Raj Sunderraman

Professor of Computer Science. His research expertise is in the areas of Databases, Data Mining, Big Data, Knowledge Representation, Semantic Web, and Reasoning with Incomplete and Inconsistent Information. He has published more than 150 research articles in these areas in leading computer science journals and conference proceedings and is the author of a popular textbook “Oracle 10g Programming: A Primer”, published by Pearson in 2007. His current research includes scalable graph storage system for big graph data, graph pattern mining, reasoning with incomplete and inconsistent information in Description Logics and Logic Programs, and machine learning/deep learning in image classification. Sunderraman received the B.E. (Honors) Electronics Engineering degree from Birla Institute of Technology and Science, Pilani, India in 1980, the M.Tech. Computer Engineering degree from Indian Institute of Technology, Delhi, India in 1982, and the Ph.D. degree in Computer Science from Iowa State University, Ames, Iowa in 1988.

Incomplete and inconsistent databases: This research deals with methods to effectively represent and query various kinds of incompleteness and inconsistencies in relational databases. In early research, pioneering work was done to represent and query relational database with disjunctive information. Later, for the first time, a data model and relational algebra was introduced to represent and query relational databases under the open world assumption with explicit negative data. More recently, data models for degrees of exclusive disjunctions as well as generalized disjunctions in paraconsistent databases have been developed.

Negation in deductive databases and logic programming: The well-founded semantics is widely accepted as the meaning assigned to logic programs and deductive databases with arbitrary negation in the body of rules. In this research, we have used the paraconsistent data model and algebra in a bottom-up approach to compute the well-founded model. Motivated by the bottom-up computation of the well-founded model, we are currently proposing program transformation methods that transform deductive databases with arbitrary negation into well-founded model equivalent deductive databases that do not have negation. For each predicate in the input database, the transformation introduces two predicates, one to keep track of all positive consequences and the other the negative consequences of the well-founded model.

Modeling and querying graph data: In this research, we have proposed a data model and query language to represent and query graph data at the conceptual level. Graph data are increasingly found in many domains including life sciences, social networks, and digital libraries. The data model extends a traditional object-oriented data model by introducing graph-specific objects and the query language provides querying capabilities with respect to the nodes, edges, paths, and sub-graph patterns. An prototype implementation is in progress and performance evaluation is in the planning stage.

Bioinformatics/neuroinformatics: Research in bioinformatics involves the invention as well as application of database and knowledge-base technologies in the life science domain. One of the projects involves the design and implementation of a knowledge-base to catalog neuronal circuitry. NeuronBank is a web-based tool that we have developed for cataloging, searching, and analyzing neuronal circuitry within and across species. Information from a single species is represented in an individual branch of NeuronBank. Users can search within a branch or perform queries across branches to look for similarities in neuronal circuits across species. The branches allow for an extensible ontology so that additional characteristics can be added as knowledge grows. In another project, we have developed a programming environment to store, query, and manipulate protein structure data. The structure data from the Protein Data Bank (PDB) is imported into an object-oriented database; a middleware system allows life scientists to work with protein structure data without having to learn much of the computer representation of the data.