Frame-constrained total variation regularization for white noise regression (Q820796)

This research introduces a family of estimators in the Gaussian white noise model, obtained by minimization of the bounded variation seminorm under a constraint on the frame coefficients of the residuals. The proposed estimates attain the minimax optimal rate of convergence in any dimension up to logarithmic factors. One extension investigated in the present research is the utilization of the proposed model to the nonparametric regression setting with discretely sampled data. The results of research are in principle extendable to inverse problems, non-Gaussian noise models and stochastic different equations based models.