C e n t r e d e r e c h e r c h e s m a t h é m a t i q u e s
Organizers:
Yoshua Bengio, Département d'Informatique et Recherche Operationnelle, Université de Montréal
It has been recognized in the last few years by many corporations that they possess an almost untapped source of information to improve themselves: the large amount of computerized data that they are collecting on their processes and their customers. Machine learning algorithms are becoming very important technological tools in many applications such as data mining, in which one wants to extract useful information from large databases, and they are particularly important when the probability distribution of that data is not known ahead of tiem. Machine learning algorithms and their analysis focus on the problem of generalization: it is not enough to extract some information from the data (e.g. to characterize the relation between some variables), we want this information to generalize well to new data, so that it becomes really useful information.
In this regard, an old question is that of "model selection", that is the choice of a class of functions, or the ways to impose a preference over functions which make the learning problem well-posed. For this, it would be very useful to estimate the expected generalization performance that would be obtained with a particular preference of function class. One could then pick the function class that is expected to yield the lowest error, or combine functions from the functions classes with the lowest expected error. For this purpose, many approaches have been proposed in the past, both in the statistics and the machine learning community.
In the area of machine learning algorithms, there has recently been a lot of interest in new ways to evaluate generalization error, to optimize it, and to combine or select models, e.g, the Structural Risk Minimization approach along with Support Vector Machines, various Boosting algorithms and the Bagging algorithm (which combine several models). These new approaches suggest that better generalization performance can be obtained using certain rather general procedures.
The following leaders in this field have tentatively accepted to participate to the workshop as invited speakers:
P. Bartlett (Australia National University), L. Breiman (Berkeley University), T. Dietterich (Oregon State University), Y. Freund (AT&T), R. Neal (University of Toronto), M. Perrone (IBM), R. Schapire (AT&T), G. Wahba (University of Wisconsin at Madison).
The workshop will comprise 50-minute lectures from the invited speakers as well as shorter presentations by participants from academia and industry, over three days April 12th to April 14th 2000, and it will be held on the premises of the Centre de Recherches Mathematiques (CRM), on the campus of the Universite de Montreal, in Montreal. The workshop will be sponsored by the CRM as well as by the MITACS (Mathematics of Information Technology And Complex Systems) Network of Centers of Excellence.
Those wishing to present a communication are invited to submit an abstract (one or 2 pages) by e-mail (in plain ASCII, postscript or pdf) to : bengioy@iro.umontreal.ca or dale@cs.uwaterloo.ca
Please register by March 27, 2000 for attendance and accommodation, or April 3, 2000 for attendance only.
Hans Brungs | University of Alberta | Sibylla Priess-Crampe | Universität München |
---|---|---|---|
Barry Green | University of Stellenbosch | Mark Spivakovsky | University of Toronto |
Werner Luetkebohmert | Universität Ulm | Bernard Teissier | École Normale Supérieure, Paris |
Alexander Prestel | Universität Konstanz |
Andrew Carson | University of Saskatchewan | Murray Marshall | University of Saskatchewan |
---|---|---|---|
Franz-Viktor Kuhlmann | University of Saskatchewan | Deirdre Haskell | College of the Holy Cross |
Salma Kuhlmann | University of Saskatchewan | Hans Schoutens | Wesleyan University |
Shreeram Abhyankar | Purdue | Francois Loeser | Paris |
---|---|---|---|
Carlos Andradas | Madrid | James Madden | Baton Rouge |
Ron Brown | Hawaii | Jan Minac | Western Ontario |
Alexandru Buium | Urbana | Freddy van Oystayen | Antwerpen |
Gilles Christol | Paris | Olivier Piltant | Paris |
Vincent Cossart | Versailles | Florian Pop | Bonn |
Michel Coste | Rennes | Patrick Popescu-Pampu | Paris |
Tom Craven | Hawaii | Victoria Powers | Emory |
Dale Cutkosky | Missouri | Ana Reguera | Valladolid |
Nikolai Dubrovin | Vladimir | Paulo Ribenboim | Kingston |
Yuri Ershov | Novosibirsk | Peter Roquette | Heidelberg |
Jose Engler | Campinas | Mohamed Saidi | Bonn |
Joachim Graeter | Potsdam | Thomas Scanlon | Berkeley |
Urs Hartl | Ulm | Claus Scheiderer | Regensburg/Duisburg |
Roland Huber | Wuppertal | Erwin Schoerner | Munich |
Sudesh Khanduja | Chandigarh | Niels Schwartz | Passau |
Hagen Knaf | Heidelberg | John Shackell | Canterbury |
Jochen Koenigsmann | Konstanz | Patrick Speissegger | Toronto |
Leung Ka Hin | Singapore | Michel Vaquie | Paris |
Quing Liu | Bordeaux | Adrian Wadsworth | San Diego |
This conference is dedicated to Paulo Ribenboim, in recognition of his extensive contributions to the subject. Tutorials will be given on July 26 and 27. The conference will be held from July 28 through August 4. There will be a special session in honor of Paulo Ribenboim on July 31, and the informal workshop will be held from August 5 through August 11.
The conference is intended to cover recent developments in valuation theory and its applications: algebraic geometry (especially local uniformization), real algebraic geometry (and quadratic forms), Galois theory, rigid analysis and curves over valuation rings, model theory of valued fields (especially in positive characteristic), o-minimal expansions of the reals (and Hardy fields), ultrametric spaces and spherically complete fields, p-adic numbers, non-commutative valuation theory.
The main topics of the Workshop will be: Local uniformization and resolution of singularities, model theory of valued fields in positive characteristic and its connections with resolution of singularities, the theory of valued function fields, approximate roots and related subjects, o-minimal expansions of the reals and Hardy fields. In addition to these subjects, the workshop will offer an opportunity to discuss other recent developments and open problems which are connected to the scientific program of the conference.
3 February 2000, webmaster@CRM.UMontreal.CA