R&D Projects - Rajendra Akerkar

Major Funding Agencies:

TWAS,
UNESCO,
Industrial (Multinational) Funding,
Department of Science & Tech., Government of India,
NBHM, Department of Atomic Energy, Government of India.

Completed Projects:

Project Title: ICT based Semantic Interoperability

Coordinator: Rajendra Akerkar

Type & Funding: Research Project (2009 - 2010)

To develop and evaluate new methods for development and evolution of the Semantic Web methodologies and tools for sustainable tourism. We will investigate Knowledge technologies that can be used to improve the accuracy of searches and obtain a superior matching between customer expectations and tourism products. Also we will examine how to accomplish semantic interoperability between web sources and services.

Project Title: Knowledge Discovery to explore correlations between Mind Conditions and Physical Elements of Human beings.

Coordinators: Rajendra Akerkar & Manish Joshi

Type & Funding: Research Project (2007 - 2009)

Man is not simply a machine,as Descartes suggested more than 300 years ago. It seems we have come full circle in our beliefs about the mind - body relationship. Body and mind clearly interact and in complex ways that we do not yet fully understand. This project aims at establishing correlation between patient symptoms and probable disturbances in patient’s mind using models of knowledge discovery in databases.

Project Title: Adaptive HTML Analysis to Automatically Wrapper Generation.

Coordinators: Rajendra Akerkar, David Camacho and María Dolores Rodríguez-Moreno

Type & Funding: Research Project /Submitted

In this project, we propose an approach, based on Artificial Intelligence techniques, to extract information from HTML pages and to add automatically semantic tags to them (extracted from a given ontology). Our project will use both, an adaptive interaction with the user, and several machine learning algorithms to adapt the wrapper generation process to the necessities of the users. The integration of both, Information Extraction and Software Agents technologies, to automatically generate adaptive wrappers could be an important issue in the evolution of new algorithms (and their related applications). These algorithms could be able to aid manage the current complexity of the Web (designed for humans not for software programs), and to alleviate the problem of extract (and manage) semantic data from the current Web systems. To study how these techniques can be adapted to the current Semantic Web-based technologies.

Project Title: Instructional Designs for Distance Education: Neural Net Enabled Distance Learning Model.

Project Coordinators: Prof. Rajendra Akerkar

Type & Funding: Industrial (2007 - 2009)

The unique “one-to-one” nature of distance learning offers a unique opportunity to tailor the instructional process for the benefit of the student as well as the organization. Neural Net Technology offers an “add-on” possibility for the computer/teacher to learn about the student, and adjust the course content as well as the pace to best suit the student. This will result in educating the student rather than issuing a pass/fail grade based on a process designed to fit all. Knowledge represented in the human brain as the strength of a very large number of interconnections between a very large number of nodes called neurons. Based on actual experience, the human brain adjusts the interconnection strength to adapt to new environments. In case of distance learning, this would entail the administrative software program adjusting the course content, through an on-going comparison of the student’s performance against a set of established “thresholds”.

Project Title: Machine Learning for Data Mining in Medicine.

Coordinators: Rajendra Akerkar

Type & Funding: Industrial Project (2003 - 2007)

The central idea of this research work was: Can useful and new knowledge be discovered from large collection of medical data by means of Data Mining, in particular by the method of machine learning from examples with artificial neural nets and case based reasoning? In this research phase, we have tested machine learning approaches used in mining of medical data, distinguishing between symbolic and sub-symbolic data mining methods and applications of these methods in medicine have been considered.

Project Title: Intelligent Natural Language Interface.

Coordinators: Rajendra Akerkar and Manish Joshi

Type: Research Project (Partially Completed in July 2007)

Natural Language Interface helps user to interact with computer in non-formal manner by answering various questions posed by user. Key-word matching based paradigm generate answers, however these answers often affected by problems caused by some language Dependant phenomena like semantic symmetry and ambiguous modification. Current methods described in literature tackles these problems using in depth parsing whereas we have formulated rules to tackle linguistic phenomena using shallow parsing. We have developed an Intelligent Natural Language Interface (ENLIGHT) comprising of newly formulated shallow parsing based algorithms in conjunction with some AI techniques to train the system. We have then tested results obtained from the ENLIGHT system. As compare to other systems which requires in depth parsing for extracting correct answer ENLIGHT system is more effective and efficient. Results are compared using different metrics like precision, mean reciprocal rank, response time and ENLIGHT system show better performance.

Our algorithms will be implemented for search engine technology in the next phase.

Project Title: A Knowledge Representation and Reasoning System

Coordinators: Rajendra Akerkar, D. Rajesh Duthie

Type: Research Project (Completed in December 2005)

The general task of work in knowledge representation and reasoning (KRR) is, trivially, the representation of knowledge. The kind of knowledge for which the representation task has been undertaken is secondary to the description of the overall task as an exercise in knowledge representation. Thus, the task of representing and using mathematical, commonsense, visual language based, and logic knowledge is "lumped" together under the KRR. We would like to argue that, in the task of natural language processing and understanding, language syntax, semantics, and pragmatics imposes constraints on any computational formalism. In this project, we present KRR system for the representation of knowledge associated with natural language dialog. We say that this may done with minimal loss of inferential power and will result in and enriched representation language capable of supporting complex natural language description, some discourse phenomena, standard first-order inference, inheritance and terminological sub-assumptions. The aim of the new logic and its implementation in a KRR system are as follows: A study of some characteristics of natural language that a KRR system should support, a KRR system to present a formalization of a propositional semantic network based knowledge representation and reasoning system that addressed our goals for NLP, a KRR system in which mapping from natural language sentences into logical form sentences be as direct as possible. The work also include other related aspects of the above mentioned problems.

Project Title: A Computational Model for Knowledge Intensive Learning

Coordinators: Rajendra Akerkar, Rajeshree Belekar

Type: Research Project (Completed in December 2007)

The framework represents a perspective on knowledge, reasoning and learning that aims at satisfying three fundamental system requirements. All requirements are based on the assumption that competent and robust systems need a thorough, relatively deep knowledge model as a fundament for understanding, i.e. as a background pool of knowledge from which plausible explanations are generated to support reasoning steps and learning decisions concerning the more shallow, associational knowledge in past cases and heuristic rules. The system developed under this project works for majority type of adult diabetic patients.