A novel associative network accommodating pattern deformation
In this paper we propose a novel associative network model which is able to associate a pattern with deformed versions of itself. The model is composed of a set of logical units (viewed as a set of marbles) and a set of regularly arranged physical units (viewed as a landscape). The marbles can move freely around the landscape under the influence of various kinds of forces arising from the landscape and other features of the world being modelled. This motion represents the evolution of the network. Information is embodied by the topological relationships among the marbles. When a network's evolution is initiated with an input pattern, topological relationships among marbles are observed and some aggregate features are preserved throughout the later evolution of the network. Evolution continues until all the marbles (logical units) match a set of physical units which corresponds to a local stable state of the network. Preliminary experiments show that the model works quite well for recognising topologically deformed letters.
Unable to display preview. Download preview PDF.
- 1.R. Durbin and D. Willshaw. An analogue approach to the travelling salesman problem using an elastic net method. Nature, 326:689–691, 1987.Google Scholar
- 2.Robert Hecht-Nielsen. Neurocomputing. Addison-Wesley Publishing Company, 1991.Google Scholar
- 3.John Hertz, Anders Krogh, and Richard G. Palmer. Introduction to the theory of neural computation. Addison-Wesley Publishing Company, 1991.Google Scholar
- 4.J.J. Hopfield. Neural networks and physical systems with emergent collective computational abilities. Proceedings of the National Academy of Sciences, 79:2554–2558, 1982.Google Scholar
- 5.Hui Wang. Towards a unified framework of relevance. PhD thesis, Faculty of Informatics, University of Ulster, N. Ireland, UK, October 1996. http://www.infm.ulst.ac.uk/~hwang/thesis.ps.Google Scholar
- 6.Christopher K.I. Williams. Combining deformable models and neural networks for handprinted digit recognition. PhD thesis, Dept. of Computer Science, University of Toronto, November 1994.Google Scholar