Nonnegative Matrix Factorization with Gaussian Process Priors
Figure 2
Underlying nonnegative
factors in the toy example: columns of (left) and rows of (right). The factors found by the LS-NMF and
the CNMF algorithms are noisy, whereas the factors
found by the GPP-NMF method are smooth. When using the correct prior, the
factors found are very similar to the true factors.