arXiv subjects

Emeric Battaglia

Publications and source records attributed to Emeric Battaglia.

4 recordsLinked to original sources

arXiv2026

Quantile Randomized Kaczmarz for Streaming Linear Systems with Massart Noise

Quantile randomized Kaczmarz (QRK) has proven to be an efficient solver for corrupted linear systems and has received much attention. It was recently shown by Cai et al. (SIAM J. Matrix Anal. Appl. 47(2):802-823, 2026) that using $O(\log T/\log(1/β))$ samples for computing the quantile is necessary and sufficient for QRK to converge linearly over $T$ iterations when solving linear systems with a $β$-fraction of arbitrary corruptions, as long as $β$ is small enough. However, it remains unclear how large the corruption level $β$ can be, and how to compute the required subsample size $D$ explicitly, without hidden constants. This paper studies streaming linear systems with Massart noise via QRK using an order-optimal batch size $D=O(\log T)$ in each update. The independence of samples from previous iterations in the streaming setting enables a sharper analysis, yielding explicit, computable bounds on both the tolerable corruption level and the required subsample size. In particular, we establish linear convergence for corruption levels of up to approximately 7%. We also discuss how the constants improve under oblivious noise.

math.NA↗

arXiv2026

Complex Stochastic Gradient Descent and Directional Bias in Reproducing Kernel Hilbert Spaces

Stochastic Gradient Descent (SGD) is a known stochastic iterative method popular for large-scale convex optimization problems due to its simple implementation and scalability. Some objectives, such as those found in complex-valued neural networks, benefit from updates like in SGD and Gradient Descent (GD) with a newly defined ``gradient'' that allows for complex parameters. This complex variant of the SGD/GD methods has already been proposed, but convergence guarantees without analyticity constraints have not yet been provided. We propose a variant of SGD (complex SGD) that allows for complex parameters, and we provide convergence guarantees under assumptions that parallel those from the real setting. Notably, these results extend to GD as well, and with the same set of assumptions, we confirm that some directional bias results extend from the real to the complex setting for kernel regression problems. We provide empirical results demonstrating the efficacy of the complex SGD in kernel regression problems utilizing complex reproducing kernel Hilbert spaces. In particular, we demonstrate we may recover superoscillation functions and Blaschke products from the Fock Space and Hardy Space, respectively, as the optimal functions for a particular choice of a loss function.

cs.LG↗

arXiv2025

Quantile-RK and Double Quantile-RK Error Horizon Analysis

In solving linear systems of equations of the form $Ax=b$, corruptions present in $b$ affect stochastic iterative algorithms' ability to reach the true solution $x^\ast$ to the uncorrupted linear system. The randomized Kaczmarz method converges in expectation to $x^\ast$ up to an error horizon dependent on the conditioning of $A$ and the supremum norm of the corruption in $b$. To avoid this error horizon in the sparse corruption setting, previous works have proposed quantile-based adaptations that make iterative methods robust. Our work first establishes a new convergence rate for the quantile-based random Kaczmarz (qRK) and double quantile-based random Kaczmarz (dqRK) methods, which, under certain conditions, improves upon known bounds. We further consider the more practical setting in which the vector $b$ includes both non-sparse ``noise" and sparse ``corruption". Error horizon bounds for qRK and dqRK are derived and shown to produce a smaller error horizon compared to their non-quantile-based counterparts, further demonstrating the advantages of quantile-based methods.

math.NA↗

arXiv2024

Reverse Quantile-RK and its Application to Quantile-RK

When solving linear systems $Ax=b$, $A$ and $b$ are given, but the measurements $b$ often contain corruptions. Inspired by recent work on the quantile-randomized Kaczmarz method, we propose an acceleration of the randomized Kaczmarz method using quantile information. We show that the proposed acceleration converges faster than the randomized Kaczmarz algorithm. In addition, we show that our proposed approach can be used in conjunction with the quantile-randomized Kaczamrz algorithm, without adding additional computational complexity, to produce both a fast and robust iterative method for solving large, sparsely corrupted linear systems. Our extensive experimental results support the use of the revised algorithm.

math.NA↗