A cross-validation procedure for stopping the EM algorithm and deconvolution of neutron depth profiling spectra

The iterative EM (expectation maximization) algorithm is used to deconvolve neutron depth profiling spectra. Because of statistical noise in the data, artifacts in the estimated particle emission rate profile appear after too many iterations of the EM algorithm is stopped using a cross-validation procedure. The data are split into two independent halves. The EM algorithm is applied to one half of the data to get an estimate of the emission rates. The algorithm is stopped when the conditional likelihood of the other half of the data passes through its maximum. The roles of the two halves of the data are then switched to get a second estimate of the emission rates. The two estimates are then averaged.<
>