NEWS & EVENTS
 A paper by Mr. Srinivasan has been accepted for International Conference on "Young Researchers Round Table on Spoken dialog system" to be held at Pittsburgh, US on September 16th, 2006.
CEN team presenting their well reviewed paper at International conference on Pattern Recognition -2006 to be held at Hong-Kong during 20-24 August.
Details of the research paper
Paper Name: "Fast single-shot Multiclass Proximal Support Vector Machines and Perceptrons”. [Download PDF Version]
Authors: Dr. Soman K.P, Loganathan R, Ajay V, Vijaya V
Significance: Support vector machines (SVM) is a very hot research topic. Companies like Microsoft, Google and Yahoo are having research teams working in this area. SVMs are basically binary classifiers used in pattern recognition task. Our research paper proposes a very simple and efficient computational approach to multiclass and hierarchical classification problems.
Conference: international conference on Pattern Recognition -2006 to be held at Hong-Kong during 20-24 August.
Workshop On “Mathematical Transform Methods in Engineering” from February 20-25, 2006 organized by Centre for Excellence in Computational Engineering and Networking (CEN). More»
AMRITA STUDENTS DEVELOP AUTOMATIC SUPER-COMPUTER CLUSTERING TOOLKIT (National Workshop on High Performance Computing) HPC 2004
 The six day long Workshop on High Performance Computing concluded on the 27th of December 2004. An Automatic supercomputing Clustering Toolkit developed by the Amrita students was released officially during the valedictory ceremony of the Workshop HPC 2004. This Toolkit would be made available as a freeware on the internet shortly.
The Centre for Excellence in Computational Engineering and Networking (CEN) & the Amrita School of Business (ASB) organized the National Level Workshop on High Performance Computing, HPC 2004 at the Amrita Vishwa Vidyapeetham, Ettimadai Campus. The workshop was co-sponsored by Bhabha Atomic Research Centre & Department of Science and Technology. It was held for 6 days, starting from December 22 to 27, 2004.
 “Dissemination of knowledge in High Performance Computing is important for the development of key technologies like nano-technology, material design, bio-informatics etc, for the economic growth of the country,” says Dr. K. P. Soman, Workshop Co-ordinator. He also said that there is a need to drive down the costs of building supercomputers so that it percolates into the universities and masses for conducting research works more effectively. To this effect, the students of CEN and ASB have been persistently working in this field. The research scholars at CEN have been working on advanced techniques in cluster computing, whereas the management students at ASB have been developing applications for HPC in business. These applications would help in solving Business problems, at both high speed and low cost.
 Supercomputing has become a major contributor to the economic competitiveness of automotive, aerospace, medical and pharmaceutical industries in the advanced countries. In this context, the Automatic Super-computing clustering toolkit, developed at CEN of the amrita university, is a landmark achievement. This would facilitate very fast building of supercomputer by clustering several ordinary Linux based machines. The key people behind this project are Mr. Ajith Peter and Mr. Rakesh Peter of Amrita Institutions. Using this toolkit, students would now be able to make supercomputers at any time they want and start learning parallel computing. They would now be able to attack massive computational problems in different domains like bioinformatics, material science etc.
The workshop included talks by various experts in the different domains that seek the power of High Performance computing. The invitees included Dr. Sekhar Majumdar (Deputy Director, CTFD Division, National Aerospace Laboratories), Dr. P. Nagabhushan (Amrita Vishwa Vidyapeetham), Dr. Balasubramaniam Sudaram (JNCASR, Dr. Sarbadhikari (Amrita Vishwa Vidyapeetham), Bangalore), Mr. Kirthiwas Neelakantan (Sun Microsystems) and Dr. Jharna Majumdar (Sc ‘G’, Aeronautical Development Establishment.
The workshop imparted training to the participants to develop parallel computing clusters, and to also equipped them to initiate research in high performance computing in their respective organizations.
Book on Wavelets By Soman K.P., Ramachandran K.I. @ PHI
The rapid growth of the theory of wavelets and their application in diverse areas-ranging from oil exploration to bioinformatics, and astrophysics-has made it imperative that engineers and scientists pursuing these areas have a working knowledge of wavelets. In the past few years, the study of wavelets and the exploration of the principles governing their behaviour have brought about sweeping changes in the disciplines of pure and applied mathematics and sciences. Keeping this in mind, the authors have specifically written this comprehensive text to fulfill the curriculum requirements of a course on wavelets. It offers an introduction to wavelet analysis, in an easy-to-understand style, without mathematical rigour. Beginning with a description of the origin of wavelets and a discussion on the recent developments, the book moves on to explain the basic concepts in Fourier series, continuous wavelet transform, discrete wavelet transform, lifting scheme and applications of wavelets in image compression, signal denoising and computer graphics. Intended primarily as a textbook for the postgraduate students of computer science, electrical/electronics and communication engineering, this book would also be useful for the practising engineers and researchers.
Massively Parallel Interior Point Method Based Linear Programming Solver from CEN which can take millions of variables and constraints.
Massively Parallel Linear Support Vector Machine based Classifier from CEN
Interior point methods (IPM)
Interior point methods are gaining popularity among researchers in the field of Operations Research for solving massively large problems. The work in the field of Interior point methods started with the work by the Indian scientist Narendra Karmarkar. In a paper presented in 1984 Karmarkar discussed an algorithm that had good theoretical convergence properties as compared to the simplex algorithm. After that, numerous papers were published in this area and the theory has been developed to a greater extent in the past few years.
The interior point algorithm, implemented more efficiently in a primal dual formulation assumes a starting point in the feasible region. The algorithm proceeds to the next point inside the feasible region by following a central path, which minimises the objective function. The feasibility of the solution after each iteration is checked using constraints imposed on duality gap, in addition to those stated in the problem. The algorithm is more efficient and faster than the simplex method. At CEN, IPM is implemented on a parallel Linux based platform. It has been applied to solve LP problems like truss design, filter design.
Applications of IPM solver
1. Engineering optimization problems in various disciplines
a. Filter design
b. Structural optimization
2. Bio informatics
3. Support vector machines for classification and regression.
Support Vector Machines (SVM)
Support Vector machines is a new learning machine for a 2-group classification problem. In this paradigm, the input vectors are mapped into a higher-dimensional feature space through an apriori chosen non linear mapping., where a linear classification decision region is constructed. This decision region when mapped back into the original feature space can take a nonlinear form. This is one of the greatest assets of SVM”s where constructing nonlinear decision regions has better classification ability than traditional classification techniques like Linear Discriminant Analysis. Furthermore, special properties of the decision surface ensure good generalization ability of this learning machine. For many 2-class problems SVM’s have been found to be more effective than other nonlinear classifiers like neural networks, k-nearest neighbor classifiers. The linear SVM is implemented in parallel cluster computing platform. Its engine is an IPM solver.
Applications:
a. Optimal control of non linear systems
b. Text identification in complex back ground
c. Intrusion detection system and I T security
d. Speech recognition systems
e. Protein fold prediction
f. Medical diagnostics
g. Marketing and finance
Parallel Implementation of interior point methods, support vector machine, PDE solver in Linux cluster computing platform with applications to protein fold prediction, Ischemic heart disease have brought in the following papers in the technical symposium, Shaastra held at IIT Madras.
|
