Lisa_seminaires

Download

lisa_seminaires@iro.umontreal.ca

January 2010

2 participants
2 discussions

UdeM-McGill-MITACS machine learning seminar Tue Feb. 2@14h30, Z-209
by Dumitru Erhan 01 Feb '10

01 Feb '10

Please note the *different location: room Z-209* in Pavillon Claire-McNicoll, located right next to Pavillon André-Aisenstadt. Next week's seminar (see http://www.iro.umontreal.ca/article.php3?id_article=107&lang=en) *Apprentissage de fonctions de classification et d’ordonnancement avec des données partiellement étiquetées* by Massih-Reza Amini Research Officer, Interactive Language Technologies NRC Institute for Information Technology Gatineau, Quebec Location: *Pavillon Claire-McNicoll (UdeM), room Z-209* Time: Tuesday, February 2, 14:30. Avec le développement des technologies d’information on assiste depuis quelques années à une nouvelle impulsion pour la conception de nouveaux cadres d’apprentissage automatique. C’est le cas par exemple du paradigme semi-supervisé qui a vu le jour vers la fin des années 90 dans la communauté apprentissage. Les premiers travaux dans ce cadre ont été motivés par le développement du web qui a entraîné une production massive de données textuelles très hétérogènes. Ces masses de données sont généralement livrées sous forme brute, sans étiquetage a priori et pour les exploiter on était alors réduit à utiliser des techniques non-supervisées. Ces approches bien que totalement génériques ne permettent cependant qu’une analyse limitée des informations de contenu et ne répondent pas ainsi aux demandes de nombreuses tâches de Recherche d’Information (RI). L’idée pragmatique développée pour l’apprentissage semi-supervisé était née de la question; "comment réduire l’effort d’étiquetage et utiliser simultanément une petite quantité de données étiquetées avec la masse de données non-étiquetées disponible pour apprendre?" Un autre exemple de l’émergence de nouveaux cadres d’apprentissage concerne le développement de méthodes automatiques pour la recherche et l’ordonnancement d’entités d’information sur des corpus de grandes tailles. Récemment beaucoup de travaux se sont intéressés à la formulation des différentes formes de la tâche d’ordonnancement. Ces travaux ont proposé des algorithmes et développé des cadres théoriques pour la prédiction d’ordres totaux ou partiels sur les exemples. La Recherche d’Information est une fois encore le domaine par excellence où les modèles d’apprentissage de fonctions d’ordonnancement jouent un rôle prépondérant. Dans notre étude nous nous sommes intéressés à deux cadres d’ordonnancement d’instances et d’alternatives. Dans le premier cas il s’agit d’ordonner les exemples (où instances) d’une collection donnée de façon à ce que les exemples jugés pertinents soient ordonnés au—dessus des exemples non—pertinents et dans le second cas nous cherchons à ordonner les alternatives d’une collection donnée par rapport à chaque exemple d’entrée. Je présenterai l’ensemble de mes travaux suivant ces deux axes en commençant par l’apprentissage semi-supervisé de fonctions de classification qui est plutôt un travail algorithmique. Je décrirai ensuite le cadre théorique que nous avons proposé pour l’apprentissage de fonctions d’ordonnancements avec comme application l’apprentissage actif de fonctions d’ordonnancement d’alternatives. -- http://dumitru.ca, +1-514-432-8435

1 1

Fwd: Canadian Conference on AI 2010 - SECOND CALL for papers
by Guillaume Desjardins 12 Jan '10

12 Jan '10

---------- Forwarded message ---------- From: Marina Sokolova <msokolova(a)ehealthinformation.ca> Date: Tue, Jan 12, 2010 at 11:56 AM Subject: SECOND CALL for papers To: Guillaume Desjardins <guillaume.desjardins(a)gmail.com> Canadian Conference on AI 2010 Graduate Students Symposium May 30th, University of Ottawa, Ontario, Canada Call For Paper - Deadline 30 January 2010 http://glass.cs.unb.ca/~ebrahim/conf/cai-gs/ AI 2010, the twenty-third Canadian Conference on Artificial Intelligence, invites graduate students to submit four-page extended abstracts of their thesis for possible inclusion in the AI 2010 Graduate Student Symposium and the Canadian AI proceedings published by Springer. Symposium Objectives The Symposium provides an opportunity for Master's and PhD students to discuss and explore their research interests and career objectives with their peers and with a panel of established researchers in Artificial Intelligence, helping to develop a supportive community of scholars and a spirit of collaborative research. The symposium will be a one-day pre-conference event, where students of accepted abstracts will be invited to give a presentation on their thesis work before a group of peers as well as a small team of expert AI researchers who would offer a critique of each presentation and provide support, advice, and mentoring. Each student may also be invited to participate in the poster session during the main conference. In addition, a small selection of the best student submissions will be invited to give a short talk during the main conference. PhD and Master's students are invited to submit original work in all areas of Artificial Intelligence. Submissions should have emphasis on work in progress, with directions for future research clearly indicated wherever appropriate. More weight will be placed on the significance of the work, the proposed ideas or solutions, and the overall presentation than on submissions which present work with complete results. The Application Package Applicants to the symposium need to submit the following materials. Please combine all materials into one PDF document: * A four-page thesis summary that outlines the problem being addressed, the proposed plan for research, and a description of the progress to date. The most successful applications directly address all three of these components. Please be sure to distinguish between work that has already been accomplished and work that remains to be done. Be sure to include a title for your work. All submissions must be written in English. Abstracts may be up to 4 pages in length and must be formatted according to Springer's LNCS style. Please follow the instructions for authors at Springer's site for authors. The use of the LaTeX2e style file available at the web site is strongly encouraged. * Include a CV (at most two pages) that describes your background and relevant experience (research, education, employment). Your CV needs to include your anticipated graduation date, or it may be rejected without review. Please include any additional items that may indicate your potential contribution to, and benefit from, the symposium. * A letter of recommendation from your thesis advisor. It must include an assessment of the current status of your thesis research, and an expected date for thesis submission. In addition, your advisor should indicate what he or she hopes you would gain from participation in the symposium. * Participant's Expectations. A short (one page or less) statement of what you expect to gain from presenting and participating in the symposium, as well as what you think you can contribute to the symposium. For best consideration, your statement must address both of these expectations. The Reviewing Process Each abstract will be reviewed by a team of program committee members. Presenting students will be selected based on clarity of the submission, stage of research, advisor's letter and evidence of promise such as published papers or technical reports. (Partial) Financial assistance for travel and accommodations may be available to the students presenting at the Symposium. All students are encouraged to attend and participate in the Symposium, whether or not they apply to present their work. All the selected student presenters are expected to actively participate in the full Symposium, as we envision participants gaining as much by interacting with their peers as by having their presentations critiqued by the faculty panel. In addition, a senior AI researcher and faculty member will be invited to give a talk on a topic of interest to graduate students. Important Dates Package submission due January 30th, 2010 Notification of acceptance March 3rd, 2010 Final paper due March 8th, 2010 Program Co-chairs: Ebrahim Bagheri, National Research Council Canada and Athabasca University Marina Sokolova, CHEO Research Institute, University of Ottawa

1 0