Justification-based belief maintenance using neural networks
Michael Gray · 2002
An implementation of justification-based belief maintenance using a Hopfield network has been proposed for relabeling belief graphs during belief maintenance. The paper analyzes the theoretical foundation of this work and discusses the source of representational and stability problems found in this system (called the Hopfield RMS). It extends that work by analyzing the advantages of a bidirectional associative memory and shows that the BAM is preferable to the Hopfield network for implementing justification based reason maintenance in intelligent agent belief systems.