Searcharxiv⌕ Search

arXiv subjects

Sofiia Shvaiko

Publications and source records attributed to Sofiia Shvaiko.

2 recordsLinked to original sources

Beyond the error horizon: averaged quantile Kaczmarz methods for corrupted and noisy linear systems

Linear systems and least-squares problems can contain measurement errors of different scales: a few measurements may be highly unreliable, while small noise affects every measurement. Residual-based iterative filtering can remove the influence of sparse corruption, but its accuracy is limited by a noise-dependent error horizon. We study how averaging and relaxation reduce this horizon in averaged quantile randomized Kaczmarz (AQRK), which filters equations by residual quantiles and averages a batch of single-row updates. Filtering, however, introduces an update bias that averaging alone cannot remove. We show that re-querying the accepted equations can remove this bias, and when the corruption is sparse enough relative to the conditioning of the matrix and large enough relative to the noise, we prove that AQRK with re-queried residuals reduces the horizon as averaging does without the corruption: its error decays to a horizon proportional to $\sqrt{α/s}$, where $α$ is the stepsize and $s$ the update batch size. Further, decreasing stepsizes or increasing batches gives convergence to the solution of the clean system. Numerical experiments support the theory and examine settings beyond it, including noise that cannot be re-queried and the corrupted low-rank tensor fitting problem. In the latter, AQRK inner solves succeed where standard alternating least-squares solvers and single-row robust updates fail.

math.NA↗

Quantile Randomized Kaczmarz Algorithm with Whitelist Trust Mechanism

Randomized Kaczmarz (RK) is a simple and fast solver for consistent overdetermined systems, but it is known to be fragile under noise. We study overdetermined $m\times n$ linear systems with a sparse set of corrupted equations, $ {\bf A}{\bf x}^\star = {\bf b}, $where only $\tilde{\bf b} = {\bf b} + \boldsymbol{\varepsilon}$ is observed with $\|\boldsymbol{\varepsilon}\|_0 \le βm$. The recently introduced QuantileRK (QRK) algorithm addresses this issue by testing residuals against a quantile threshold, but computing a per-iteration quantile across many rows is costly. In this work we (i) reanalyze QRK and show that its convergence rate improves monotonically as the corruption fraction $β$ decreases; (ii) propose a simple online detector that flags and removes unreliable rows, which reduces the effective $β$ and speeds up convergence; and (iii) make the method practical by estimating quantiles from a small random subsample of rows, preserving robustness while lowering the per-iteration cost. Simulations on imaging and synthetic data demonstrate the efficiency of the proposed method.

math.NA↗