Deterministic walks as an algorithm of pattern recognition.

New tools for automatically finding data clusters that share statistical properties in a heterogeneous data set are imperative in pattern recognition research. Here we introduce a deterministic procedure as a tool for pattern recognition in a hierarchical way. The algorithm finds attractors of mutually close points based on the neighborhood ranking. A memory parameter mu acts as a hierarchy parameter, in which the clusters are identified. The final result of the method is a general tree that represents the nesting structure of the data in an invariant way by scale transformation.