- Lisa_seminaires - mailman.iro.umontreal.ca

Pre-ISMIR09 talks: Tomorrow(!), October 12th @ 12h30-13h30 in AA 3195
by Guillaume Desjardins 12 Oct '09

12 Oct '09

I hope everyone had a good thanksgiving, filled with family, lots of food (of your choice) and gravy. Unfortunately, I just returned to Montreal only to realize that the official announcement for tomorrow's talks was never sent ! (thanks for the heads up Isabelle). I hope some of you got wind of it by browsing Lisa's Google calendar. The talks will be given by Philippe Hamel and François Maillet, and are meant to capture material which will be presented at this year's International Society for Music Information Retrieval (ISMIR) conference, held in Kobe, Japan. I apologize for the late notice and hope many of you can attend ! ------------- Speaker: Philippe Hamel Title: Automatic identification of instrument classes in polyphonic and poly-instrument audio ------------- We present and compare several models for automatic identification of instrument classes in polyphonic and poly-instrument audio. The goal is to be able to identify which categories of instrument (Strings, Woodwind, Guitar, Piano, etc.) are present in a given audio example. We use a machine learning approach to solve this task. We constructed a system to generate a large database of musically relevant poly-instrument audio. Our database is generated from hundreds of instruments classified in 7 categories. Musical audio examples are generated by mixing multi-track MIDI files with thousands of instrument combinations. We compare three different classifiers: a Support Vector Machine (SVM), a Multilayer Perceptron (MLP) and a Deep Belief Network (DBN). We show that the DBN tends to outperform both the SVM and the MLP in most cases. ------------- Speaker: Francois Maillet Title: Steerable Playlist Generation by Learning Song Similarity from Radio Station Playlists ------------- This paper presents an approach to generating steerable playlists. We first demonstrate a method for learning song transition probabilities from audio features extracted from songs played in professional radio station playlists. We then show that by using this learnt similarity function as a prior, we are able to generate steerable playlists by choosing the next song to play not simply based on that prior, but on a tag cloud that the user is able to manipulate to express the high-level characteristics of the music he wishes to listen to. -- Guillaume Desjardins

1 0

Fwd: Next Math-neuro presentation
by Yoshua Bengio 30 Sep '09

30 Sep '09

Pour ceux que les modèles motivés par la neuroscience intéresse, je connais Chris Eliasmith et il est bon: Begin forwarded message: > From: David THURA <david.thura(a)umontreal.ca> > Date: September 30, 2009 9:52:25 AM EDT (CA) > > Subject: Next Math-neuro presentation > > Dear Math-Neuro group, > > This message to remind you that a math-neuro session is > exceptionally planned tomorrow, *October the 1st (Thursday)*, 4-6 > PM, room 4106 of Pavillon Paul-G. Desmarais, Université de Montréal. > > Please, check out the map: > http://maps.google.com/maps/ms?client=safari&oe=UTF-8&ie=UTF8&hl=en&t=k&lci… > > Chris Eliasmith (University of Waterloo) will be giving an exciting > talk entitled "*How to build a brain: From single cells to > cognition.*" (see the abstract below) > > Please, join us for the stimulating discussion... > > Best regards, > > David > > Abstract: > > A central goal of theoretical neuroscience is to develop methods > that allow us to connect high-level hypotheses about brain function > to specific, detailed neural mechanisms that can be compared to > neural data. In this talk I will present the Neural Engineering > Framework (NEF), which we have developed to address this goal. I > will demonstrate how the framework relates to detailed physiological > data, and how it aids the construction of complex, large-scale > spiking neural models. I will discuss its application to > perceptual, motor, and cognitive modelling using specific examples. > Finally, I will provide a brief introduction to Nengo, the GUI > software environment we have developed that implements > the principles of the NEF. > > -- > David THURA, PhD > Department of physiology > University of Montreal > 2960 chemin de la tour > Montreal (Québec) H3T 1J4 CANADA > > Phone : 514-343-6111 x3302 > FAX : 514-343-2111 > E-mail : david.thura(a)umontreal.ca > Web : http://sites.google.com/site/thuradavidphd/ > Math-Neuro@UdM : http://sites.google.com/site/mathneuroatudm/

1 0

Reminder: UdeM-McGill-MITACS machine learning seminar Fri, Sep 18th
by Guillaume Desjardins 17 Sep '09

17 Sep '09

Here's a friendly reminder for tomorrow's talk given by Dr. Massih-Reza Amini and entitled "Learning from Multiple Partially Observed Views" (see below for abstract). This talk will be held at the University of Montreal, Pavillon André-Aisenstadt (UdeM), room 3195 @ 11:00AM. See you there ! ---------- Forwarded message ---------- From: Guillaume Desjardins <guillaume.desjardins(a)gmail.com> Date: Fri, Sep 11, 2009 at 11:47 AM Subject: [Lisa_seminaires] RESCHEDULED: UdeM-McGill-MITACS machine learning seminar Fri, Sept To: lisa_seminaires(a)iro.umontreal.ca This is to advise you that the upcoming MITACS seminar has been rescheduled for Friday, September 18th from 11:00-12:00, due to some scheduling conflicts. I apologize for the short notice and hope many of you can make it. ---------- Forwarded message ---------- From: Guillaume Desjardins <guillaume.desjardins(a)gmail.com> Date: Tue, Sep 8, 2009 at 5:36 PM Subject: UdeM-McGill-MITACS machine learning seminar (NEW DATE & TIME) Mon, Sept 18th @11h00 To: lisa_seminaires(a)iro.umontreal.ca We start off the new academic year with a talk from Dr. Massih-Reza Amini entitled "Learning from Multiple Partially Observed Views". Speaker: Dr. Massih-Reza Amini Associate Professor of Pierre & Marie Curie University / Research Officer, NRC Institute for Information Technology Location: Pavillon André-Aisenstadt (UdeM), room 3195 Time: September 18th 2009 at 11h00 Title: Learning from Multiple Partially Observed Views Résumé: We address the problem of learning classifiers when observations have multiple views, some of which may not be observed for all examples. We assume the existence of view generating functions which may complete the missing views in an approximate way. This situation corresponds for example to learning text classifiers from multilingual collections where documents are not available in all languages. In that case, Machine Translation (MT) systems may be used to translate each document in the missing languages. We derive a generalization error bound for classifiers learned on examples with multiple artificially created views. Our result uncovers a trade-off between the size of the training set, the number of views, and the quality of the view generating functions. As a consequence, we identify situations where it is more interesting to use multiple views for learning instead of classical single view learning. An extension of this framework is a natural way to leverage unlabeled multi-view data in semi-supervised learning. Experimental results on a subset of the Reuters RCV1/RCV2 collections support our findings by showing that additional views obtained from MT may significantly improve the classification performance in the cases identified by our trade-off. _______________________________________________ Lisa_seminaires mailing list Lisa_seminaires(a)mercure.iro.umontreal.ca https://webmail.iro.umontreal.ca/mailman/listinfo/lisa_seminaires

1 0

RESCHEDULED: UdeM-McGill-MITACS machine learning seminar Fri, Sept
by Guillaume Desjardins 11 Sep '09

11 Sep '09

This is to advise you that the upcoming MITACS seminar has been rescheduled for Friday, September 18th from 11:00-12:00, due to some scheduling conflicts. I apologize for the short notice and hope many of you can make it. ---------- Forwarded message ---------- From: Guillaume Desjardins <guillaume.desjardins(a)gmail.com> Date: Tue, Sep 8, 2009 at 5:36 PM Subject: UdeM-McGill-MITACS machine learning seminar (NEW DATE & TIME) Mon, Sept 18th @11h00 To: lisa_seminaires(a)iro.umontreal.ca We start off the new academic year with a talk from Dr. Massih-Reza Amini entitled "Learning from Multiple Partially Observed Views". Speaker: Dr. Massih-Reza Amini Associate Professor of Pierre & Marie Curie University / Research Officer, NRC Institute for Information Technology Location: Pavillon André-Aisenstadt (UdeM), room 3195 Time: September 18th 2009 at 11h00 Title: Learning from Multiple Partially Observed Views Résumé: We address the problem of learning classifiers when observations have multiple views, some of which may not be observed for all examples. We assume the existence of view generating functions which may complete the missing views in an approximate way. This situation corresponds for example to learning text classifiers from multilingual collections where documents are not available in all languages. In that case, Machine Translation (MT) systems may be used to translate each document in the missing languages. We derive a generalization error bound for classifiers learned on examples with multiple artificially created views. Our result uncovers a trade-off between the size of the training set, the number of views, and the quality of the view generating functions. As a consequence, we identify situations where it is more interesting to use multiple views for learning instead of classical single view learning. An extension of this framework is a natural way to leverage unlabeled multi-view data in semi-supervised learning. Experimental results on a subset of the Reuters RCV1/RCV2 collections support our findings by showing that additional views obtained from MT may significantly improve the classification performance in the cases identified by our trade-off.

1 0

UdeM-McGill-MITACS machine learning seminar Mon, Sept 14th @15h00
by Guillaume Desjardins 08 Sep '09

08 Sep '09

We start off the new academic year with a talk from Dr. Massih-Reza Amini entitled "Learning from Multiple Partially Observed Views". Speaker: Dr. Massih-Reza Amini Associate Professor of Pierre & Marie Curie University / Research Officer, NRC Institute for Information Technology Location: Pavillon André-Aisenstadt (UdeM), room 3195 Time: September 14th 2009 at 15h00 Title: Learning from Multiple Partially Observed Views Résumé: We address the problem of learning classifiers when observations have multiple views, some of which may not be observed for all examples. We assume the existence of view generating functions which may complete the missing views in an approximate way. This situation corresponds for example to learning text classifiers from multilingual collections where documents are not available in all languages. In that case, Machine Translation (MT) systems may be used to translate each document in the missing languages. We derive a generalization error bound for classifiers learned on examples with multiple artificially created views. Our result uncovers a trade-off between the size of the training set, the number of views, and the quality of the view generating functions. As a consequence, we identify situations where it is more interesting to use multiple views for learning instead of classical single view learning. An extension of this framework is a natural way to leverage unlabeled multi-view data in semi-supervised learning. Experimental results on a subset of the Reuters RCV1/RCV2 collections support our findings by showing that additional views obtained from MT may significantly improve the classification performance in the cases identified by our trade-off.

1 0

Reminder: Invited Speaker: Chris Pal. Thursday, July 16th, 15:00 AA
by Guillaume Desjardins 15 Jul '09

15 Jul '09

Here's a friendly reminder for tomorrow's talk given by Christopher Pal (Ecole Polytechnique de Montreal). See below for abstract. The room has been confirmed and it will be held in AA 3195, as usual. On Fri, Jul 10, 2009 at 3:40 PM, Guillaume Desjardins<guillaume.desjardins(a)gmail.com> wrote: > Hi Everyone, > > Please mark your calendars for next week's talk by Christopher Pal, > from the Ecole Polytechnique de Montreal. > The talk will be given in both french (first half) and english. > English and french abstracts and bios can be found below. > > See you all there ! > > Titre: Probabilité, reconnaissance des formes et techniques > d'optimisation pour multimédias interactifs > Présentateur: Christopher Pal > Date: jeudi 16 juillet, 3:00pm > Place: pav. André-Aisenstadt, UdeM. Salle à confirmer (mais > probablement au AA 3195) > > Pour cette présentation, je démontrerai comment les modèles probabilistes et > les méthodes d'optimisation sont employés pour créer certaines technologies > clé pour : vision par ordinateur, photographie computationnelle, traitement > des langues naturelles et forage de données qui permettent la création de > nouvelles formes de médias interactifs. > > Je vais commencer par un bref examen des tâches facilitant la création des > représentations riches à haute résolution de scènes visuelles par la > combinaison des images, telles que les représentations panoramiques et avec > structure tridimensionnelle. J¹indiquerai comment ces représentations > peuvent ensuite être utilisées en tant qu'interfaces interactives pour > accéder aux informations. > > En examinant les étapes pour associer l'information sémantique des textes à > ces représentations visuelles riches, je montrerai comment des modèles > probabilistes basés sur les champs aléatoires sont employés pour classifier > les documents et pour extraire l'information sémantique. Je soulignerai > certains de nos travaux de recherche récents concernant la construction de > structures de modèles innovateurs et de méthodes innovatrices pour optimiser > les champs aléatoires. En particulier, je vais discuter des méthodes > hybrides génératives et discriminantes, leurs applications à l'apprentissage > semi-supervisé, les architectures basées sur les machines de Boltzmann et > l¹apprentissage avec les méthodes basées sur « contrastive divergence ». > > Biographie : > > Christopher Pal est professeur adjoint dans le département de génie > informatique et génie logiciel à l¹École Polytechnique de Montréal. Ses > intérêts de recherche incluent : la vision par ordinateur, la reconnaissance > de formes et l'apprentissage automatique avec des applications à > l'infographie, l'analyse du langage naturel et l'exploration des données. > > Il était auparavant professeur adjoint en informatique à l'Université de > Rochester. Il a également occupé des postes à l'Université du Massachusetts, > de l'Université de Toronto, « Interval Research » et « the Interactive > Visual Media Group of Microsoft Research ». Il a obtenu son doctorat de > l'Université de Waterloo au Canada. > > ===== > > Titre: Probability, Pattern Recognition and Optimization Techniques > for Interactive Media > Speaker: Christopher Pal > Date: Thursday, July 16th, 3:00pm > Place: André-Aisenstadt, UdeM. Room to be confirmed (probably in AA 3195). > > In this talk I show how probability models and optimization techniques can > be used to create core technologies for: computer vision, computational > photography, natural language processing and data mining which enable new > forms of interactive media. > > I¹ll begin by briefly examining some key tasks required to create rich, high > resolution representations of visual scenes by combining images to: create > high quality panoramic representations and infer three dimensional structure > in a scene. I'll motivate and show how such representations can then be used > as interactive interfaces to information. > > Examining the tasks involved with associating semantic information in text > with rich visual representations, I'll show how probability models based on > random fields can be used to classify documents and extract semantic > information. I'll highlight some of our recent work on constructing novel > model structures and novel approaches for optimizing random field based > models. In particular, I¹ll discuss hybrid generative and discriminative > techniques, applications to semi-supervised learning, Boltzmann machine > based architectures and learning with contrastive divergence based methods. > > Biography : > > Christopher Pal is an Assistant Professor of Computer and Software > Engineering at the École Polytechnique of Montreal. His research interests > include computer vision, pattern recognition and machine learning with > applications to computer graphics, natural language analysis and data > mining. > > He was previously an assistant professor of Computer Science with the > University of Rochester. He has also held positions with the University of > Massachusetts, the University of Toronto, Interval Research and Microsoft > Research¹s Interactive Visual Media Group. He earned his doctorate from the > University of Waterloo in Canada. > _______________________________________________ > Lisa_seminaires mailing list > Lisa_seminaires(a)mercure.iro.umontreal.ca > https://webmail.iro.umontreal.ca/mailman/listinfo/lisa_seminaires >

1 0

Invited Speaker: Chris Pal. Thursday, July 16th, 15:00, tentative
by Guillaume Desjardins 10 Jul '09

10 Jul '09

Hi Everyone, Please mark your calendars for next week's talk by Christopher Pal, from the Ecole Polytechnique de Montreal. The talk will be given in both french (first half) and english. English and french abstracts and bios can be found below. See you all there ! Titre: Probabilité, reconnaissance des formes et techniques d'optimisation pour multimédias interactifs Présentateur: Christopher Pal Date: jeudi 16 juillet, 3:00pm Place: pav. André-Aisenstadt, UdeM. Salle à confirmer (mais probablement au AA 3195) Pour cette présentation, je démontrerai comment les modèles probabilistes et les méthodes d'optimisation sont employés pour créer certaines technologies clé pour : vision par ordinateur, photographie computationnelle, traitement des langues naturelles et forage de données qui permettent la création de nouvelles formes de médias interactifs. Je vais commencer par un bref examen des tâches facilitant la création des représentations riches à haute résolution de scènes visuelles par la combinaison des images, telles que les représentations panoramiques et avec structure tridimensionnelle. J¹indiquerai comment ces représentations peuvent ensuite être utilisées en tant qu'interfaces interactives pour accéder aux informations. En examinant les étapes pour associer l'information sémantique des textes à ces représentations visuelles riches, je montrerai comment des modèles probabilistes basés sur les champs aléatoires sont employés pour classifier les documents et pour extraire l'information sémantique. Je soulignerai certains de nos travaux de recherche récents concernant la construction de structures de modèles innovateurs et de méthodes innovatrices pour optimiser les champs aléatoires. En particulier, je vais discuter des méthodes hybrides génératives et discriminantes, leurs applications à l'apprentissage semi-supervisé, les architectures basées sur les machines de Boltzmann et l¹apprentissage avec les méthodes basées sur « contrastive divergence ». Biographie : Christopher Pal est professeur adjoint dans le département de génie informatique et génie logiciel à l¹École Polytechnique de Montréal. Ses intérêts de recherche incluent : la vision par ordinateur, la reconnaissance de formes et l'apprentissage automatique avec des applications à l'infographie, l'analyse du langage naturel et l'exploration des données. Il était auparavant professeur adjoint en informatique à l'Université de Rochester. Il a également occupé des postes à l'Université du Massachusetts, de l'Université de Toronto, « Interval Research » et « the Interactive Visual Media Group of Microsoft Research ». Il a obtenu son doctorat de l'Université de Waterloo au Canada. ===== Titre: Probability, Pattern Recognition and Optimization Techniques for Interactive Media Speaker: Christopher Pal Date: Thursday, July 16th, 3:00pm Place: André-Aisenstadt, UdeM. Room to be confirmed (probably in AA 3195). In this talk I show how probability models and optimization techniques can be used to create core technologies for: computer vision, computational photography, natural language processing and data mining which enable new forms of interactive media. I¹ll begin by briefly examining some key tasks required to create rich, high resolution representations of visual scenes by combining images to: create high quality panoramic representations and infer three dimensional structure in a scene. I'll motivate and show how such representations can then be used as interactive interfaces to information. Examining the tasks involved with associating semantic information in text with rich visual representations, I'll show how probability models based on random fields can be used to classify documents and extract semantic information. I'll highlight some of our recent work on constructing novel model structures and novel approaches for optimizing random field based models. In particular, I¹ll discuss hybrid generative and discriminative techniques, applications to semi-supervised learning, Boltzmann machine based architectures and learning with contrastive divergence based methods. Biography : Christopher Pal is an Assistant Professor of Computer and Software Engineering at the École Polytechnique of Montreal. His research interests include computer vision, pattern recognition and machine learning with applications to computer graphics, natural language analysis and data mining. He was previously an assistant professor of Computer Science with the University of Rochester. He has also held positions with the University of Massachusetts, the University of Toronto, Interval Research and Microsoft Research¹s Interactive Visual Media Group. He earned his doctorate from the University of Waterloo in Canada.

1 0

Fwd: Séminaire demain, 3 juillet - École Poly, M-2110 @ 14h
by Dumitru Erhan 03 Jul '09

03 Jul '09

---------- Forwarded message ---------- From: Chris Pal <christopher.pal(a)polymtl.ca> Date: 2009/7/2 Subject: Séminaire demain, 3 juillet - École Poly, M-2110 @ 14h To: Dumitru Erhan <dumitru.erhan(a)umontreal.ca> Bonjour, Le 3 juillet à École Polytechnique, M-2110 @ 14h, Nicholas Morsillo étudiant au doctorat à l'université de Rochester qui travaille avec moi présentera son article dans IJCAI 2009 intitulé: "Semi-Supervised Learning of Visual Classifiers from Web Images and Text" Vous êtes cordialement invité à assister à la présentation et à échanger avec Nick suite à la présentation. La présentation elle-même sera d'environ 25 minutes. - Chris Pal Title of Talk: Semi-Supervised Learning of Visual Classifiers from Web Images and Text By: Nicholas Morsillo To appear in: Proceedings of the Twenty-First International Conference on Artificial Intelligence Abstract: The web holds tremendous potential as a source of training data for visual classification. However, web images must be correctly indexed and labeled before this potential can be realized. Accordingly, there has been considerable recent interest in collecting imagery from the web using image search engines to build databases for object and scene recognition research. While search engines can provide rough sets of image data, results are noisy and this leads to problems when training classifiers. In this paper we propose a semi-supervised model for automatically collecting clean example imagery from the web. Our approach includes both visual and textual web data in a unified framework. Minimal supervision is enabled by the selective use of generative and discriminative elements in a probabilistic model and a novel learning algorithm. We show through experiments that our model discovers good training images from the web with minimal manual work. Classifiers trained using our method significantly outperform analogous baseline approaches on the Caltech-256 dataset. ---- Christopher J. Pal, Ph. D. Professeur adjoint Département de génie informatique et génie logiciel Ecole Polytechnique de Montréal Téléphone : (514) 340-5121, poste 7174

1 0

Fwd: Séminaire RALI-OLST du 8 juillet 2009
by Yoshua Bengio 03 Jul '09

03 Jul '09

Begin forwarded message: > From: Webmestre RALI <critter(a)iro.umontreal.ca> > Date: July 2, 2009 8:30:28 PM EDT (CA) > To: rali-olst(a)iro.umontreal.ca > Subject: Séminaire RALI-OLST du 8 juillet 2009 > > > Annonce de séminaire RALI-OLST > > Prédiction des mots cibles en traduction statistique à l'aide d'un > réseau de neurones > Alexandre Patry > RALI, DIRO > > Le 8 juillet 2009 à 11h30 > Salle 3195, Pavillon André-Aisenstadt > > ---- > Une famille de systèmes de traduction statistique de pointes > traduisent > une phrase source donnée segment par segment. Cette approche capture > très bien les dépendances à l'intérieur d'un segment mais néglige les > dépendances entre un segment et le reste de la phrase. Lors de ce > séminaire, je présenterai ce que je crois être une piste de solution à > ce problème: un modèle évaluant la probabilité de chaque mot cible > étant donnée tous les mots de la phrase source considérées de manière > jointe. > ---- > > > Consulter l'annonce du séminaire sur le Web : > http://rali.iro.umontreal.ca/sem/display_sem.php?semid=3344

1 0

Reminder: UdeM-McGill-MITACS machine learning seminar Mon, June 22nd,
by Dumitru Erhan 22 Jui '09

22 Jui '09

Here's a reminder of today's talk and a new abstract by Andrew Ng (the topic has changed!): Reinforcement learning and apprenticeship learning for robotic control by Andrew Y. Ng Location: Pavillon André-Aisenstadt (UdeM), room 3195 Time: June 22nd, 13h30 Drawing from examples in autonomous helicopter flight, legged robot locomotion, and autonomous driving, in this talk I’ll describe a class of reinforcement learning and apprenticeship learning algorithms—methods which learn by watching an expert demonstration of a task—for robotic control. Specifically, we’ll see that for many robots it Is challenging to develop a controller because (i) It is hard to write down, in closed form, a formal specification of the control task (for example, what is the cost function for "driving well"?), and (ii) It is difficult to learn good models of the robot's dynamics. I’ll present formal results showing how apprenticeship learning methods, when given access to a human demonstration of a task, can be used to efficiently address these problem. Further, I’ll present results on the application of these ideas to controlling several different robots. Bio: Andrew Ng is an Assistant Professor of Computer Science at Stanford University. His research interests include machine learning, reinforcement learning/control, and broad-competence AI. His group has won best paper/best student paper awards at ACL, CEAS, 3DRR and ICML. He is also a recipient of the Alfred P. Sloan Fellowship, and the IJCAI 2009 Computers and Thought award.

1 0