Sutanu Chakraborti
I
am an Assistant Professor at the Department of Computer Science and Engineering
at IIT Madras. My research interests are broadly in the areas of language,
memory and learning. I offer three electives in these areas: Natural Language
Processing (NLP), Memory Based Reasoning in AI (MBR) and Introduction to
Machine Learning (IML, to be launched next semester -- August 2010). The first
two courses have been offered jointly in the past with Dr. Ravindran
Balaraman and Prof. Deepak Khemani.
Dr. Ashish Tendulkar is
sharing the IML course with me next semester.
Here
are some fairly open-ended questions that drive my research. I am not equally
active on all of these, but I wish I were. I have also been involved in shorter
term problems, some derived out of these long term challenges, and some others
which are independently motivated.
- How can we build real
world applications that can help big organizations make best use of the
wealth of wisdom hidden in their unstructured data repositories (project
documentations, proposals, white papers, user groups, blogs,
research reports, employee feedback, resumes)?
- Can machines reasoning
over unstructured text learn by introspecting over failures? Instead of
working hard at solving problems end to end, can they seek our help
intelligently at times? Semi supervised learning is particularly of
interest in this context. The broader question is: how best can we exploit
the complementary abilities of humans and machines in building
conversational problem solvers that can communicate in natural language?
This may need us to take a broader look at systems, the way they represent
knowledge to enable transparency and interpretability, the way they
interact with us (user interfaces and visualizations) and the way they
facilitate learning and evolution - ours and theirs.
- How do language and
memory interact? Even before you read an assassination report in your
morning newspaper, you would have glossed through the news heading and
begun anticipating the questions the report aims at answering (Who killed
whom? Where? How? Why? What is unique about the manner in which the crime
is committed?) Memory helps in understanding language, understanding
language in turn helps in building up and organizing memory. Can we build
cognitive/computational models that model learning through interaction of
memory and language?
- How can different
sources of knowledge (statistical, background, linguistic) integrate to
facilitate better text analysis over tasks like classification or question
answering? Can we do a comparative analysis of different modes of
combination?I see this line of
work deriving inspiration from cognitive studies on humor. I believe jokes
illustrate beautifully what goes wrong when we fail to combine the
knowledge sources appropriately.
- How can knowledge
mined from text be represented at various levels of abstraction (general
through specific)? How can a system identify which level is appropriate
for a task? A classification task may need knowledge at a very different
level of abstraction, when compared to a question answering task which may
need a very high level of granularity. Also, when we talk to novices, we
use level of abstraction very different from the one we use when we talk to
experts. So the problem of choosing the right level of representation is
as important to generation (NLG) as it is to understanding (NLU).
- How do we measure
"complexity" of a collection of documents, in the supervised as well as
unsupervised learning scenario? Tasks could be classification, retrieval,
question answering. In a FAQ dataset, we may be interested in determining
how well the problem and solution components are "aligned" .This is of
interest in the area of Case Based Reasoning (CBR) where a case is
typically a recorded instance of a successful problem solving episode, and
can be viewed as a problem-solution pair. In this context, the alignment
problem maps onto the question : how strongly can
we say that similar problems have similar solutions? The complexity (or
alignment) is a function of our choice of representation, so this can help
us in choosing between representations (those that give higher alignment/
lower complexity are better). A local level (case specific) analysis of
complexity can also have implications in "casebase
maintenance" - knowing which cases to retain and which to throw away.
- How do the physics of
matter relate to the physics of information? There have been very
interesting but isolated attempts at relating theories of nonlinear
dynamics, quantum mechanics, statistical mechanics and thermodynamics to
cognitive processes. Is there really a deeper link that can be exploited
in simulating interesting cognitive phenomena? I have sketched some
preliminary ideas in a recent workshop paper that explores models of human
memory founded on nonlinear dynamics. I would love to collaborate with
physicists on this front.
|
Degree |
Year |
University
/ Institute |
Specialization |
|
Ph.
D. |
2004-2007 |
The
Robert Gordon University, Aberdeen, U.K. |
Computer
Science |
|
M.S.
(by Research) |
2002-2004 |
Indian
Institute of Technology, Madras |
Computer
Science and Engineering |
|
B.E.
(Hons.) |
1994-1998
|
Regional
Engineering College (now NIT), Rourkela |
Applied
Electronics and Instrumentation Engineering |
Deepak
Padmanabhan (PhD, jointly with Prof. Deepak Khemani)
Anil
Patelia (MS by Research)
Kevin
Joseph (MS by Research, jointly with Prof. Deepak Khemani)
Karthik Jayanthi (Dual Degree)
Ravi Kiran (Dual Degree)
Vasanth
R. (B.Tech)
K Sujit Kumar Reddy(B.Tech.)
1 semester at
Courses
Taught: Soft Computing, Memory-Based Reasoning in AI, Natural Language
Processing (all graduate level electives), Principles
of Communication (an undergraduate level core course).
Approximately
7 years at Tata Research Development and Design Centre as a scientist and later
as project leader of the Case-Based Reasoning Research Team.
Publications< link >
Half
day tutorial at International Conference on Natural Language Processing
(December 2009, Hyderabad) on "Mining Concepts from Words"
Half-day
tutorial on invitation at Knowledge-Based Computer Systems (KBCS) Conference
(India) in December 2002, on "Case-based Reasoning: Retrieval Algorithms and
TRDDC experiences" (jointly with Mr. Vivek Balaraman)
Received the Annual Excellence Award of TRDDC, for
innovative contributions in the area of Case-based Reasoning that led to the
conception, design, development and deployment of a CBR-based Directory
Assistance System run by a leading teleservices company in
Selected
for inclusion in Marquis Who's Who in Science and Engineering, 10th
Anniversary Edition (2007) for contributions to Computer Science and
Engineering.
Other
Mentions: 2nd in the state of Tripura in Higher Secondary
Examination, and 1st in the state of Tripura in ICSE exam (10th
standard), 2nd in state in National Talent Search Examination, All
India Topper in the subject Computer Science in ICSE Exam (10th
Standard) as declared by ICS (the coordinator of courses in several
schools)
Announcement: Project Position
Once
upon a time, I wished I could write poems. Here is one attempt.
Book of Life
My book of life gets written
Every passing day
By things I couldn't do
Words I couldn't say.
The undone and the unsaid
A silent life do share
Where the said and the done
Seek meaning in despair.
I can feel it not so quite,
I can hear it not so clear;
How my deal for days to come,
Their silent life does steer.
That indeed is silence within,
That indeed is joy profound;
The book is written all over,
And the author never found.
Snail
mail:
Room No.
BSB 304
Department of Computer Science and Engineering
IIT Madras,
Chennai 600 036
Email:
(This page was last updated on 1st
June 2010).