Abstract
We construct a pattern recognition system by modeling the structure of the visual cortex. The complexities of the visual cortex can be simplified by understanding that the neurons of this region are distinguished by the features of input image that each neuron detects. We propose a neural network(NN) model of the simple structure based on the function and structure of the visual cortex. Moreover, a lot of ideas of manufactured products with NN were proposed. One of issues for productization is uncertainty of the behavior of nonlinearity of NN. Accordingly, it is important to analyze the internal representation of NN. In this paper, we discuss the recognition and training mechanism of a NN model by use of Alopex algorithm. Alopex algorithm, which is an iterative and stochastic processing to minimize or maximize a cost function. processing to minimize or maximize a cost function. By this method, the receptive fields of the units in the output layer are obtained.
We have proposed a four-layered feed-forward NN model for pattern recognition and analyzed the recognition mechanism as well as the performance of the model. We proposed a modified Alopex algorithm and calculated the receptive fields of the output unit. In the case of simple training character set, the receptive field changes according to the values of initial weight vectors. If the initial values are large, NN uses small amounts of input values for the classification. In contrast, if the initial values are small, NN uses whole input image. Moreover, it was seen that as the complexity of the set of training patterns increased the receptive field of the output unit changed. Smaller initial values of weight vector have advantage to get the more features. Alopex algorithm is an effective method to find the characteristics of images.
Keywords
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Preview
Unable to display preview. Download preview PDF.
References
K. Fukushima “Cognitron, A self-organizing multilayered neural network”. Biol. Cybern, Vol. 20, pp.121–136, 1975
K. Fukushima “Neocognitron: A Self-organizing Neural Network Model for Mecanism of Pattern Recognition Unaffected by Shift in Position,” Biol. Cybern, 36, pp.193–202, 1980
B. D. Ripley: Pattern recognition and neural networks, Cambridge University Press, Cambridge, 1996
D.E. Rumelhart., J.L. McClelland and the PDP Research Group “Parallel distributed processing.” Vol. 1, Cambridge: The MIT Press, 1986
Tadashi Sugihara, Shimon Edelman and Keiji Tanaka: Representation of objective similarity among three-dimensional shapes in the monkey, Biol. Cybern.,78, pp.1–7, 1998
Eucaly Kobatake, Gang Wang and Keiji Tanaka: Effects of Shape-Discrimination Training on the Selectivity of Inferotemporal Cells in Adult Monkeys, The Journal of Neurophysiology, vol. 80, pp.324–330, 1998
E. Tzanakou, R. Michalak, and E. Harth “The Alopex process: visual receptive fields by response feedback,” Biol. Cybern., 35, pp.161–174, 1979
Author information
Authors and Affiliations
Corresponding author
Editor information
Rights and permissions
Copyright information
© 2007 International Federation for Medical and Biological Engineering
About this paper
Cite this paper
Shintani, H., Akutagawa, M., Nagashino, H., Pandya, A.S., Kinouchi, Y. (2007). Analysis of Multi-Layer Neural Network’s Recognition Mechanism Using Alopex Algorithm. In: Magjarevic, R., Nagel, J.H. (eds) World Congress on Medical Physics and Biomedical Engineering 2006. IFMBE Proceedings, vol 14. Springer, Berlin, Heidelberg. https://doi.org/10.1007/978-3-540-36841-0_41
Download citation
DOI: https://doi.org/10.1007/978-3-540-36841-0_41
Publisher Name: Springer, Berlin, Heidelberg
Print ISBN: 978-3-540-36839-7
Online ISBN: 978-3-540-36841-0
eBook Packages: EngineeringEngineering (R0)