Operational fault tolerance of the ADAM neural network system

The author investigates the fault tolerance of the Advanced Distributed Associative Memory (ADAM), focusing on its operational use. The effect of the reliability of recall of variously configured ADAM systems is examined by injecting faults individually, and also in various combinations since correlations between them will influence their overall effect on the system. Analysis of the results indicates that fault tolerance actually arises in a similar manner to N-modular redundancy. An enhancement is made to the ADAM system which improves storage capacity and recall, and allows the results to be generally applied. It is concluded that to achieve a fault tolerant ADAM system, a large number (at least 20) of small tuple units (around three inputs) should be used. With such a configuration, an ADAM system would be very likely to provide failure-free operation for up to about 30000 h before gracefully degrading