Growing grid

Pytorch implementation of 3-D Growing Grid (extension of Kohonen Self-Organizing Map) with Gaussian neighbourhood function and exponentially decreasing learning rate. For a description of 2-D model see: Growing Grid - a self-organizing network with constant neighborhood range and adaptation strength, Fritzke, B. Neural Process Lett (1995) 2:9. https://doi.org/10.1007/BF02332159.

Code is based on Giannis Nikolentzos' 2-D SOM implementation, who modified Sachin Joglekar's Tensorflow implementation.

As an example I use the model to perform voxel clustering of a single frame of a fMRI brain scan.

Run

jupyter notebook growing_grid.ipynb

Data for training was obtained from the OpenfMRI database. Its accession number is ds000221.