Research Overview – Prof.

Amitabha Sanyal

What is research – It’s about critical thinking. Knowing something well enough is completed if one
extends its frontiers. Students first identify their area of interest. They should choose courses according
to their interest and as well as might take some basic courses. For deciding area of interest talking to
seniors, faculties or attending talks may help. Students should take research problems which are useful,
interesting and doable.

Area of interest – Program Analysis (Analyze programs to find useful properties).

HEAP REFERENCE ANALYSIS – Traditionally garbage collection of the heap data is found by analyzing
reachability. But critical thinking and reasoning can be if one can guarantee that some nodes would not
be accessed in future then they can also be garbage collected. So the path of research can be data
gathering, work out the analysis, prove the correctness of the analysis, implementation and then writing
a paper. There is thinking going on if this method can be applied to Scheme as this type of language has
better opportunity for such optimization.

TRANSLATION VALIDATION – Validate the program produced by the compiler. There can be direct
correspondence between the original program and the compiler output. But most of the time the
scenario is more complicated. This is a very interesting area of research and needs a lot of critical
thinking.

R & D Overview – Prof. D M Dhamdhere

R & D is the most important component in the M.Tech course. The courses are to create the background
for the R & D projects. And the other component is the TAship which is only to help the system to run
conveniently and ensure that the faculties have more time to devote to R & D. R & D is a creative work
aimed at extending the frontiers of the field. There are many kinds of R & D. At the first stage of the art
it may be desired only to make it work. Then make it ‘better’. Then it is necessary to prove correctness.
Next to have a method to perform it. All these steps are necessary for research.

Not every student is going to choose Research as their career. But R & D is still useful as it makes an
individual think individually and analyzes its feasibility, effectiveness, limitation etc. This makes one to
think open mindedly.

Area of research – Code optimization, Operating Systems, Distributed Systems.

OPERATING SYSTEMS – Operating systems was a difficult area until the open source OSs came. Recently
he is working in file systems. He also works for improving execution performance of processes in Linux.
This needs kernel-level works.

DISTRIBUTED SYSTEMS – Distributed algorithms need critical and intellectual thinking. Lazy Vector global
clocks in the distributed systems are used for knowing temporal behaviors. They can be used for
discovering race conditions. This is also a current interest of Prof. DMD.

Research Overview - Prof. Puru (puru@cse)

Many faculties are in this group – Varsha Apte, Kameswari Chebrolu, Bhaskaran Raman, Puru, Anirudh
Sahoo, Umesh Bellur, Sridhar Iyer, Krithi Ramamritham, and some EE faculties.

Get more information from http://www.cse.iitb.ac.in/synerg

Research areas – Wireless networks, Distributed systems, Performance analysis, Virtualization, Sensor
systems and E-learning.

1st three projects are developed for developing regions.

FRACTEL – Extending WiFi for long distance links. This would find a solution for the rural area
connectivity problem. This project have issues like TDMA based MAC implementation, study long
distance behaviors etc. In future they would incorporate power optimization. IIT Bombay has a mesh
network test bed between hostels and the academic buildings and some of these links are
approximately 1k.m. long.

Lo3 (Low cost, Low power, Local voice and messaging) – This project is aimed at filling the gap of cellular
connectivity in the rural area. It is a 802.15.4 based network. This would benefit the rural users as there
would be no cumulative costs to pay to the cellular distributors and the technology itself have low
deployment cost.

Congestion detection and /Traffic Estimation in Developing Regions – The traffic estimation is beneficial
for both the people and the road designers. State of art is to fix sensors in the road or have mobile or
GPS tracking systems to have the estimation. But these techniques are not the direct measurement of
the vehicle speed or number of vehicles. New idea is Horn-OK-Please. Record honk and then analyze it
to get the information about the vehicle.
Resource Provisioning and Management in Virtual Execution Environments – Clients can hire computing
service. This is possible by the virtue of virtualization.

Many other areas of research @ SynerG are like Esense etc.

Students should not miss two things in IITB – nature@iitb and Mumbai.

Computer Science-driven Interdisciplinary Research – Prof. Srinivas

Many fields need computer science to solve their problems. For solving these problems many core
computer science problems are addressed. To formulate the problem one have to interact with the
experts of that area. Then develop algorithms and implement them and in this way this ends up with
extending core areas of computer science.

One interesting problem might be genome sequencing. Human beings have billions of genomes so to
sequence them we need powerful computing systems. High throughput sequencing is gaining
momentum these days. Performance is increased about 10 folds every year. Where CERN project has
deployed super computers to analyze its results a few years ago, now comparable amount of data is
being generated by a single bio-technology laboratory. So some open issues are how to query petabytes
of data efficiently, compare many genomes etc.

Research Overview – Prof. Abhiram Ranade

His research interest covers scheduling problems, packing problems and other problems like graph
partitioning etc. For example, given a train schedule decide the timings of new trains that can be
introduced or find the number of new trains that can be introduced. He will offer seminar in the next
semester and later MTP but those who are interested should take at least one theory course in this
semester.

Prof. Ranade is also interested in implementation. Mumbai Navigator gives Mumbai bus routes.

His recent interest is in DrawCAD – alternative approach to machine drawing. Use mouse to draw and let
the machine guess rightly what the user want to draw and ‘beautify’ accordingly. Input method can also
be stylus etc instead of mouse.
Why this DrawCAD is an interesting topic – this would make typing Indic language easy, smart device
power can be exploited farther.

Issues like removing noise, and context dependent beautification would come in the path. These can be
solved using image processing, machine learning and math programming concepts. There can be a
comparison with AutoCAD which does not use a convenient way to draw. Currently work is going on in
this project to find the right specification. And there is a large opportunity for future work. Prof. Ranade
will offer seminar based on these issues.

Natural Language Processing and associated machine learning research

Based on the core areas of search, reasoning and learning of artificial intelligence, fields like Natural
Language Processing developed. NLP has two main components like Machine learning and linguistics. To
reveal the semantic information in the texts linguists create graphs and discovers for words with which it
is bind to. It is challenging to create parser for natural language as compared to structured languages.
This is where machine learning comes into play. The machines need to be trained with examples.

At IITB various works are going on in the field of machine learning and natural language processing like
Information extraction, IR, Machine Translation, Web 2.0 Application. IITB got a United Nations project
to build UNL. Govt. of India gave NLP group in IITB the project of developing technologies for Indic
languages. Now they are working on building Indowordnet and in fields like sentiment analysis, text
entailment etc. One of the main aims of this NLP group is to build multilingual WSD with Indicwordnet.

Sometimes it happens that some query doesn’t fetch the desired result. So if the client is given the
freedom of expressing the query more freely and in any language then searching might be lot easier and
effective.

NLP group is funded by United Nations, Govt. of India and many companies like IBM, Microsoft, HP Labs
etc.

For more information please visit – http://www.cfilt.iitb.ac.in or http://www.cse.iitb.ac.in/~pb

Research activities at GCC Resource Center – Prof. Uday Khedker

One of the main aims of the GCC Resource center is to improve GCC by adding new optimization,
cleaning machine descriptions of GCC and by many other means.
Machine Independent Optimizations in GCC – Primitive algorithms have too many passes and many
repetitive works in the passes. Version by version number of passes is reduced. Goal is to implement
scalable flow dependent pointer analysis and generation of optimizer s from specifications.

Another ongoing project is to have an interprocedural data flow analysis to have optimizations across
procedure boundary. Challenge is to get the context of the called procedures while doing analysis.
Further goal is to come up with a technique to analyze large amount of codes in a very little time.

Another research problem is to solve heap reference analysis by a static analysis method where it is now
a day’s done by profiling.

Improving Instruction Selection and Machine Desciption - There are problems regarding state of the art
in this area as the designs and specification mechanisms are quite adhoc. Research interest might be to
come with ideas to solve relevant problems in this area.

Research Directions – Prof. Bernard Menezes

Prof. Bernard’s primary research interest is in security and forecasting.

Security related areas – This incorporates software related security and as well as web security. ECC is
used for security. Some interesting aspects in this field are double and add property, half and add
etc.NAF and window NAF are some of the best known techniques. Software and application security is a
race a never ending race between attacker and the defender. Their effort is to understand new efforts
and do ethical hacking. And create defenses against those attacks.

Demand Forecasting – Important part of supply chain management. Given the data upto x+h months the
problem is to predict (x+a+h) th month information. Their approach is to decompose the series and then
use multiple experts to forecast each point in the time series and after that combine the forecasts of the
corresponding points.

Theory

This is the study of abstract objects. The abstract objects model the real world objects. The abstract may
not exist in reality but the study of them reveals the truth about the objects they model. Some abstract
objects are numbers, words, theories, graphs etc.

These abstracts deal with Theorem, Proof, and Conjecture. Finding all theorems is not possible as there
can be infinitely many theories. So the 1 st task is to find the interesting ones. There are many interesting
theories like number theory, graph theory, set theory etc. Besides there are several number of
computer science theories like automata theory, complexity theory database theory etc.

Why deal with these abstract objects? Properties studied as the abstract objects can be approximated in
the real object. Some abstract objects can model many different real objects. But the real reason is that
abstract objects are more interesting than the real ones.

Prof. Ganesh Ramakrishnan

His area of interest is machine learning. He is interested in rule based systems. Some of the problems in
this area are Information extraction – extract data from unstructured text with the help of rules. For
example Word Sense Disambiguation (WSD) – identify the correct sense of the words. There are three
problems concerning rules – statistical rule mining, statistical rule consolidation and rule engines for IE.
Existing approaches in solving these have redundant computations and large overheads and includes the
overhead of rule changes. They have devised a new approach using inverted index.

Regular expressions can also be solved using inverted index and inverted index is 20-30 times faster.
Probabilistic rules can be expressed using HMM. This is an emerging area of research and Prof. Ganesh is
finding new ideas from his students in this area.

Prof. Ganesh is also interested in building a scalable framework for Annotator development (RAD).
Companies like IBM are also interested in this field.

Area of relational learning is rich but there is also a lot of opportunity of research there. Machine
learning problems generally take more time in feature construction. So a good idea would be to try to
reduce the time of feature construction.