1: \begin{abstract}

2: In this paper, we consider the framework of privacy amplification via iteration, which is originally proposed by Feldman et~al.~and subsequently simplified by Asoodeh et~al.~in their analysis via the contraction coefficient.

3: This line of work focuses on the study of the privacy guarantees obtained by the projected noisy stochastic gradient descent (PNSGD) algorithm with hidden intermediate updates.

4: A limitation in the existing literature is that only the early stopped PNSGD has been studied, while no result has been proved on the more widely-used PNSGD applied on a shuffled dataset.

5: Moreover, no scheme has been yet proposed regarding how to decrease the injected noise when new data are received in an online fashion.

6: In this work, we first prove a privacy guarantee for shuffled PNSGD, which is investigated asymptotically when the noise is fixed for each sample size $n$ but reduced at a predetermined rate when $n$ increases, in order to achieve the convergence  of privacy loss. We then analyze the online setting and provide a faster decaying scheme for the magnitude of the injected noise that also guarantees the convergence of privacy loss.

7: \end{abstract}

8: