Mailing Address:	Department of Computer Science, Room# 3302, 10 King's College Road Toronto, ON M5S 3G4, Canada
E-MAIL:
WEB:	http://queens.db.toronto.edu/~amit/

Data quality is a serious concern in every organization that relies on data. The quality of data is commonly poor due to a multitude of reasons including, but not limited to, spelling mistakes, abbreviations, lack of standards and inconsistent notations.
SPIDER is a declarative data cleaning tool. It incorporates a set of algorithms that can be used to aid the improvement of data quality on any relational data source SPIDER can be used for flexible querying, approximate joins, schema matching and data exploration.

Built and maintained specialized BDD-based logical indices on the relational tables and described query re-write rules for efficient utilization of logical indices to quickly identify the violating relational constraint. Implemented this approach in C++ on top of Postgres database using ODBC and tested it on large collections of real and synthetic data sets.

We consider the problem of speeding up Entity Recognition systems that exploit existing large databases of structured entities to improve extraction accuracy. These systems require the computation of the maximum similarity scores of several overlapping segments of the input text with the entity database. We formulate a Batch-Top-K problem with the goal of sharing computations across overlapping segments. Our proposed algorithm performs a factor of three faster than independent Top-K queries and only a factor of two slower than an unachievable lower bound on total cost. We then propose a novel modification of the popular Viterbi algorithm for recognizing entities so as to work with easily computable bounds on match scores, thereby reducing the total inference time by a factor of eight compared to stateof- the-art methods.

Data Integration from Web-Pages, Feb 2005 - Apr 2005
Advisor: Prof. Soumen Chakrabarti, IIT Bombay

Designed a technique to extract publication entries from web-pages and storing these entries into a structured database. The creation of structured database is performed in two steps: first step identifies individual publication entry and second step performs fine grained information extraction. For the first step, we implemented a classifier on DOM nodes, while for the second step we implemented an efficient inference algorithm using A* technique.

Designed an intelligent system to automatically detect possible events of network intrusion. The system monitored network logs generated by tcpdump (per-packet activity) for anomalies and raised a flag whenever observed behavior deviated significantly from normal. We employed stide-methodology for classifying, where we used sequences of consecutive log-records (over a sliding window of fixed size) to represent activity. The basic approach is to construct a normal dictionary from data collected when there was no intrusion. This dictionary is used to compute anomaly count of incoming log-data. Stide-methodology has been previously shown to be effective in system intrusion detection problems. We proposed a novel encoding scheme for sequences in network activity log that enabled us to use same technique in this domain as well. Our results were verified by experimenting on real world datasets.

We have developed an algorithm for constructing a summary of an XML document to discover the structural aspect of its schema, and to use the summary for other tasks like - query result size estimation, structural compression and exploration. The summary is capable of preserving various kinds of quantificational information, which can be used to extract knowledge on number of edges or paths following a certain label pattern.

We designed methods and tools to assist the building of database applications to be used on mobile devices keeping in view their frequent communication breakdowns. The key idea is to maintain partial weakly consistent view of the central database on the mobile device during disconnectivity and synchronize the data when the connection is available.

Presented an in-depth study of various systems like DBXplorer, DISCOVER, DTL's DataSpot, BanKS and XRank, which enable keyword search over relational databases highlighting important features like relavence ranking and proximity search.

Presented a talk to demonstrate the use of Back Propagation Neural Networks for stock market prediction with an overview of Back Propagation NN and the design of the prediction model.

Developed a GUI with web front-end to locate different people, places and locations of various ongoing events in a region, showing the shortest path to the destination on a map.

Provided a common web based interface between instructors, students and teaching assistants in an institute for doing mundane tasks such as giving and submitting assignments, assigning projects and demo scheduling, course information, notices, grading and messages. Implemented using servlets, JDBC, SQL and Java.