- Lisa_seminaires - mailman.iro.umontreal.ca

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

UdeM-McGill-MITACS machine learning seminar Tues, Nov. 24th
by Guillaume Desjardins 22 Nov '09

22 Nov '09

There is another MITACS talk this week, given by Karol Gregor from NYU. Come one, come all ! Speaker: Karol Gregor, NYU Title: Emergence of Complex-Like Cells in a Temporal Product Network with Local Receptive Fields Location: Pavillon André-Aisenstadt (Université de Montréal), room 3195 Time: November 24th 2009, 12h30-13h30 Abstract: We introduce a new neural architecture and an unsupervised algorithm for learning invariant representations from temporal sequence of images. The system uses two groups of complex cells whose outputs are combined multiplicatively: one that represents the content of the image, constrained to be constant over several consecutive frames, and one that represents the precise location of features, which is allowed to vary over time. The architecture uses an encoder to extract features, and a decoder to reconstruct the input from the features. The method was applied to patches extracted from consecutive movie frames and produces orientation and frequency selective units analogous to the complex cells in V1. An extension of the method was used to train a network composed of units with local receptive field spread over a large image of arbitrary size. Pooling over local neighborhoods was also used which produces orientation-selective cells that are organized in pinwheel patterns similar to those observed in the mammalian visual cortex. By adding logistic regression classifier applied to complex cell layer, the complete system becomes a very fast, biologically inspired object recognition system - it contains locally-connected simple and complex like cells, it can be applied to full sized images, exhibits pinwheel patterns and has a feed-forward encoder for efficient feature computation. -- Guillaume Desjardins

1 0

UdeM-McGill-MITACS machine learning seminar Fri Nov. 20@14h30,
by Dumitru Erhan 19 Nov '09

19 Nov '09

Next week's seminar (see http://www.iro.umontreal.ca/article.php3?id_article=107&lang=en) Unlocking Brain-Inspired Computer Vision: a Multi-Disciplinary, High-Throughput Approach by Nicolas Pinto Department of Brain and Cognitive Sciences Massachusetts Institute of Technology Location: Pavillon André-Aisenstadt (UdeM), room 3195 Time: Friday, November 20, 14h30 The construction of artificial vision systems and the study of biological vision are naturally intertwined as they represent simultaneous efforts to forward and reverse engineer systems with similar goals. While exploration of the neuronal substrates of visual processing provides clues and inspiration for artificial systems, artificial systems can in turn serve as important generators of new ideas and working hypotheses. However, while systems neuroscience has so far provided inspiration for some of the "broad-stroke" properties of the visual system (e.g. hierarchical organization, synaptic integration of inputs and threshold, normalization, plasticity, etc), much is still unknown. Even for those qualitative properties that most biological-inspired models hold in common, experimental data currently provide little constraint on their key parameters. Consequently, it is difficult to truly evaluate a set of computational ideas, since the performance of any one model depends strongly on its particular instantiation - e.g. the size of the pooling kernels, the number of units per layer, exponents in normalization operations, etc. Since the number of such parameters (explicit or implicit) is very large, and the typical computational cost of evaluating one particular model is high, the space of possible model instantiations usually goes largely unexplored. Compounding the problem, even if a set of computational ideas are on the right track, the instantiated "scale" of those ideas is typically small (e.g. in terms of dimensionality and amount of learning experience provided). Thus, when a model fails to approach the abilities of the visual system, we are left uncertain whether this failure is because we are missing a fundamental idea, or because the correct "parts" have not been tuned correctly, assembled at sufficient scale, or provided with sufficient natural experience. To pave a possible way forward, we have begun developing a high-throughput approach to expansively explore a large range of biologically-inspired models - including models of larger, more realistic scale - leveraging recent advances in commodity stream processing hardware (high-end GPUs and Playstation 3’s Cell processors) and scientific cloud computing (e.g. Amazon EC2). In analogy to high-throughput screening approaches in molecular biology and genetics, we generated and trained thousands of potential network architectures and parameter instantiations, and "screened" the visual representations produced by these models using an object recognition task. From these candidate models, the most promising were selected for further analysis. We have shown that this approach can yield significant, reproducible gains in performance across an array of basic object recognition tasks, consistently outperforming a variety of state-of-the-art purpose-built vision systems from the literature, and that it can offer insight into which computational ideas are most important for achieving this performance. As the scale of available computational power continues to expand, we believe that this approach holds great potential both for accelerating progress in artificial vision, and for generating new, experimentally-testable hypotheses for the study of biological vision.

1 1

Fwd: Colloque DIRO aujourd'hui à 15:30
by Dumitru Erhan 19 Nov '09

19 Nov '09

Reminder for today's LISA & DIRO seminar: ---------- Forwarded message ---------- From: Miklós Csürös <csuros(a)iro.umontreal.ca> Date: 2009/11/19 Subject: Colloque DIRO aujourd'hui à 15:30 To: seminaires(a)iro.umontreal.ca COLLOQUE DU DIRO --- http://www.iro.umontreal.ca/~csuros/ColloquesDIRO/<http://www.iro.umontreal.ca/%7Ecsuros/ColloquesDIRO/> Conférencier: David Stork (Ricoh Innovations et Stanford U.) Titre: Computer vision in the study of art: New rigorous approaches to the study of paintings and drawings Date: jeudi, le 19 novembre 2009, 15:30 Lieu: Salle 6214, Pavillon André-Aisenstadt Résumé: New rigorous computer algorithms have been used to shed light on a number of recent controversies in the study of art. For example, illumination estimation and shape-from-shading methods developed for robot vision and digital photograph forensics can reveal the accuracy and the working methods of masters such as Jan van Eyck and Caravaggio. Computer box-counting methods for estimating fractal dimension have been used in authentication studies of paintings attributed to Jackson Pollock. Computer wavelet analysis has been used for attribution of the contributors in Perugino's Holy Family and works of Vincent van Gogh. Computer methods can dewarp the images depicted in convex mirrors depicted in famous paintings such as Jan van Eyck's Arnolfini portrait to reveal new views into artists' studios and shed light on their working methods. New principled, rigorous methods for estimating perspective transformations outperform traditional and ad hoc methods and yield new insights into the working methods of Renaissance masters. Sophisticated computer graphics recreations of tableaus allow us to explore what if scenarios, and reveal the lighting and working methods of masters such as Caravaggio. ------------------ Miklós Csürös -- http://dumitru.ca, +1-514-432-8435

1 0

Fwd: LISA and DIRO talk: David Stork
by Aaron Courville 17 Nov '09

17 Nov '09

---------- Forwarded message ---------- From: Aaron Courville <courvila(a)iro.umontreal.ca> Date: Tue, Nov 17, 2009 at 12:51 PM Subject: LISA and DIRO talk: David Stork To: Lisa Labo <lisa_labo(a)iro.umontreal.ca> The "Colloques du DIRO" this week is our visitor, David Stork. He will be visiting all day Thursday, please email me if you are interested in meeting with him. Cheers, Aaron Speaker: David G. Stork (Ricoh Innovations and Stanford University_ Time/Location: 15h30 Thursday Nov 19th / AA6214 Title: Computer vision in the study of art: New rigorous approaches to the study of paintings and drawings Abstract: New rigorous computer algorithms have been used to shed light on a number of recent controversies in the study of art. For example, illumination estimation and shape-from-shading methods developed for robot vision and digital photograph forensics can reveal the accuracy and the working methods of masters such as Jan van Eyck and Caravaggio. Computer box-counting methods for estimating fractal dimension have been used in authentication studies of paintings attributed to Jackson Pollock. Computer wavelet analysis has been used for attribution of the contributors in Perugino's Holy Family and works of Vincent van Gogh. Computer methods can dewarp the images depicted in convex mirrors depicted in famous paintings such as Jan van Eyck's Arnolfini portrait to reveal new views into artists' studios and shed light on their working methods. New principled, rigorous methods for estimating perspective transformations outperform traditional and ad hoc methods and yield new insights into the working methods of Renaissance masters. Sophisticated computer graphics recreations of tableaus allow us to explore "what if" scenarios, and reveal the lighting and working methods of masters such as Caravaggio. Biographical sketches: (taken from http://www.diatrope.com/stork/FAQs.html) (Short) Dr. David G. Stork, Chief Scientist of Ricoh Innovations, is a graduate in physics of the Massachusetts Institute of Technology and the University of Maryland at College Park, studied art history at Wellesley College and was Artist-in-Residence through the New York State Council of the Arts. He is a Fellow of the International Association for Pattern Recognition and has published six books/proceedings volumes and has one forthcoming, including Seeing the Light: Optics in nature, photography, color, vision and holography (Wiley), Computer image analysis in the study of art (SPIE), Pattern Classification (2nd ed., Wiley), and HAL's Legacy: 2001's computer as dream and reality (MIT). (Long) Dr. David G. Stork is Chief Scientist of Ricoh Innovations and has held academic appointments and taught in five departments at Stanford University frequently over the last 20 years. The breadth of his interests and contriburtions is revealed through the academic departments and programs in which he has held faculty positions in leading liberal arts colleges and research universities: Physics, Mathematics, Electrical Engineering, Statistics, Computer Science, Neuroscience, Psychology, and Art and Art History. He is a Fellow of the International Association for Pattern Recognition, "For contributions to pattern recognition education, machine learning, speech recognition, and the application of computer vision to the study of art" and Chair of its Technical Committee on Computer Vision in Cultural Heritage Applications. He has published six books/proceedings volumes and has another in production, including Seeing the Light: Optics in nature, photography, color, vision and holography (Wiley), the leading textbook on optics in the arts, Computer image analysis in the study of art (SPIE), the first volume in this discipline, Computer vision and image analysis in the study of art (forthcoming), Pattern Classification (2nd ed., Wiley), the world's all-time best-selling textbook in the field, translated into three languages and used in courses in over 250 universites worldwide, and HAL's Legacy: 2001's computer as dream and reality (MIT), the source of his PBS television documentary 2001: HAL's Legacy. A graduate in physics of the Massachusetts Institute of Technology and the University of Maryland at College Park, he also studied art history at Wellesley College and was Artist-in-Residence through the New York State Council of the Arts. He has performed research and taught courses such as "Light, color and visual phenomena," "The physics of aesthetics and perception," "Optics, perspective and Renaissance painting," and "Computer vision and image analysis in the study of art" over the last quarter century variously at leading liberal arts and research universities such as Wellesley College, Swarthmore College, Clark University, Boston University and Stanford University. He holds 38 US patents and has published numerous technical papers on human and machine learning and perception of patterns, physiological optics, image understanding, concurrency theory, theoretical mechanics, optics, image processing. He has served on the editorial boards of five international journals and has delivered over 58 plenary, invited or distinguished lectures at universities and conferences (atop over 230 traditional invited colloquia and seminars). He was one of four scientists invited to comment on Mr. Hockney's theory at the December 2001 "Art and Optics" Symposium at the New York Institute for the Humanities and one of two scientists invited to present a lecture in the symposium exploring the possible use of optics by early Renaissance painters at the Optical Society of America's Annual Meeting in Rochester, NY, October 2004. -- Aaron C. Courville Département d’Informatique et de recherche opérationnelle Université de Montréal email:Aaron.Courville@gmail.com -- Aaron C. Courville Département d’Informatique et de recherche opérationnelle Université de Montréal email:Aaron.Courville@gmail.com

1 0

Fwd: LISA and DIRO talk: David Stork
by Aaron Courville 17 Nov '09

17 Nov '09

---------- Forwarded message ---------- From: Aaron Courville <courvila(a)iro.umontreal.ca> Date: Tue, Nov 17, 2009 at 12:51 PM Subject: LISA and DIRO talk: David Stork To: Lisa Labo <lisa_labo(a)iro.umontreal.ca> The "Colloques du DIRO" this week is our visitor, David Stork. He will be visiting all day Thursday, please email me if you are interested in meeting with him. Cheers, Aaron Speaker: David G. Stork (Ricoh Innovations and Stanford University_ Time/Location: 15h30 Thursday Nov 19th / AA6214 Title: Computer vision in the study of art: New rigorous approaches to the study of paintings and drawings Abstract: New rigorous computer algorithms have been used to shed light on a number of recent controversies in the study of art. For example, illumination estimation and shape-from-shading methods developed for robot vision and digital photograph forensics can reveal the accuracy and the working methods of masters such as Jan van Eyck and Caravaggio. Computer box-counting methods for estimating fractal dimension have been used in authentication studies of paintings attributed to Jackson Pollock. Computer wavelet analysis has been used for attribution of the contributors in Perugino's Holy Family and works of Vincent van Gogh. Computer methods can dewarp the images depicted in convex mirrors depicted in famous paintings such as Jan van Eyck's Arnolfini portrait to reveal new views into artists' studios and shed light on their working methods. New principled, rigorous methods for estimating perspective transformations outperform traditional and ad hoc methods and yield new insights into the working methods of Renaissance masters. Sophisticated computer graphics recreations of tableaus allow us to explore "what if" scenarios, and reveal the lighting and working methods of masters such as Caravaggio. Biographical sketches: (taken from http://www.diatrope.com/stork/FAQs.html) (Short) Dr. David G. Stork, Chief Scientist of Ricoh Innovations, is a graduate in physics of the Massachusetts Institute of Technology and the University of Maryland at College Park, studied art history at Wellesley College and was Artist-in-Residence through the New York State Council of the Arts. He is a Fellow of the International Association for Pattern Recognition and has published six books/proceedings volumes and has one forthcoming, including Seeing the Light: Optics in nature, photography, color, vision and holography (Wiley), Computer image analysis in the study of art (SPIE), Pattern Classification (2nd ed., Wiley), and HAL's Legacy: 2001's computer as dream and reality (MIT). (Long) Dr. David G. Stork is Chief Scientist of Ricoh Innovations and has held academic appointments and taught in five departments at Stanford University frequently over the last 20 years. The breadth of his interests and contriburtions is revealed through the academic departments and programs in which he has held faculty positions in leading liberal arts colleges and research universities: Physics, Mathematics, Electrical Engineering, Statistics, Computer Science, Neuroscience, Psychology, and Art and Art History. He is a Fellow of the International Association for Pattern Recognition, "For contributions to pattern recognition education, machine learning, speech recognition, and the application of computer vision to the study of art" and Chair of its Technical Committee on Computer Vision in Cultural Heritage Applications. He has published six books/proceedings volumes and has another in production, including Seeing the Light: Optics in nature, photography, color, vision and holography (Wiley), the leading textbook on optics in the arts, Computer image analysis in the study of art (SPIE), the first volume in this discipline, Computer vision and image analysis in the study of art (forthcoming), Pattern Classification (2nd ed., Wiley), the world's all-time best-selling textbook in the field, translated into three languages and used in courses in over 250 universites worldwide, and HAL's Legacy: 2001's computer as dream and reality (MIT), the source of his PBS television documentary 2001: HAL's Legacy. A graduate in physics of the Massachusetts Institute of Technology and the University of Maryland at College Park, he also studied art history at Wellesley College and was Artist-in-Residence through the New York State Council of the Arts. He has performed research and taught courses such as "Light, color and visual phenomena," "The physics of aesthetics and perception," "Optics, perspective and Renaissance painting," and "Computer vision and image analysis in the study of art" over the last quarter century variously at leading liberal arts and research universities such as Wellesley College, Swarthmore College, Clark University, Boston University and Stanford University. He holds 38 US patents and has published numerous technical papers on human and machine learning and perception of patterns, physiological optics, image understanding, concurrency theory, theoretical mechanics, optics, image processing. He has served on the editorial boards of five international journals and has delivered over 58 plenary, invited or distinguished lectures at universities and conferences (atop over 230 traditional invited colloquia and seminars). He was one of four scientists invited to comment on Mr. Hockney's theory at the December 2001 "Art and Optics" Symposium at the New York Institute for the Humanities and one of two scientists invited to present a lecture in the symposium exploring the possible use of optics by early Renaissance painters at the Optical Society of America's Annual Meeting in Rochester, NY, October 2004. -- Aaron C. Courville Département d’Informatique et de recherche opérationnelle Université de Montréal email:Aaron.Courville@gmail.com -- Aaron C. Courville Département d’Informatique et de recherche opérationnelle Université de Montréal email:Aaron.Courville@gmail.com

1 0

Fwd: call for papers (Canadian Conference on AI 2010)
by Guillaume Desjardins 09 Nov '09

09 Nov '09

---------- Forwarded message ---------- From: Marina Sokolova <msokolova(a)ehealthinformation.ca> Date: Mon, Nov 9, 2009 at 8:39 AM Subject: 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

Fwd: Reminder: Sridhar Mahadevan colloquium TODAY (Fir Oct. 30)
by Guillaume Desjardins 30 Oct '09

30 Oct '09

---------- Forwarded message ---------- From: Doina Precup <dprecup(a)cs.mcgill.ca> Date: Fri, Oct 30, 2009 at 9:26 AM Subject: Fwd: Reminder: Sridhar Mahadevan colloquium TODAY (Fir Oct. 30) 3:30pm, MC 13 To: Guillaume Desjardins <guillaume.desjardins(a)gmail.com> Please forward to your seminar list Thanks, Doina Begin forwarded message: > From: Doina Precup <dprecup(a)cs.mcgill.ca> > Date: October 30, 2009 9:19:51 AM EDT (CA) > To: profs(a)cs.mcgill.ca, grads(a)cs.mcgill.ca, labrl(a)cs.mcgill.ca > Subject: Reminder: Sridhar Mahadevan colloquium TODAY (Fir Oct. 30) 3:30pm, MC 13 > > > 2009/10/30, MC13, 15:30 - 16:30 > > New Frontiers in Representation Discovery > Sridhar Mahadevan, University of Massachusetts-Amherst > > Abstract: > A longstanding challenge in artificial intelligence is how to automatically construct representations by analyzing state spaces. This problem has attracted renewed interest in a variety of areas, including compression and dimensionality reduction, planning and reinforcement learning, and semi-supervised and unsupervised learning. This talk will describe a family of spectral approaches to constructing new representations by diagonalizing or dilating Laplacian operators over a state space. A number of challenges and tradeoffs will be discussed, including task-specific vs. task-invariant representations; flat vs. multi-scale approaches; orthogonal vs. redundant bases; and dealing with the curse of dimensionality. A range of case studies from our recent research will be presented, including a paradigm for solving Markov decision processes where representation and control are learned simultaneously; a spectral method for clustering of text documents where the topic hierarchy is automatically constructed; and a compression method for computer graphics based on multiscale analysis of object geometry. > > > Biography of Speaker: > Sridhar Mahadevan is Co-Director of the Autonomous Learning Laboratory at the Department of Computer Science, University of Massachusetts, Amherst. His research interests are in artificial intelligence and machine learning. He is an associate editor of the Journal of Machine Learning Research, and was a tutorial speaker at AAAI 2007, IJCAI 2007, and ICML 2006. In 2008, he authored a book on representation discovery as part of the Synthesis Lectures in AI and Machine Learning series published by Morgan Claypool. > > As usual, refreshments will be served in the SOCS lounge after the talk. > > Hope to see you there! Feel free to forward to others who may be interested > Doina

1 0

Soutenance de
by vincentp＠iro.umontreal.ca 15 Oct '09

15 Oct '09

Vous êtes tous cordialement invités à assister à la soutenance de thèse de doctorat de Nicolas Chapados, qui aura lieu ce Vendredi 16 octobre 2009 à 13:00, salle 3195 Pavillon André Aisenstadt, 2920 Chemin de la Tour. TITRE: "Sequential Machine Learning Approaches for Portfolio Management" (un résumé en Français suit le résumé Anglais). ABSTRACT: This thesis considers a number of approaches to make machine learning algorithms better suited to the sequential nature of financial portfolio management tasks. We start by considering the problem of the general composition of learning algorithms that must handle temporal learning tasks, in particular that of creating and efficiently updating the training sets in a sequential simulation framework. We enumerate the desiderata that composition primitives should satisfy, and underscore the difficulty of rigorously and efficiently reaching them. We follow by introducing a set of algorithms that accomplish the desired objectives, presenting a case-study of a real-world complex learning system for financial decision-making that uses those techniques. We then describe a general method to transform a non-Markovian sequential decision problem into a supervised learning problem using a K-best paths search algorithm. We consider an application in financial portfolio management where we train a learning algorithm to directly optimize a Sharpe Ratio (or other risk-averse non-additive) utility function. We illustrate the approach by demonstrating extensive experimental results using a neural network architecture specialized for portfolio management and compare against well-known alternatives. Finally, we introduce a functional representation of time series which allows forecasts to be performed over an unspecified horizon with progressively-revealed information sets. By virtue of using Gaussian processes, a complete covariance matrix between forecasts at several time-steps is available. This information is put to use in an application to actively trade price spreads between commodity futures contracts. The approach delivers impressive out-of-sample risk-adjusted returns after transaction costs on a portfolio of 30 spreads. RÉSUMÉ: Cette thèse envisage un ensemble de méthodes permettant aux algorithmes d'apprentissage statistique de mieux traiter la nature séquentielle des problèmes de gestion de portefeuilles financiers. Nous débutons par une considération du problème général de la composition d'algorithmes d'apprentissage devant gérer des tâches séquentielles, en particulier celui de la mise-à-jour efficace des ensembles d'apprentissage dans un cadre de validation séquentielle. Nous énumérons les desiderata que des primitives de composition doivent satisfaire, et faisons ressortir la difficulté de les atteindre de façon rigoureuse et efficace. Nous poursuivons en présentant un ensemble d'algorithmes qui atteignent ces objectifs et présentons une étude de cas d'un système complexe de prise de décision financière utilisant ces techniques. Nous décrivons ensuite une méthode générale permettant de transformer un problème de décision séquentielle non-Markovien en un problème d'apprentissage supervisé en employant un algorithme de recherche basé sur les K meilleurs chemins. Nous traitons d'une application en gestion de portefeuille où nous entraînons un algorithme d'apprentissage à optimiser directement un ratio de Sharpe (ou autre critère non-additif incorporant une aversion au risque). Nous illustrons l'approche par une étude expérimentale approfondie, proposant une architecture de réseaux de neurones spécialisée à la gestion de portefeuille et la comparant à plusieurs alternatives. Finalement, nous introduisons une représentation fonctionnelle de séries chronologiques permettant à des prévisions d'être effectuées sur un horizon variable, tout en utilisant un ensemble informationnel révélé de manière progressive. L'approche est basée sur l'utilisation des processus Gaussiens, lesquels fournissent une matrice de covariance complète entre tous les points pour lesquels une prévision est demandée. Cette information est utilisée à bon escient par un algorithme qui transige activement des écarts de cours (price spreads) entre des contrats à terme sur commodités. L'approche proposée produit, hors échantillon, un rendement ajusté pour le risque significatif, après frais de transactions, sur un portefeuille de 30 actifs. COMPOSITION DU JURY: Président-Rapporteur: Pascal Vincent Directeur de Recherche: Yoshua Bengio Examinateur Externe: Doina Precup Membre du Jury: Michel Gendreau Représentante du Doyen: Yves Sprumont

1 1

UdeM-McGill-MITACS machine learning seminar Tues, Oct. 20th -
by Guillaume Desjardins 15 Oct '09

15 Oct '09

Yoshua Bengio will be giving a talk next Tuesday here at the Université de Montréal, entitled "On Training Deep Neural Networks". Hope many of you can make it ! Speaker: Yoshua Bengio, Université de Montréal Title: On Training Deep Neural Networks Location: Pavillon André-Aisenstadt (Université de Montréal), room 3195 Time: October 20th 2009, 12h30-13h30 Whereas theoretical work suggests that deep architectures might be more efficient at representing highly-varying functions, training deep architectures was unsuccessful until the recent advent of algorithms based on unsupervised pre-training. Even though these new algorithms have enabled training deep models, many questions remain as to the nature of this difficult learning problem. We attempt to shed some light on these questions in several ways, by comparing different successful approaches to training deep architectures and through extensive simulations investigating explanatory hypotheses. The experiments confirm and clarify the advantage (and sometimes disadvantage) of unsupervised pre-training. They demonstrate the robustness of the training procedure with respect to the random initialization, the positive effect of pre-training in terms of optimization and its role as a regularizer (in both cases in unusual ways). We explore explanatory hypotheses based on the notion that early growth of the model parameters is determinant, and in particular that early use of unsupervised learning places the dynamics of supervised learning in attractors associated with local minima with good generalization properties. We discuss how several training approaches for deep architecture may exploit the principle of continuation methods in order to find good local minima. In particular we suggest that this is the case of shaping or the use of a curriculum, showing that it has both an effect on the speed of convergence of the training process to a minimum and, in the case of non-convex criteria, on the quality of the local minima obtained. Finally, we investigate the nature and evolution of gradients at different levels of a deep supervised neural networks in an attempt to understand why training is sometimes slowed down, and sometimes possibly stuck in apparent local minima. -- Guillaume Desjardins

1 0