Research
I work on machine learning by probabilistic methods
and reasoning in uncertainty. I also create algorithms that make these
tasks efficient for large high-dimensional data sets.
Here are some problems that interest me right
now:
- Clustering by eigenvalues and eigenvectors is an emerging
technique with roots in graph theory. It already has applications in
image segmentation, web and document clustering, bioinformatics and
linguistics but its full potential is still unexplored.
- Understanding
the problem of multiclass classification; how do we take
classification from "flat" decisions to hierarchical decisions?
- Comparing clusterings Given two clusterings, or two partial
clusterings, how different are they? There is more than
one way of measuring the distance between two clusterings,
and some of them have exciting connections with
combinatorics and the lattice of partitions. This work
relates closely to the question: is a clustering algorithm
(significantly) better than another one?
- Intransitivity in classification and choice.
It has been often noted that people's choices are not transitive: in
other words, their preferences between K objects are not consistent
with an ordering. Economic theory of choice has introduced various
theories explaining how the observed intransitivity may
arise. However, there is no work to date on how one may infer these
models from data. Among the things I want to do: to formulate
estimation problems for the hidden context and other models of
intransitivity that are relevant to practical domains; to define when
the model is identifyable (it may not be when the number of components
K is large) and to design rigorously founded algorithms to estimate
it.
This problem is also of relevance to artificial systems, like
multiclass classifiers. One common approach to deciding between K
classes is to construct several binary classiffiers and then to
combine their output. Since there often is no way to constrain the
binary classifiers' otuputs to be consistent with an ordering, the
problem is naturally one of dealing with intransitive
``preferences''. Joint work with Jeff Bilmes.
- Proteomics Interpreting the very complex signature of an
aminoacid sequence that is subjected to collision induced dissociation
(CID). Probabilistic identification of the protein composition of a
complex mixture from high troughput mass spectrometry data.
In the past I have worked on these topics as well: approximate
inference and structure finding in belief networks, fast algorithms
for learning graphical models in high dimensions, optimal
triangulation of Bayes nets, transfer of learning, reinforcement
learning, mixtures of experts.
For Romanian PhD candidates only
Teaching
One time offerings
-
STAT 591/EE 596 Modern methods of machine learning: multiway classification, preferences, intransitivity with Jeff
Bilmes, Autumn 2006
- STAT 593 BProbabilistic reasoning
with graphical models
with Jeff
Bilmes, Spring 2005
STAT
592 B/ CSE 590 MM Classic methods of Machine Learning (with Alejandro Murua),
Winter 2004
STAT 592 C Winter 2002 (with Jeff
Bilmes and Thomas Richardson)
Click here for the course schedule
STAT 390 B Probability and Statistics for Computer
Science Spring 2001 Version V.2.0 of this course is being
taught now as STAT 391.
STAT
593 C Information Theory, Statistics and Machine Learning Spring
2001, joint with Jeff
Bilmes
Editorial duties
- Action Editor, Journal of Machine Learning Research
- Associate Editor, IEEE Transactions on Pattern Analysis and Machine Intelligence
- Associate Editor, Book Reviews, Journal of the American Statistical Association & The American Statistician
- Editorial Board, Foundation and Trends in Machine Learning
- Co-chair with Xiaotong Shen, the 11th Artificial Intelligence and Statistics Conference, AISTATS 2007
- Area co-chair, Neural Information Processing Systems conference, NIPS 2001, 2002
"Clustering by weighted cuts in directed graphs" with William Pentney, presented at the 2007 SIAM Conference on Data Mining SDM 2007