I wanted to highlight some recent work by Emmanuel Candes, Carlos Sing-Long, and myself on unbiased risk estimates for singular value thresholding (SVT). Amongst other things, the models developed in this work can be used to automatically tune SVT-based denoising models for image series (e.g., cardiac MRI). The manuscript, code, and data needed to replicate all experiments can be found at: http://www-stat.stanford.edu/~candes/SURE/index.html. If you think this work would be of interest to the Nuit Blanche community, we'd appreciate it being featured.
All the best,Thanks Josh for the heads-up:
In an increasing number of applications, it is of interest to recover an approximately low-rank data matrix from noisy observations. This paper develops an unbiased risk estimate—holding in a Gaussian model—for any spectral estimator obeying some mild regularity assumptions. In particular, we give an unbiased risk estimate formula for singular value thresholding (SVT), a popular estimation strategy which applies a soft-thresholding rule to the singular values of the noisy observations. Among other things, our formulas oﬀer a principled and automated way of selecting regularization parameters in a variety of problems. In particular, we demonstrate the utility of the unbiased risk estimation for SVT-based denoising of real clinical cardiac MRI series data. We also give new results concerning the diﬀerentiability of certain matrix-valued functions.
Liked this entry ? subscribe to Nuit Blanche's feed, there's more where that came from. You can also subscribe to Nuit Blanche by Email, explore the Big Picture in Compressive Sensing or the Matrix Factorization Jungle and join the conversations on compressive sensing, advanced matrix factorization and calibration issues on Linkedin.