Data clustering and visualization using cellular automata ants

Abstract

This paper presents two novel features of an emergent data visualization method coined cellular ants: unsupervised data class labeling and shape negotiation. This method merges characteristics of ant-based data clustering and cellular automata to represent complex datasets in meaningful visual clusters. Cellular ants demonstrates how a decentralized multi-agent system can autonomously detect data similarity patterns in multi-dimensional datasets and then determine the according visual cues, such as position, color and shape size, of the visual objects accordingly. Data objects are represented as individual ants placed within a fixed grid, which decide their visual attributes through a continuous iterative process of pair-wise localized negotiations with neighboring ants. The characteristics of this method are demonstrated by evaluating its performance for various benchmarking datasets. © Springer-Verlag Berlin Heidelberg 2006.