Embedding Observations into Higher Dimensional Spaces
This is kind of the opposite of Modeling Dynamical Systems on Manifolds. We try to embed the observations into a higher dimensional space sucht that the learning improves by utilizing more output-target comparisons.
This means we use a randomly chosen and frozen matrix with Uniformly Distributed values
to embed the original values in a high dimensional space and we learn a matrix to transform back to the original values. The values for should lie in the above interval to be in a reasonable range of the function.
