1: \begin{abstract}

2:   This work examines under what circumstances adaptivity for truncated SVD

3:   estimation can be achieved by an early stopping rule based on the smoothed

4:   residuals \( \| ( A A^{\top} )^{\alpha / 2} ( Y - A \widehat{\mu}^{( m )})

5:   \|^{2} \). 

6:   Lower and upper bounds for the risk are derived, which show that moderate

7:   smoothing of the residuals can be used to adapt over classes of signals with

8:   varying smoothness, while oversmoothing yields suboptimal convergence rates.

9:   The range of smoothness classes for which adaptation is possible can

10:   be controlled via $ \alpha $. 

11:   The theoretical results are illustrated by Monte-Carlo simulations.

12: \end{abstract}