Open Research Projects
Open Undergraduate or Graduate Research Projects
Projects can lead to a 4th year project or a Master's Thesis
Some funding is available
Project 1: Combining Supervised and Unsupervised Learning
Supervised Learning consists of learning from experience using the outcome
of that experience. For example, assume that investing a certain amount of
money in a
given type of stock under some particular market condition proved very
successful since the value of that investment tripled in 3 months.
Conversely, assume that other investments proved disastrous as their value
40% in the span of one month. The purpose of supervised learning is to
discover the rules that, when applied, will allow us to invest properly so
as to make a maximum amount of profit, based on previous good and bad
Unsupervised learning, on the other hand, does not make use of such outcomes,
but it learns how to group similar types of data together. It is the type of
learning that gaves rise to zoological or botanical classifications:
from observing different animals carefully, it became clear after a while
that dogs and wolves share similar caracteristics, while cats, lions and
tigers form another group. These groups are different from those of birds and
so on. Such classifications did not use any type of outcome, but they grew
out of studies of these animals' various features. That's what
unsupervised learning is all about.
Although closely related and somewhat complementary, Supervised and
Unsupervised learning are usually treated as different research areas.
Yet, there could be great advantages in combining both approaches.
On the one hand, unsupervised learning could be used to study the data
prior to feeding it to a supervised classifier. This would have the advantage
of helping us tune the supervised learner or the data set in an intelligent way
that would help us improve its accuracy. On the other hand, unsupervised
learning could be used simultaneously with supervised learning so as to
bias the supervised learner in a meaningful way.
The purpose of this project will be to experiment with either or
both schemes. After being made familiar with both supervised and unsupervised
learning, the student will be given the opportunity to decide which of the two
problems, type of classifier and combination technique s/he would like to try.
Alternatively s/he could be given more guidance as to how to conduct the
Project 2: Learning how to Translate Locative Preposition from
English into French from Large Text Corpora
Machine translation of locative prepositions is not straightforward,
even between closely related languages. For example, while English
speakers will say: "The boy is ON the bus", French speakers will say:
``Le garcon est DANS le bus" (Literally, "The boy is IN the bus"). Similarly,
English speakers will say:
"The picture is ON the wall" while French Speakers will say "Le tableau est
AU mur" (Literally, "The picture is AT the wall"). The purpose of this project is to use statistical techniques and
Reference tools such as Dictionnaries, Thesauri, or WordNet to
browse through large corpora of bilingual texts and attempt to learn
translation rules that will associate the correct French translation to the
English locative preposition used in a given context.
Since this project can be managed in two phases (a French and an English phase)
separately, although preferable, knowledge of both English and French is not
required. After learning the various text processing tools, the student will
be given the freedom to choose the direction for his/her project that interests
him/her most. Alternatively, s/he will be given more specific guidance.
Project 3: Your Idea
If you have an idea related to Artificial Intelligence, Machine Learning
and/or Text Processing, I will be glad to discuss it with you. Hopefully,
we will be able to negotiate a topic of interest to us both!