Preliminary Evaluation of Batch-Learning Self-Organizing Map Algorithm on a Graphic Processor

A. Shitara, Y. Nishikawa, M. Yoshimi, T. Abe, T. Ikemura, and H. Amano (Japan)


GPU, GPGPU, CUDA, Self-organizing Map


In this paper, we introduce a GPU implementation and evaluation of batch learning self-organizing maps (BL-SOM) algorithm, which improves Kohonen’s original SOM algorithm by making input data sequence independent from learning process. We used CUDA provided by NVIDIA Corporation for parallel programming, profiling, and data flow optimization so as to exploit inherent data level parallelism of the algorithm. With various parameter combinations, implementation on GTX280 achieved 250 times higher performance compared to Intel’s Core2Quad Q6600 2.40GHz when parameters of map size, dimension of vectors, learning size and iteration of learning were 960×960, 136, 70 and 1, respectively.

Important Links:

Go Back