Estimating the inverse nonlinear functions

To estimate the inverse nonlinearities in [13], Theis and Amari consider that there should be a linear relationship between the same component of different clusters. Here, we exploit that there should be a linear relationship between the different components of the same cluster. Both approaches can lead to a reliable estimation of the inverse nonlinearities, but the latter allows to operate directly on the available data. By doing so it avoids the interpolation needed in [13] that can be problematic in cases of strong nonlinearities or in the presence of noise.