| Projects and Technical Reports
- Simulation of a Reconfigurable Fovea Designed
for Object Tracking - The project dealt with the
simulation of a reconfigurable fovea designed for single
object tracking. Foveal parameters like gate size, number
of perifoveal levels and initial fovea location were varied
and the system performance in terms of object tracking accuracy
versus noise was studied. The case of multiple object tracking
was also looked into and interesting observations were made.
Project Report
- Text Independent Speaker Identification
- This project implemented text independent speaker recognition
for a biometrics application. Data pre-processing using
pre-emphasis and feature extraction using mel cepstrum co-efficients
were performed. The speaker models were generated using
two methods; the Linde, Buzo, Gray Algorithm and a PCA variant
approach, both types of vector quantization for codebook
generation. The classification accuracies for the two methods
were compared and system performance was determined in terms
of the ROC curves.
Project Team - Achint Oommen Thomas and Juan Li (CSE, SUNY
- Buffalo)
Project Presentation
- Unsupervised Classification of Treatment Resistant
and Non-Resistant Cancerous Cells – The project
involved mining relevant features for the binary classification
task of treatment resistant cancerous tissue and treatment
non-resistant tissue.
Project Guide - Liana
M. Lorigo
Project Report
- Simulation of Short Term Memory and Long Term
Memory using Hopfield Networks – The objective
was to build a software system to simulate the human memory
system w.r.t. short term memory and long term memory, and
explore the interaction between them. This was achieved
using hopfield networks and hebbian learning.
Project Team - Achint Oommen Thomas, Adithya Joag, Sarvesh
Telang and Vinuchandar Krishnaswamy (CSE, SUNY - Buffalo)
Project Report
- Maximum Ball Skeleton Object Recovery
- The objective was to design an algorithm to compute maximum
ball skeletons of all objects in a given binary image. It
had to be shown that reconstruction of the exact original
image from its labeled skeletons was possible.
Project Team - Achint Oommen Thomas and Kiran
Garaga Lokeswarappa (CSE, SUNY - Buffalo)
Project Report
- Design of a Two-Class Classifier - This
project involved a classification task on a spam email dataset
that represented a two-class problem. The objective was
to build the best possible classifier to classify the datapoints
with highest accuracy on an unseen test dataset. After analysis
of the data, a feed forward neural network was designed
to perform the classification task.
Project Report
- Kernel Modification of NACHOS - The project
involved working with an educational software system, NACHOS,
which has the basic functionalities of an operating system.
Further enhancements like file system calls were added and
modifications were made to support multi-programming and
to include demand paging. The programming was based on the
Linux platform.
- Online Generation of Association Rules
– Implementation of an efficient online association
rule generation algorithm for data mining, in the Linux
environment. The implementation was based on the paper,
C.C. Aggarwal and Philip S. Yu, “A New Approach to
Online Generation of Association Rules”, in IEEE Transactions
on Knowledge and Data Engineering, Vol. 13, No. 4, pp. 527-540,
August 2001.
Project Team - Achint Oommen Thomas (CSE, SUNY - Buffalo)
and Amrinder Singh (CSE, VT)
Project Abstract
- INSTAQUEST – Question Bank Software
that generates question papers. Entered for the Intel Science
Talent Discovery Fair 2001.
- CRYPT – An Encryption Algorithm
with a novel method of generating the encrypting key from
a user-specified password. Entered for the Kishore Vaigyanik
Protsahan Yojana 2002 by Indian Institute of Science in
India.
- Developed applications in C++, Java, Visual Basic, CGI,
RDBMS systems for the Windows and Linux platforms.
|
