A comparative analysis-based on basins of attraction for neural associative memories

This paper presents a comparative analysis of basins of attraction for an associative memory implemented with a recurrent neural network (RNN) and other implemented with a network known as GBSB (Generalized-Brain-State-in-a-Box″). To compare the performance of both associative memories is considered the storage of patterns corresponding to a prototype example. The RNN network weights are tuned using the training algorithm for optimal margin of support vector machines (SVM). The GBSB network weights are determined using various algorithms proposed in the literature. Associative memories implemented with the RNN and GBSB undergo a performance analysis is the convergence of different initial states to each of the stored patterns.