GIF-Animation of BSOM 1

This GIF file is large (250Kbyte), so it may take long times to be loaded. Moreover, some browsers can not view GIF-animation. If you have difficulties loading the GIF file, you should press the stop button. You can use MPEG version (65Kbyte) instead.

This animation demonstrates density-learning of a Bayesian self-organizing map (BSOM). Green asterisks denote artificially generated data points. Blue circles and links denote the positions of central parameters and a topology of the BSOM respectively. The BSOM has a probability density function specified by the central parameters and a common variance. In the animation, this density is expressed using a black-red-yellow-white color map. The learning of BSOM is regarded as a search process for the optimal density explaining data generation by improving parameters and hyperparameters.


The initial values of the central parameters are positioned on a straight line whose direction is parallel with the first principal component vector of the data distribution. Starting from a density similar to the density estimated by the principal component analysis (PCA), the BSOM has been improving the density.

Akio Utsugi

Jul 19 1996 update