Characteristics of Hebbian-type associative memories with quantized interconnections

The effects of quantized Hebbian-type associative memories (HAM's) on storage capacity and hardware implementations are explored in this paper. The quantizations include a two-level strategy and a three-level strategy using a cutoff threshold in which the interconnection weights of both strategies are restricted to binary memory points on each synapse. The restricted interconnection weights enable a digital or an analog HAM to be implemented more space efficiently. The three-level quantized network with sparse connectivity among neurons has further simplified analog circuits in the dedicated hardwares. Results of simulation and theory show that the three-level quantized network with a properly selected cutoff threshold possesses higher storage capacity than the two-level quantized network. >