I am currently working as Research Liaison on IBM's Global Innovation Outlook (GIO) Team. The GIO challenges some of the brightest minds on the planet -- from the worlds of business, politics, academia, and non-profits – to collaboratively address some of the most vexing challenges on earth.
In the past I have worked on and led several innovative research projects at IBM in the area of content analysis, machine learning and bio-surveillance.
Semantic Learning, Standardization, Benchmarking and Collaborative Innovation
- MARVEL: I was the pioneering member of this technology and responsible for enabling its semantic capabilities. MARVEL won the Wall Street Journal's Technology Innovation Award in the Multimedia Category in 2004. I have been actively involved in applying the MARVEL technology to solve the content management problems for IBM's clients.
- IBM TRECVID Semantic Retrieval System: I led IBM's TRECVID Concept Detection design and implementation since the inception of the TRECVID concept detection benchmark. The system I led was based on a generic scalable learning architecture and topped performance evaluations for several years.
- Large Scale Concept Ontology for Multimedia Understanding -- LSCOM : I was the principal investigator for this DTO Challenge Workshop Series and led a diverse team of more than 50 experts across various academic, industrial and public sector research and development sites as well as experts from the intelligence community and media corporations in designing the first of its kind and the largest multimedia concept ontology with more than one thousand concepts. This is the first such effort where the design has taken into consideration the simultaneous constraints of utility, feasibility of automation and observability
- DTO Video Analysis for Content Exploitation Phase III Project -- : I am co-principal investigator of IBM's VACE Phase III project titled "An Analyst-Centric Workbench for Large Scale Cross Domain Video Intelligence." This project involves collaboration with our colleagues at the University of Illinois at Urbana Champaign and at Columbia University
- DTO Video Analysis for Content Exploitation Phase II Project -- : I was a key contributor to IBM's VACE Phase II project titled "Reconstructing and Mining of Semantic Threads Across Multiple Video Broadcast News Sources using Multi-Level Concept Modeling"
- The Multimedia Mining Adventurous Research Project is a 3 year research project within IBM. This is collaboration between members of the Pervasive Media Management Group and the Audio-visual Speech Technologies Group. The research aim is to develop methods and algorithms for generic trainable semantic concept detection. My research towards achieving this goal involves work in application of discriminant classification techniques to the modeling of semantic concepts as binary hypotheses, modeling of semantic context using factor graphical networks, reducing the amount of supervision necessary for attaining specified performance through the application of active learning, multiple instance learning and such other novel learning techniques.
- DARPA Bio-Surveillance Project: Bio-Alirt: This project deals with the early warning and advanced detection of bio-terrorist attacks and outbreaks from non-traditional heterogeneous data sources such as over the counter medicine transactions, absentee records and other such non-traditional data sources. The main requirement is to generate an alert by analyzing patterns of unusual behavior in the non-traditional data sources and the requirement is that the alert should be raised before the outbreak causes a wide-spread public health catastrophe.