Capacity of second-ordered bidirectional associative memory in the presence of noise bits

This paper describes error probability of the second-ordered BAM in the presence of noise bits in initial input. A calculation method is proposed to estimate the error probability at the second iteration to retrieve a library pair. Theoretical estimation suggests that the BAM with a larger number of neurons are more robust for the noise bits: the error probability of 10^-6 is obtained for the BAM with 500 neurons in a layer even the initial input has 90 noise bits (15% of numbers of neurons). Dependence of number of library pairs on number of neuron is estimated for the various error probabilities by assuming 15% noise bits. If the error rate of 10^-6 is permitted, capacity of the BAM can be increase by about 40%