Further Studies on Distributed Adaptation in Neuromime Networks
Report Number: AMRL TR 67-121
Author(s): Griffith, V. V., Wuerthele, W. E., Bolen, G. H.
Corporate Author(s): Goodyear Aerospace Corporation
Laboratory: Aerospace Medical Research Laboratories
Date of Publication: 1967-08
Pages: 144
Contract: AF 33(615)-1891
DoD Project: 7232
DoD Task: 723203
Identifier: AD0660848
Abstract:
This report describes investigations of networks with adaptive ability distributed through them. It is thought that large-scale adaptive systems can be constructed of adaptive building blocks. These adaptive systems would be flexible in function, reliable and would resist severe damage characteristics of living creatures. Neuron models were tested by interconnecting them into various networks to perform simple control tasks. The test results were evaluated and the evaluation used to improve the theory and the neuron model. The distributed adaption concept was analyzed from an abstract algebraic approach, using optimal control theory. The combined approach, when studied in depth, contributed to the understanding of the problem. Although the conclusions of this report are at best tentative, one conclusion seems reasonably valid: any required adaptive controller can be built using iterative elements provided only that all terminal segments of optimal trajectories of the process are themselves optimal trajectories, and that the process is controllable and observable.
Provenance: RAF Centre of Aviation Medicine
Author(s): Griffith, V. V., Wuerthele, W. E., Bolen, G. H.
Corporate Author(s): Goodyear Aerospace Corporation
Laboratory: Aerospace Medical Research Laboratories
Date of Publication: 1967-08
Pages: 144
Contract: AF 33(615)-1891
DoD Project: 7232
DoD Task: 723203
Identifier: AD0660848
Abstract:
This report describes investigations of networks with adaptive ability distributed through them. It is thought that large-scale adaptive systems can be constructed of adaptive building blocks. These adaptive systems would be flexible in function, reliable and would resist severe damage characteristics of living creatures. Neuron models were tested by interconnecting them into various networks to perform simple control tasks. The test results were evaluated and the evaluation used to improve the theory and the neuron model. The distributed adaption concept was analyzed from an abstract algebraic approach, using optimal control theory. The combined approach, when studied in depth, contributed to the understanding of the problem. Although the conclusions of this report are at best tentative, one conclusion seems reasonably valid: any required adaptive controller can be built using iterative elements provided only that all terminal segments of optimal trajectories of the process are themselves optimal trajectories, and that the process is controllable and observable.
Provenance: RAF Centre of Aviation Medicine