Introduction - If you have any usage issues, please Google them yourself
This is a learning model for matrix-based data. It can be more efficient than standard vector-based models when the training sample is small or the inputs are matrices, since this model has less parameters than standard ones and exploits an inner structure of the inputs