Research Interests
I am interested in building algorithms that
learn from experience,
to be able to perform their tasks better.
Most of my current work has a strong application pull
-- i.e., is motivated by some specific tasks.
Some other projects are more technology push --
where the goal is more exploring some foundation or mathematical framework,
rather than solving some application.
(Each research project is placed in only one list, based on its main emphasis;
almost all of the project actually involve both aspects.)
See also
Research Summary 2006-2011 (NSERC Form 100),
and miscellaneous extended webpages.
- Patient-Specific Survival Prediction:
a novel algorithm for learning patient-specific survival time distribution,
based on all available patient attributes -- basically a personalized version
of Kaplan-Meier curve, that can be used to visualize the survival rate of
an individual patient, or to predict median survival time, or whatever.
- Explaining
the Gene Signature Anomaly:
formally investigating the overlap of the top ranked features in two lists
whose elements are ranked by their respective Pearson correlation coefficients
with the same outcome.
- Budgeted learning:
deciding which features of which training instances to purchase,
to produce an effective classifier,
when the learner has a fixed budget for such purchases
- Learning belief nets:
Webpages with Details
Education and Popularizing
Earlier work
|