Aperçu général du programme
Programme RFIA 2004 (Version HTML)
Programme RFIA 2004 (Version PDF)
Conférenciers Invités:
Toward True 3D Object Recognition
Jean Ponce
ponce@cs.uiuc.edu
Beckman Institute and Department
of Computer Science
University of Illinois at Urbana-Champaign
Résumé:
This talk addresses the problem of recognizing three-dimensional (3D) objects
in photographs and image sequences, revisiting viewpoint invariants as a -local-
representation of shape and appearance. The key insight is that, although
smooth surfaces are almost never planar in the large, and thus do not (in
general) admit global invariants, they are always planar in the small---that
is, sufficiently small surface patches can always be thought of as being comprised
of coplanar points---and thus can be represented locally by planar invariants.
This is the basis for a new, unified approach to object recognition where
object models consist of a collection of small (planar) patches, their invariants,
and a description of their 3D spatial relationship. I will illustrate this
approach with two fundamental instances of the 3D object recognition problem:
(1) modeling rigid 3D objects from a small set of unregistered pictures and
recognizing them in cluttered photographs taken from unconstrained viewpoints;
and (2) representing, learning, and recognizing non-uniform texture patterns
under non-rigid transformations. I will also briefly discuss extensions to
the analysis of video sequences and the recognition of object categories.
Joint work with Svetlana Lazebnik, Frederick Rothganger, and Cordelia Schmid.
Bio:
Jean Ponce received his Thèse d'État from the University of
Paris Orsay in 1988. He is an associate professor in the Department of Computer
Science at UIUC and a full-time Beckman Institute faculty member in the Artificial
Intelligence Group. His fields of professional interest are computer vision
and robotics.
Statistical model for geometric inference from images
Kenichi Kanatani
kanatani@suri.it.okayama-u.ac.jp
Department of Information Technology
Okayama University, Okayama
Résumé:
We investigate the meaning of "statistical methods" for geometric
inference based on image feature points. Tracing back the origine of feature
uncertainty to image processing operations, we discuss the implications of
asymptotic analysis in reference to "geometric fitting" and "geometric
model selection". We point out that a correspondence exists between the
standard statistical analysis and the geometric inference problem. We also
compare the capability of the "geometric AIC" and the "geometric
MDL" in detecting degeneracy.
Bio:
Kenichi Kanatani was born on August 12, 1947 in Okayama, Japan. He received
his B.S., M.S, and Ph.D. in applied mathematics from the University of Tokyo,
Japan, in 1972, 1974, and 1979, respectively. He joined the Department of
Computer Science, Gunma University, Kiryu, Japan, in April 1979 as Assistant
Professor. He became Associate Professor and Professor there in April 1983
and April 1988, respectively. From April 2001, he is Professor of Information
Technology, Okayama University, Okayama, Japan. He was a visiting researcher
at the University of Maryland, U.S.A., the University of Copenhagen, Denmark,
the University of Oxford, U.K., and INIRA at Rhone Alpes, France. He is the
author of ``Group-Theoretical Methods in Image Understanding'' (Springer,
1990), ``Geometric Computation for Machine Vision'' (Oxford University Press,
1993) and ``Statistical Optimization for Geometric Computation: Theory and
Practice'' (Elsevier Science, 1996).
His his research career started with studies of theoretical continuum mechanics
(elasticity, plasticity, and fluid) and its application to mechanics of granular
materials such as powder and soil, but his research interested has shifted
to mathematical analysis of images and 3-D reconstruction from images. Currently,
he is devoted to mathematical analysis of statistical reliability of computer
vision and optimization procedures. He is an IEEE Fellow.
Local
and Descriptive Data Mining: A Robust and Scalable
Alternative to Global Modeling
Prof. Dr. Stefan Wrobel
stefan.wrobel@ais.fraunhofer.de
Fraunhofer Institute for
Autonomous Intelligent Systems (AIS)
Sankt Augustin
Résumé:
In the fields of Machine Learning (ML) and Knowledge Discovery in Databases
(KDD), research and applications for a long time have primarily been concentrating
on the induction of global predictive models, i.e., hypotheses that are capable
of producing a prediction of a value of interest for any unknown object with
which they are confronted. If such a model can be induced, it is a complete
model of a phenomenon of interest, and thus constitutes a comparatively strong
discovery. On the downside, such models are difficult to produce in many applications,
for two reasons. Firstly, the available descriptors may not be sufficient
to actually express the complete functional relation of interest. In this
case, learning will fail to produce a good and usable model. Secondly, due
to the requirement of having a complete model, components of a model are all
interrelated, greatly adding to the complexity of search.
For these reasons, there is strong
interest in the data mining community in techniques which produce local,
descriptive models instead of global predictive models. Such models
do not guarantee complete coverage of all possible situations, but try to
find subspaces about which useful statements or models can be formulated.
Such local models can often be found in situations in which global modeling
is impossible, thus allow interesting discoveries in domains which are inaccessible
to global learning techniques. Secondly, since patterns are local and descriptive,
they can be discovered independently of each other, offering enormous potential
for speeding up discoveries; in fact, the fastest data mining algorithms available
are local descriptive discovery algorithms.
In the talk, we will introduce the topic of local descriptive discovery and illustrate it with some of the basic algorithms and tasks in this area, namely discovery of association rules and discovery of subgroups. We will present sample applications and discuss extensions of the method to probabilistic discovery and structured and geographic datatypes.
Bio:
Prof. Dr. Stefan Wrobel, M.S., studied computer science in Bonn and Atlanta,
GA, USA (M.S. degree, Georgia Institute of Technology), receiving his doctorate
from Univ. of Dortmund in 1993. He has been in active in Machine Learning
since 1986, first at Technical Univ. of Berlin, then from 1989 at GMD as a
research scientist and later leader of the Machine Learning/Data Mining Group.
In 1996, he co-founded Dialogis GmbH, also serving as one of the company's
technical directors. In 1998, he became professor of computer science at Univ.
of Magdeburg, leading the group "Knowledge Discovery and Machine Learning".
Since 2002 he is a professor of computer science at Univ. of Bonn and institute
director at Fraunhofer AIS, leading the "Knowledge Computing" area.
Prof. Wrobel has continuously been publishing on subjects in Machine Learning
and Data Mining, has (co-)organized conferences and workshops and is serving
on the editorial board of several journals and conferences. He is elected
founding member of the International Machine Learning Society (IMLS), and
a member of the management board of KDnet, the European network of excellence
on Knowledge Discovery.
Application
des methodes d'Intelligence Artificielle et d'Apprentissage (Modeles Graphiques,
Reseaux de Neurones Recursifs,
Kernel Methods, etc) a la chimie organique au sens large (prediction de la
structure des proteines, docking and drug design/screening, prediction de
toxicite, etc) et les enjeux futurs.
Pierre Baldi
pfbaldi@ics.uci.edu
University of California,
Irvine
Research focus:
Bioinformatics/ Computational Biology
Machine Learning/AI/Data Mining
Communication Networks
Projects in my group include developing machine learning and other statistical
methods for AI and large-scale data analysis, understanding and predicting
protein structures, computationally screening and designing new drugs and
chemical interactions, modeling and understanding metabolic, signaling, and
regulatory networks (systems biology), building a computer GO player, understanding
genome evolution, analyzing and designing communication networks (Internet,
Ultra Wide Band Radio).
Bio:
Pierre Baldi is a Professor in the School of Information and Computer Science
and the Department of Biological Chemistry at the University of California,
Irvine and the Director of the Institute for Genomics and Bioinformatics.
Born and raised in Europe, he received his PhD from the California Institute
of Technology in 1986. From 1986 to 1988 he was a postdoctoral fellow at the
University of California, San Diego. From 1988 to 1995 he held faculty and
member of the technical staff positions at the California Institute of Technology
and at the Jet Propulsion Laboratory. He was CEO of a startup company from
1995 to 1999 and joined UCI in 1999. He is the recipient of a 1993 Lew Allen
Award at JPL and a Laurel Wilkening Faculty Innovation Award at UCI. Dr. Baldi
has written over 100 research articles and four books: Modeling the Internet
and the Web-Probabilistic Methods and Algorithms, Wiley, (2003); DNA Microarrays
and Gene Regulation-From Experiments to Data Analysisand Modeling, Cambridge
University Press, (2002); The Shattered Self-The End of Evolution, MIT Press,
(2001); Bioinformatics: the Machine Learning Approach, MIT Press, Second Edition
(2001). His research focuses in AI, machine learning, and bioinformatics.
RoboCup: New Scientific and Technical Advancements
Enrico Pagello 
Department of Information Engineering
University of Padua, Italy
Résumé:
After
seven years of successful RoboCup international meetings, it is possible to
set a first balance of the scientific and technical advancements, that have
been made possible in the field of AI and Robotics, thanks to the RoboCup
community. We will discuss this issue, with particular attention to the last
RoboCup-2003 Event held in Padua (Italy). We will illustrate also some specific
research topics that are conducted by the Team "Artisti Veneti",
of the University of Padua, that is participating to RoboCup since its beginning.
Bio:
Enrico
Pagello was born in Vicenza, Italy, on Nov. 17, 1946. Full Professor, Dept.
of Information Engineering (DEI), The University of Padua and part-time Research
Scientist at Institute of Biomedical Enginering of the National Research Council
(ISIB-CNR)
Fields of activity: Artificial Intelligence, Robotics, Distributed