de Netart en France. Il a été professeur-invité au Fresnoy (2004-2005), à l’UQAM (2007-2014), aété récipiendaire d’une chaire internationale de recherche à l’Université de Paris VIII (2015) et poursuit ses recherches à l’Ecole normale supérieure de Paris.

Béatrice Joyeux-Prunel est maître de conférences HDR en histoire de l’art contemporain à l’Ecole normale supérieure de Paris. Elle travaille sur l’histoire des arts dans une perspective mondiale, transnationale et sociologique, tout en coordonnant à l’ENS l’enseignement de l’histoire de l’art contemporain et des humanités numériques. Spécialiste de la mondialisation culturelle, elle s’intéresse également à l’histoire culturelle et visuelle du pétrole, et à l’imaginaire des technologies numériques. Béatrice Joyeux-Prunel a fondé et dirige le projet ARTL@S. Elle coordonne avec Grégory Chatonsky le projet Postdigital. Parmi ses publications : Les avant-gardes artistiques – une histoire transnationale 1848-1918, Paris, Gallimard Folio histoire (inédit poche), 2016 ; volume 2 (1918-1945), paru en 2017 ; et volume 3 (1945-1970), à paraître en 2019.

Talk about difference between Actor-Critic and Advantage Actor-Critic (A2C and A3C)
A2C methods
Peters et al, Natural Actor-Critic, (2005)
homes.cs.washington.ed.....c.pdf
Bhatnagar et al, Natural Actor–Critic Algorithms, (2009)
hal.inria.fr/hal-00840470/document
Grondman et al, A survey of actor-critic reinforcement learning: standard and natural policy gradients
(2012)
hal.archives-ouvertes......ument
Volodymyr et al, Asynchronous Methods for Deep Reinforcement Learning, (2016), ICML
proceedings.mlr.press/v48/mniha16.html
lien google drive
drive.google.com/open?.....sFDsT

NB : dejeuner avec Vincent puis séminaire
Vincent Cohen-Addad
##
Title : Hierarchical clustering: Objective functions and algorithms
Date : Apris 9 - 14h00
Room : Salle C (2eme étage)
Abstract : Hierarchical clustering is a recursive partitioning of a dataset into clusters at an increasingly finer granularity. Motivated by the fact that most work on hierarchical clustering was based on providing algorithms, rather than optimizing a specific objective, Dasgupta framed similarity-based hierarchical clustering as a combinatorial optimization problem, where a `good’ hierarchical clustering is one that minimizes some cost function. He showed that this cost function has certain desirable properties.We take an axiomatic approach to defining `good’ objective functions for both similarity and dissimilarity-based hierarchical clustering. We characterize a set of “admissible” objective functions (that includes Dasgupta’s one) that have the property that when the input admits a `natural’ hierarchical clustering, it has an optimal value. Equipped with a suitable objective function, we analyze the performance of practical algorithms, as well as develop better algorithms. For similarity-based hierarchical clustering, Dasgupta showed that the divisive sparsest-cut approach achieves an O(log^{3/2} n)-approximation. We give a refined analysis of the algorithm and show that it in fact achieves an O(\\sqrt{log n})-approx. (Charikar and Chatziafratis independently proved that it is a O(\\sqrt{log n})-approx.). This improves upon the LP-based O(logn)-approx. of Roy and Pokutta. For dissimilarity-based hierarchical clustering, we show that the classic average-linkage algorithm gives a factor 2 approx., and provide a simple and better algorithm that gives a factor 3/2 approx..Finally, we consider `beyond-worst-case’ scenario through a generalisation of the stochastic block model for hierarchical clustering. We show that Dasgupta’s cost function has desirable properties for these inputs and we provide a simple 1 + 
o(1)-approximation in this setting.
Joint work with Varun Kanade, Frederik Mallmann-Trenn, and Claire Mathieu.

Pour en savoir plus :
www.di.ens.fr/~vcohen/
##