|
Courses and Workshops
|
| Title: | Statistical learning for natural language processing |
| Lecturer(s): | Eric Gaussier (Xerox Research Centre Europe) and Nicola Cancedda (Xerox
Research Centre Europe) |
| Type: | Advanced Course |
| Section: | Language and Computation |
| Week: | First | | Time: | 11.00-12.30 (Slot 2) |
| Webpage: | http://www.xrce.xerox.com/people/gaussier/
|
| Reader: | Course will not have a reader | | Room: | A3 |
Description Statistical Learning provides an array of methods for classes of
problems such as categorization, regression, clustering, and probability
density estimation. This course will try to convey the reflex of
abstracting away from details and asking: "is this NLP problem an
instance of a general class for which effective solutions exist?"
To this end, the course will review major learning algorithms and
principles (including kernel methods and SVMs), and will present how some
NLP research problems relate to them. A particular attention will be
given to issues common in NLP such as the representation of
the data, the high dimension of feature spaces, the use of models with
hidden variables, and the need for combining annotated and non-annotated
data.
These considerations will provide the ground for discussing central
theoretical issues (bias-variance trade-off, notion of model capacity,
discriminative vs generative models), which will be illustrated on
standard NLP tasks.
PREREQUISITES
Basic notions of linear algebra:
- vector spaces
- linear transformations
- eigenvalues and eigenvectors of a square matrix
Basic notions of probability theory:
- random variables, density funcitons,
- standard probability distributions (binomial, multinomial, Gaussian)
BRIEF CV OF THE LECTURERS
Dr. Eric Gaussier is Project Leader of the XRCE MLTIA (Machine Learning
for Textual Information Access) group. In 1995, he received a PhD from
University Paris 7 in theoretical computer science. His initial
projects at XRCE centred on NLP analysis for (cross-language)
information retrieval. He is currently working on new models for
categorizing and clustering documents and entities, and new methods for
extracting information from corpora for cross-language applications. He
published numerous research papers on natural language processing and
textual information access, served as co-editor for the International
Journal of Corpus Linguistics, co-edited a book on Intelligent
Assistance to Information Retrieval, is a member of the editorial board
of the journal Traitement Automatique des Langues.
Dr. Nicola Cancedda joined XRCE as a Research Engineer in 1999 after
receiving a PhD in Computer Engineering from the University of Rome
"La Sapienza" for a dissertation on automatic summary
generation.
Before joining XRCE he was also international fellow at the Artificial
Intelligence Center of SRI International in Menlo Park, California, and
consultant for Finsiel SpA in Rome, Italy. His main research interest
is the application of Machine Learning methods to natural language
processing and textual information access. He currently acts as XRCE
site manager for the IST project KerMIT.
|
© 2003, 2004 LORIA, Carlos Areces. |
2003-11-20
| |
