About Me

0101010101 to public static void main …

At 5 I decided to be an soldier, at 13 a cricketer, at 16 tennis player, at 18 went to Engineering college at 23 decided to do an MBA and now I am a ‘Techie’ … Too many dreams just come and fly by, only few stay.

4+ yrs of exp working for IT services company in india. Worked as a Java programmer for Investment Bank. Developed window based application and few small web based application. Other responsibilities were code review( toughest part),testing and interacting with the client at different SDLC phases. Currently I am pursing MS in Comp Sci @ USC in Los Angeles and will be graduating in May 2011. I am supporter of Open source project and deeply inclined towards new technologies like Spring,Rails and Python.  I love reading article on any stuff particularly articles related to Java world.I like exploring new technologies. Also I am avid fan of sports and personally play Cricket,badminton,table tennis…

My experience at Master Degree so far :

is simply awesome. At my coursework I have learnt a lot , so much that at times it become difficult to digest. I started my course with few of my favorite courses : Natural Language Processing ,Information Retrieval from Web ,Operating Systems and Artificial Intelligence.
Some of the Interesting projects:

Bootstrapping based approach for Transliteration :Objective of project is on the general purpose transliteration from English into Hindi. Lot of work has been done in Machine transliteration using various approaches varying from different types of modeling to learning approaches used in the system. Used noisy channel model wherein we first try to map Hindi grapheme sequence into English grapheme sequence. The language model learns bigram probabilities for Hindi grapheme.The channel model probabilities are learnt using semi-supervised approach and use Expected Maximization algorithm to learn the channel model probabilities. In this approach instead of choosing uniform prior or random prior as EM starting point, we try to bootstrap initial starting point us-ing hand-labeled data.

Finding a better ranking system for images database :The retrieval of images relevant to a query is an extremely challenging and difficult problem, because
(1) query terms can be ambiguous and
(2) the amount of tags describing the content of the image might be insufficient.
The main goal of research was to surmount the aforementioned problems and provide a better and more accurate ranking of the images searched by a user. For experimental purpose, used Flickr – a very popular web site, which contains millions of images and is used by users on a daily basis.