Linear matrix-based neural network

A novel learning algorithm for matrix-based data

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. This model can be interpreted as linear neural network with matrix inputs. You can see theory at http://pd.hacker.lt/linear_mnn.pdf
(only a draft at this time)