Memory Properties of Artificial Neural Networks with Different Types of Dilutions and Damages

Memory properties of the Hopfield type neural networks with four different types of dilution of synaptic connections (dilution inside blocks, dilution outside blocks and dilution of excitory/inhibitory synapses) as well as damaging of a part of neurons, are numerically investigated. Number of stored bits per neuron an stored bits per synapse for these networks were calculated and compared. Influence of the type of dilution on the memory properties of the network is discussed. (author)