SOM computational cost reduction in ROI video ex- traction

MARTINEZ ARCA, JORGE; PASSONI LUCIA ISABEL; GONZÁLEZ, MARIELA AZUL; MONTINI PABLO

Torreon, Coahuila

Workshop; Workshop on compensatory fuzzy logic for knowledge management and decision making; 2017

Eurekas

Reducing the computational cost is of great importance in applica-tions that handle large volumes of data (Data Mining, Big Data, etc.). Severalstudies have demonstrated the ability of self-organized networks (SOM) to dis-cover data natural groupings. In cases of large data volumes, the computationalcost of SOM training increases, limiting the efficiency. The proposal of this work is to compare two process pipelines: the first comprises two steps: a first one of parallel processes that performs clustering on partitions of original data and a sec-ond step: training a SOM with the computed centroids. The second pipeline the SOM is trained completely with the original data. Efficiency and computational cost of both pipelines applied to the segmentation of images obtained from video sequences of biological dynamic patterns are analyzed. In the first pipeline the grouping on partitions used k-means algorithm; hence the decreasing of the amount of SOM training data is noticeable. As a hypothesis, we assume that the computational cost of the first pipeline decrease compared to the direct use of SOM. Modeling of the computational costs is performed using statistical parameters of experimental designs.