ASPHD_CSC Archives

PhD Student


Options: Use Forum View

Use Monospaced Font
Show Text Part by Default
Condense Mail Headers

Message: [<< First] [< Prev] [Next >] [Last >>]
Topic: [<< First] [< Prev] [Next >] [Last >>]
Author: [<< First] [< Prev] [Next >] [Last >>]

Print Reply
PhD Student <[log in to unmask]>
Mon, 11 Oct 2010 09:22:54 -0400
PhD Student <[log in to unmask]>
text/plain; charset=windows-1252; format=flowed
Tammie Dudley <[log in to unmask]>
text/plain (59 lines)
Dear all,

Departmental Colloquium (10/15/2010 Friday 1:30 pm, Department 
Conference Room)

Dr. Nagabhushan N. Tagadur
Professor and Chairman
Department of Information Science and Engineering
S.J. College of Engineering (affiliated with Visveshwaraya Technological 

Radial Basis Function (RBF) neural networks have emerged as an 
alternative to multilayer proceptron neural networks for classification 
and function approximation. There has been a greater interest among 
researchers to evolve RBF architectures dynamically during the learning 
phase for a given application . Many significant works have appeared in 
the literature on building efficient auto-configuring RBF networks 
(Moody and Darken 1989, Platt 1991, Fritzke 1995, Chitra 2002, Wallace 
2005, and many more). While synthesizing an RBF network during learning, 
there is a need to look into certain objectives such as:

* The network must be compact.
* The network have good generalization.
* The network must learn fast.

This talk addresses some novel ideas such as:

* Positioning the RBF kernels at an optimal location in the input 
pattern space
* Pruning for any redundant neurons
* Computation of the width of RBF kernels based on edge information
* Training the network with only ”significant patterns”

while achieving the above objectives. Experiments have been conducted on 
five benchmark data sets from the UCI machine learning repository and 
the results are compared.

About the Speaker: Dr. Nagabhushan N. Tagadur holds masters and doctoral 
degrees from the Indian Institute of Science, Bangalore, India. He has 
been working with S.J. College of engineering, Mysore, India, since 
1984. He was a visiting professor at Korea University, Seoul, in 2007 
and Hannan University, Japan, in 2004. He has supervised five doctoral 
theses and has over 40 publications in conferences and journals. He has 
delivered a number of invited talks. Dr. Nagabhushan also served as the 
head of the e-learning center of Visveswvaraya Technological University 
(VTU) from January 2007 to August 2010. He is a reviewer for many 
journals and has chaired many conferences.

Dr. Nagabhushan’s research areas include dynamic learning algorithms for 
neural networks and the Industrial application of neural networks for 
automobiles and in communication networks. He is a member of the 
Institution of Engineers (India), the Indian Society for Technical 
Education, and the Computer Society of India. He has a research 
collaboration with mobile giant Nokia on local language support for 
N900-series devices.