The performance of dummy augmentation encoding

Given the number of additional neurons, the author has estimated a lower bound of memory capacity of a bidirectional associative memory (BAM) with dummy augmentation encoding (DAE). The lower bound is min (L_A(1+r_ca)²/4 log L_A, L_B(1+r_db)²/4 log L_B), where (L_A, L_B) are the original dimensions of the library pairs (A&oarr;_k B&oarr;_k), and the numbers of additional neurons are denoted as r_caL_A and r_dbL_B (r_ca and r_db are positive constants). The extension of the author´s result on a Hopfield network with DAE is also discussed. Finally, on the basis of the lower bound, the author has evaluated the efficiency of DAE.