Abstract
Purchase PDF (179 K)
E-mail Article
Add to my Quick Links
Cited By in Scopus (1)
Purchase PDF (916 K)
doi:10.1016/S0925-2312(00)00350-7 
Copyright © 2001 Elsevier Science B.V. All rights reserved.
Parallel implementation of neocognitron on star topology: Theoretical and experimental evaluation
References and further reading may be available for this article. To view references and further reading you must purchase this article.
Accepted 1 November 2000
Available online 26 July 2001.
Abstract
This paper presents the parallel implementation of the neocognitron neural network paradigm on star topology. A detailed theoretical timing analysis has been presented along with a comparison between the experimental and theoretical results. These studies demonstrate that a linear speedup can be achieved by mapping the neocognitron onto a star topology.
Author Keywords: Artificial neural networks; Neocognitron; Parallel architectures; Parallel virtual machine (PVM); Pattern recognition.
Article Outline
- 1. Introduction
- 2. Neocognitron
- 3. Implementation of the neocognitron model
- 3.1. Parallel virtual machine (PVM)
- 3.2. Mapping onto star topology
- 3.3. Testing and functioning of the neocognitron model
- 4. Theoretical analysis of parallel implementation of neocognitron model
- 4.1. Time for sequential implementation
- 4.1.1. Time taken to initialise the weights
- 4.1.2. Time to calculate the S-plane outputs
- 4.1.3. Time to calculate the C-plane output
- 4.1.4. Time to calculate the inhibitory cell output
- 4.1.5. Time to calculate the outputs of excitatory, and inhibitory synapses
- 4.2. Time for parallel implementation
- 4.3. Theoretical analysis of communication time
- 4.4. Theoretical performance study based on the silicon graphics indigo 2 machine
- 5. Results of experimental implementation of neocognitron on star topology
- 6. Conclusions
- Acknowledgements
- Appendix
- References
- Vitae
Corresponding author. Tel.: +91-80-3342451; fax: +91-80-3341683; email: lalit@micro.iisc.ernet.in
