Clustering data by inhomogeneous chaotic map lattices

A new approach to clustering, based on the physical properties of inhomogen eous coupled chaotic maps, is presented. A chaotic map is assigned to each data point and short range couplings are introduced. The stationary regime of the system corresponds to a macroscopic attractor independent of the ini tial conditions. The mutual information between pairs of maps serves to par tition the data set in clusters, without prior assumptions about the struct ure of the underlying distribution of the data. Experiments on simulated an d real data sets show the effectiveness of the proposed algorithm.