I want to use four matrices as predictors, with a single matrix as an output. All values in the matrices are real floating points. I've tried using element (i, j) of each predictor matrix to predict the (i, j) element of the output matrix, but (although I can get it to run) this won't work because I expect each element of the output matrix to be at least somewhat correlated to every value in all four predictor matrices. I used the Regression Learner app within the Statistics and Machine Learning Toolbox 11.6 to do this.
I've also looked into image recognition a little, but it seems that although it can handle 2D data, it works as a classifier, whereas I want a matrix of floats.