Multiscale maximum penalized likelihood estimators
Robert D. Nowak, Eric D. Kolaczyk · 2003
We present a new class of maximum penalized likelihood estimators which are analogues of popular wavelet denoising methods. The new estimators move beyond the standard signal plus Gaussian noise model to handle a much broader class of nonparametric function estimation problems including Poisson and multinomial data types. The estimators share the same sort of adaptivity and near-optimality properties as wavelet denoising methods.