Hash-based Associative Memory-like Pattern Classifier

A pattern classifier named Hash-based Associative Memory-like Pattern Classifier (HAPC) based on hashing technique, associative memory, and human deep-thinking logic is proposed in this paper to solve pattern classification problems without complicated mathematical computation. In the hashing layer of the designed system, one of the similarity-preserve hashing is used to convert the input samples to hashed data which dramatically reduces the dimensionality of the data but keeps the similarity information. In the clustering layer, hashed samples are divided into different clusters, which can be considered a feature extraction process based on the similarity of the hash value. In the associate memory layer, each cluster will be trained as an associate memory to store the hashed data stably. In the logical layer, Bayesian inference is used to perform the decision-making process to classify the testing sample without complicated mathematical computation. In the simulation, we are able to achieve a good result, especially on small numbers of training data.