Neural computing on a system of parallel organized transputers: Software implementation and hardware configuration
✍ Scribed by Diethelm Würtz; Georg Hartung
- Publisher
- Elsevier Science
- Year
- 1989
- Tongue
- English
- Weight
- 883 KB
- Volume
- 56
- Category
- Article
- ISSN
- 0010-4655
No coin nor oath required. For personal study only.
✦ Synopsis
We show how a reconfigurable network of transputers can be used to serve as a fast neural computing machine. Software implementation and hardware configuration are presented. The individual updating of the neurons in the neural network is performed following the parallel synchronous or maximum field dynamic. We use an atray of 6 parallel groups of work transputers each consisting of 3 workers. Two control transputer process the incoming data from the workers and supply the master transputer with the results, which are managed and interpreted in the host environmçnt. All 21 computing elements are housed in a computing surface module named SPOT (System of Parallel Organized Transputers) developed at BBC Central Research Heidelberg.