An Auto-Associative Unit-Merge Network

This paper describes a new auto-associative network called a Unit-Merge Network. It is so-called because novel compound keys are used to link 2 nodes in 1 layer, with 1 node in the next layer. Unit nodes at the base store integer values that can represent binary words. The word size is critical and specific to the dataset and it also provides a first level of consistency over the input patterns. A second cohesion network then links the unit nodes list, through novel compound keys that create layers of decreasing dimension, until the top layer contains only 1 node for any pattern. Thus, a pattern can be found using a search and compare technique through the memory network. The Unit-Merge network is compared to a Hopfield network and a Sparse Distributed Memory (SDM). It is shown that the memory requirements are not unreasonable and that it has a much larger capacity than a discrete Hopfield network, for example. It can store sparse data, deal with noisy input and a complexity of O(log n) compares favourably with these networks. This is demonstrated with test results for 4 benchmark datasets. Apart from the unit size, the rest of the configuration is automatic, and its simplistic design could make it an attractive option for some applications.