Reliability and Speed of Recall in an Associative Network

Previous investigations of the storage capacity of associative nets have not explicitly considered quantitative aspects of the tradeoff between storage capacity and reconstructive power in these systems. Furthermore, few comparisons have been made between theoretical estimates and experimental results (simulations). In this correspondence, we describe some results recently obtained and relevant to these issues. It is shown that a high storage capacity is possible, without sacrificing reliability in the recall process. Furthermore, an efficient algorithm for retrieval of the information stored is presented, and the speed of recall employing various degrees of parallelism is discussed.