One of the key benefits of fu h.dwarey ,ple)ntations of certain ficai Neural Networks (ANNs) is their apparen 'built-in'fault'tolerance, which make thewpotentiO candidates for critical tasks with hi h reliability requirements. This pape inveitigates the fault-tolerance characteristics df time continuous, recurrent ANNs that c be used to solve optimization problems. The perform ce of these networks is first illus, ated by using ellknown model problems like the TraveIng Salesman Problem and theAssignment Prgblem. The ANNs are then subjected to up to 3 simultaneous "stuck-at-1 or "stuck-at-0" faults for network sizes of up to 900 "neurons." The effect of these faults on the performance is demonstrated and the cause for the observed fault-tolerance is discussed. An application is presented in which a network performs a critical task for a realftime distributed processing system by generating new task allocations during the reconfigpration of the system. The performance degradation of the ANN under the presence of fa4lts is investigated by largescale simulations and the potential benefits of delegating a critical task to a fault-tolerant network are discussed.

No abstract