... alternately.

The procedure studied in [7] can be seen as the zero-noise limit of the EM algorithm for a probabilistic PCA model.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... entries

We make the typical assumption that values are missing at random, that is, the missingness does not depend on the unobserved data. An example where the assumption does not hold is when out-of-scale measurements are marked missing.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... all

It could be filled by finding a value that maximizes the determinant of the covariance matrix (and thus the entropy of the underlying Gaussian distribution).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...

The PCA approach has been considered by other Netflix contestants as well (see, e.g., [15,16]).

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.