Title :
On the capacity of a Markov-chain encoded associative memory
Author :
Shirazi, Mehdi N.
Author_Institution :
Common. Res. Lab., Minist. of Posts & Telecommun., Kobe, Japan
Abstract :
The author proposes a Markov-chain encoded associative memory as a general autocorrelation associative memory which can embrace hitherto-proposed encodings such as the conventional (biased or unbiased Bernoulli trial) and sparse encodings if the chain´s parameters are chosen appropriately. The proposed associative memory is a network of n fully interconnected two-state formal neurons. One chooses m realizations (patterns) of a Markov chain at random (independently), and then stores them in the network by adjusting the network synaptic matrix according to the Hebbian rule. A general condition which must be satisfied by the number of the stored patterns m is derived. Then, the capacity of the network is given as the solution of an optimization problem
Keywords :
Markov processes; content-addressable storage; encoding; matrix algebra; optimisation; Bernoulli trial; Hebbian rule; Markov-chain encoded associative memory; autocorrelation associative memory; neural nets; optimization; storage capacity; synaptic matrix; two-state formal neurons; Associative memory; Biological information theory; Codes; Computer simulation; Crosstalk; Encoding; Neurons; Random variables; Sparse matrices; Stability;
Conference_Titel :
Neural Networks, 1991., IJCNN-91-Seattle International Joint Conference on
Conference_Location :
Seattle, WA
Print_ISBN :
0-7803-0164-1
DOI :
10.1109/IJCNN.1991.155353