Date of Submission


Date of Award


Institute Name (Publisher)

Indian Statistical Institute

Document Type

Doctoral Thesis

Degree Name

Doctor of Philosophy

Subject Name

Computer Science


Computer Vision and Pattern Recognition Unit (CVPR-Kolkata)


Parui, Swapan Kumar (CVPR-Kolkata; ISI)

Abstract (Summary of the Work)

Abstract: Self-organizing neural network models constitute the main theme of this thesis. Some well-known self-organizing models are surveyed and their properties are discussed. The application areas on which the thesis focuses are briefly described.This thesis deals with Artificial Neural Network models, in particular, Self- organizing (unsupervisnd) models. We develop here a few self-organizing neural net- work models to solve certain problems which are well studied in the areas of Image Processing and Computationel Geometry and have wide applications in shape eztrac- tion and optimization.1.1 Artificial neural networkThe study of Biological Neural Networks originally comes under biological sciences. They deal with the brain funetions in living organisms. A lot of research work in this area have established that human brains are composed of a huge number of neurons (elementary processing elements) with massive paralel interconnections (forming a network). Such networks have learning capabilities from external stimuli (input sig- nals) and they can store, in sorme manner, what they have learned. Artificial Neural Networks are man-made (simplified) models of the biological neural networks. Such artificial models have been of great interest for some yeors in various arens lika op timization, pattern recognition, computer viston, image processing, robotica, classi fication, industrtes ete. (10, 15, 42, 43, 45, 69, 76, 86, 96, 99, 101). Artilicial neural network models ce simply neural netwurks am massively parallet inter-connections of simple computational elements (processors - also called nodes or units) that work as a collective system (we shall use these terms interchangeably). Instead of perform- ng operations sequentially, rmural network models (benenforth, by neural networek modela we shull mean artificial neural network models) are capabie of doing the same simultaneously using massively parallel networks composed of a mamber of proceisora consected by links, and thus provide high computational rates for real-time process- ing This is why neural network modela are also called comnectionist models and parallel distributed processing (PDP) models.The ability of parallel computation is one of the major motivationa of using neural network aa a tool for solving various problems in the areas mentioned abova Noural networks are albo popular due to their adaptive nature. They provide a new form of parallel computing, called neocompating as against conventional computing. in an adoptive manner as observed in living beings. Starting from an initial set of weights (usually random) neural network models can odapt or adjust the initial umghts to improve perforrmance. The adaptation is a major forus of seural netwoek research. With the help of a number of processons, rerving as the neuroos in biolopical systems, connected by hnks, the neural network can artificially work in a similar way as the brain doer. Moreover, in neural network technology, a complex problem can be solved by a number of processoen, working collectively, while each processor needs 1o do much simpler computations only. Thus simple processora can work collectively and can solve a much complex problem. Neural retwork models, by sta nature, provide robustness to some extant. Dumage to a few procasora or links may not alfect the overall performance and thus such models have sometimes higher fault :olerance. Moreover, in some caasi, nmurocomputing ia found to an efficient tool uhere conventional compating poses problema or performs poorly.


ProQuest Collection ID:

Control Number


Creative Commons License

Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.


Included in

Mathematics Commons