A Robust Neural Gas algorithm for clustering analysis

In this paper, we present a novel robust neural gas (RNG) algorithm. While retaining the essence of the original neural gas (NG) algorithm, the RNG algorithm effectively tackles the robustness problems associated with NG and it present variants, such as sensitivity to input sequence ordering and presence of many outliers. In addition, through combining the competitive Hebbian learning strategy and minimal description length framework, our new algorithm can establish the topology relationships among the prototypes and ensure that all prototypes can represent a meaningful region in the data set. Our algorithm has shown encouraging experimental results compared with 9 prototype based clustering algorithms and their robust variants in static clustering tasks with the fixed number of prototypes.