Muhammad Aziz Muslim, Masumi Ishikawa, Tetsuo Furukawa
In our previous study, task segmentation was done by mnSOM, using prior information that winner modules corresponding to subsequences in the same class share the same label. Since this prior informatiom is not available in real situation, segmentation thus obtained should be regarded as the upper bound for the performance, not as a candidate for performance comparison. Present paper proposes to do task segmentation by applying various clustering methods to the resulting mnSOM, without using the above prior information. Firstly, we use the conventional hierarchical clustering. It assumes that the distances between any pair of modules are provided with precision, but this is not the case in mnSOM. Secondly, we used a clustering method based on only the distance between spatially adjacent modules with modification by their temporal contiguity. In the robotic field 1, the segmentation performance by the hierarchical clustering is very close to the upper bound for novel data. In the robotic field 2, the segmentation performance by clustering with the spatio-temporal contiguity is very close to the upper bound for novel data. Therefore, the proposed methods demonstrated their effectiveness in segmentation.
Department of Electrical Engineering, University of Brawijaya, Malang, 65145, Jl. MT Haryono 167, Indonesia; Department of Brain Science and Engineering, Kyushu Institute of Technology, Wakamatsu, Kitakyushu 808-0196, 2-4 Hibikino, Japan
