Annonce November 2017

annonce@lrde.epita.fr

2 participants
3 discussions

[Seminaire-LRDE] Séminaire du LRDE le mercredi 13 décembre 2017: Camille Couprie, Facebook AI research

by Guillaume Tochon

Chers collègues, La prochaine session du séminaire Performance et Généricité du LRDE (Laboratoire de Recherche et Développement de l'EPITA) aura lieu le Mercredi 13 décembre 2017 (11h--12h), Amphi 4 de l'EPITA. Vous trouverez sur le site du séminaire [1] les prochaines séances, les résumés, captations vidéos et planches des exposés précédents [2], le détail de cette séance [3] ainsi que le plan d'accès [4]. [1] http://seminaire.lrde.epita.fr [2] http://seminaire.lrde.epita.fr/Archives [3] http://seminaire.lrde.epita.fr/2017-12-13 [4] http://www.lrde.epita.fr/wiki/Contact Au programme du Mercredi 13 décembre 2017 : * 11h--12h: Vers l'apprentissage d'un sens commun visuel -- Camille Couprie, Facebook AI research https://perso.esiee.fr/~coupriec/ Les réseaux de neurones convolutifs connaissent depuis quelques années un franc succès dans de nombreuses applications de reconnaissance visuelle. Nous reviendrons sur les premiers travaux en la matière en segmentation sémantique (étiquetage de chaque pixel des images par une catégorie sémantique). Nous explorerons ensuite une piste d'amélioration visant à réduire la quantité de données labelisées utilisée, à base d'entraînement de réseaux adversaires. Dans un second temps, nous nous intéresserons au problème de la prédiction d'images suivantes dans les vidéos: s'il nous parait simple d'anticiper ce qu'il va se passer dans un futur très proche, c'est un problème difficile à modéliser mathématiquement étant données les multiples sources d'incertitude. Nous présenterons nos travaux de prédiction dans le domaine des images naturelles, puis dans l'espace plus haut niveau des segmentations sémantiques, nous permettant de prédire plus loin dans le futur. -- Camille Couprie est chercheuse à Facebook Artificial Intelligence Research. Elle a obtenu son doctorat en informatique de l'Université Paris Est en 2011, sous la direction de Hugues Talbot, Laurent Najman et Leo Grady, avec une recherche spécialisée dans la formulation et l'optimisation de problèmes de vision par ordinateur dans les graphes. En 2012, elle a travaillé comme postdoc a l'institut Courant de New York University avec Yann LeCun. Après un poste IFP new energies, organisme de recherche français actif dans les domaines de l'énergie, des transports et de l'environnement, elle a rejoint Facebook en 2015. L'entrée du séminaire est libre. Merci de bien vouloir diffuser cette information le plus largement possible. N'hésitez pas à nous faire parvenir vos suggestions d'orateurs. -- Guillaume TOCHON Maître de conférences // Assistant Professor LRDE, EPITA 18, rue Pasteur 94270 Le Kremlin-Bicêtre _______________________________________________ Seminaire mailing list Seminaire(a)lrde.epita.fr https://lists.lrde.epita.fr/listinfo/seminaire

7 years

[Seminaire-LRDE] Séminaire du LRDE le mercredi 29 novembre 2017: Barbara Jobstmann, Cadence Design Systems

by Guillaume Tochon

Chers collègues, La prochaine session du séminaire Performance et Généricité du LRDE (Laboratoire de Recherche et Développement de l'EPITA) aura lieu le Mercredi 29 novembre 2017 (10h--11h), Amphi 4 de l'EPITA. Vous trouverez sur le site du séminaire [1] les prochaines séances, les résumés, captations vidéos et planches des exposés précédents [2], le détail de cette séance [3] ainsi que le plan d'accès [4]. [1] http://seminaire.lrde.epita.fr [2] http://seminaire.lrde.epita.fr/Archives [3] http://seminaire.lrde.epita.fr/2017-11-29 [4] http://www.lrde.epita.fr/wiki/Contact Au programme du Mercredi 29 novembre 2017 : * 10h--11h: Industrial Formal Verification – Cadence’s JasperGold Formal Verification Platform -- Barbara Jobstmann, Cadence Design Systems Formal verification (aka Symbolic Model Checking) is becoming a mainstream technology in system on a chip (SoC)/intellectual property design and verification methodologies. In the past, the usage of formal verification was limited to a small range of applications; it was mainly used to verify complex protocols or intrinsic logic functionality by formal verification experts. In recent years, we saw a rapid adoption of formal verification technology and many new application areas, such as checking of configuration and status register accesses, SoC connectivity verification, low power design verification, security applications, and many more. In this talk, we give an overview of the JasperGold Formal Verification Platform. The platform provides a wide range of formal apps, which ease adoption of formal verification by offering property generation and other targeted capabilities for specific design and verification tasks. In addition, JasperGold offers a unique interactive debug environment (called Visualize) that allows the user to easily analyze the verification results. We present JasperGold from a user’s point of view, showcase selected apps, and discuss features that were essential for their wide adoption. -- Barbara Jobstmann is a field application engineer for Cadence Design Systems and a lecturer at the École Polytechnique Fédérale de Lausanne (EPFL). She joined Cadence in 2014 through the acquisition of Jasper Design Automation, where she worked since 2012 as an application engineer. In the past, she was also a CNRS researcher (chargé de recherche) in Verimag, an academic research laboratory belonging to the CNRS and the Communauté Université Grenoble Alpes in France. Her research focused on constructing correct and reliable computer systems using formal verification and synthesis techniques. She received a Ph.D. degree in Computer Science from the University of Technology in Graz, Austria in 2007. L'entrée du séminaire est libre. Merci de bien vouloir diffuser cette information le plus largement possible. N'hésitez pas à nous faire parvenir vos suggestions d'orateurs. -- Guillaume TOCHON Maître de conférences // Assistant Professor LRDE, EPITA 18, rue Pasteur 94270 Le Kremlin-Bicêtre _______________________________________________ Seminaire mailing list Seminaire(a)lrde.epita.fr https://lists.lrde.epita.fr/listinfo/seminaire

7 years

Vcsn 2.6 is released!

by Akim Demaille

The Vcsn team is happy to announce the long overdue release of Vcsn 2.6. Most of our work was devoted to providing a better, smoother, user experience. This includes improvements in the build system, better performances, extended consistency, and more legible error messages. Multitape expressions also received a lot of attention. For more information see the detailed news below. We warmly thank our users who made valuable feedback (read "bug reports") including Victor Miller, Dominique Soudière and Harald Schilly. Dominique and Harald helped integrating Vcsn into CoCalc (formerly SageMathCloud). People who contributed to this release: - Akim Demaille - Clément Démoulins - Clément Gillard - Sarasvati Moutoucomarapoulé Our next stop should be Vcsn 3.0, which will include extensive changes from Sarasvati. Release page: https://vcsn.lrde.epita.fr/Vcsn2.6 Tarball: https://www.lrde.epita.fr/dload/vcsn/2.6/vcsn-2.6.tar.bz2 Home page: https://vcsn.lrde.epita.fr The various packages (Docker, Debian, MacPorts, etc.) will appear soon. ## New features ### Promotion from single to multitape It is now possible to mix single-tape expressions with multitape expressions. This is especially handy when using labels classes (`[abc]`). For instance: In [2]: zmin2 = vcsn.context('lat<lan, lan>, zmin') In [3]: zmin2.expression(r'([ab] + ⟨1⟩(\e|[ab] + [ab]|\e))*') Out[3]: (a|a+b|b+<1>(\e|(a+b)+(a+b)|\e))* is a easy means to specify an expression generating an automaton that computes the edit-distance (between words or languages). Or, with a wide alphabet: In [4]: zmin = vcsn.context('lan(a-z), zmin') In [5]: zmin2 = zmin | zmin In [6]: zmin2 Out[6]: {abcdefghijklmnopqrstuvwxyz}? x {abcdefghijklmnopqrstuvwxyz}? -> Zmin In [7]: zmin2.expression(r'([^] + ⟨1⟩(\e|[^] + [^]|\e))*') In the future the display of expressions will also exploit this approach. ### vcsn compile now links with precompiled contexts If you write a program using dyn::, then it will benefit from all the precompiled algorithms. ### vcsn compile now supports --debug Use this to avoid the removal of intermediate object files. On some platforms, such as macOS, the object file contains the debug symbols, so their removal makes the use of a debug harder. ### vcsn compile now uses -rpath Programs/libraries generated by `vcsn compile` needed to be told where the Vcsn libraries were installed. In practice, it meant that `vcsn run` was needed to execute this program, and in some situations it was not even sufficient. Now, `vcsn compile my-prog.cc` generates `my-prog` which can be run directly. ### random_expression supports the tuple operator It is now possible to generate multitape expressions such as `(a|x*)*`. Before, random_expression was limited to expressions on multitape labels such as `(a|x)*`. In [1]: import vcsn In [2]: c = vcsn.context('lan(abc), q') In [3]: c2 = c | c In [4]: for i in range(10): ...: e = c2.random_expression('|, +, ., *=.2, w.=.2, w="min=-2, max=2"', length=10) ...: print('{:u}'.format(e)) ...: a|a+(ε|b)*+⟨1/2⟩ε|ε (ε+c)a|(⟨2⟩ε+c) (ε|b+(c|a)*)* ε|a+a|b c|b+⟨2⟩ε|(ε+b) ε|a+((a|ε)(ε|b)(ε|c))*(b|ε) (a|b+c|ε)(a|b) ε*|ε ε+a|a+⟨2⟩(b|ε) (ε+ε*)|ε Generating expressions with compose operators is also supported. In [5]: for i in range(10): ...: e = c2.random_expression('|=.5, +, ., @=2', length=10, identities='none') ...: print('{:u}'.format(e)) ...: ((b|ε)ε)(ε(ε|a)) ε(ε|c)+((ε|a)(a|b)+ε|c) ((ε|a)(ε|b))((b|a)(c|c)@b|b) (b|b@ε+a|b)@(c|ε@b|ε) (ε+ε|a)@(ε|c@ε|a+b|ε) (ε|a+ε|c)((ε|b)(c|ε)) (a|ε)((ε+c)|(a+ε)) c|ε@(a|ε)(ε|b@b|ε) (ε+ε|b)@ε|b+ε|a b(ε²)|((ε+ε)+a) ### New algorithms A few algorithms were added: - context.compose Composing `{abc} x {xyz}` with `{xyz} x {ABC}` gives, of course, `{abc} x {ABC}`. - expansion.normalize, expansion.determinize These are rather low-level features. The latter is used internally when derived-term is asked for a deterministic automaton. - expansion.expression The "projection" of an expansion as an expression. - label.compose For instance `a|b|x @ x|c` -> `a|b|c`. - polynomial.compose Additive extension of monomial composition. For instance composing `a|e|x + a|e|y + a|e|\e` with `x|E|A + y|E|A + \e|E|A` gives `<3>a|e|E|A`. - polynomial.shuffle and polynomial.infiltrate The missing siblings of polynomial.conjunction. ### Doxygen is no longer needed By default, `make install` generated and installed the C++ API documentation using Doxygen. It is now disabled by default, pass `--enable-doxygen` to `configure` to restore it. ## Bug Fixes ### Severe performance regression when reading daut Version 2.5 introduced a large penalty when reading daut files, especially large ones. ### Tupling of automata could be wrong on weights As a consequence, `expression.inductive` was also producing incorrect automata from multitape expressions. ### Polynomials featuring the zero-label Under some circumstances it was possible to have polynomials featuring monomials whose label is the zero of the labelset (e.g., `\z` with expressions as labels). This is fixed. ### Portability issues Newest compilers are now properly supported. ## Changes ### Divisions The division of expressions and automata now compute the product of the weights, not their quotient. ### Handling of the compose operator The support of the compose operators in expansions was completely rewritten. As a result, the derived-term automata are often much smaller, since many useless states (non coaccessible) are no longer generated. For instance the derived-term automaton of `(\e|a)* @ (aa|\e)*` has exactly two states. It used to have three additional useless states. ### Better diagnostics Many error messages have been improved. The Daut automaton format now treats `->` as a keyword, so `0->1 a` is now properly read instead of producing a weird error message because Vcsn thought your state was named `0->1`.

7 years, 1 month

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

2005

2004

Annonce November 2017