arXiv subjects

Alexander Vinogradov

Publications and source records attributed to Alexander Vinogradov.

5 recordsLinked to original sources

arXiv2026

Ky Fan Norms and Beyond: Dual Norms and Combinations for Matrix Optimization

In this article, we explore the use of various matrix norms for optimizing functions of weight matrices, a crucial problem in deep learning. Moving beyond the spectral norm that underlies the Muon update, we leverage the duals of the Ky Fan norms to introduce the Fanion family of linear minimization oracle (LMO) algorithms, which are closely related to Muon, $ν$-SAM, and Dion. Staying inside the LMO, we construct the families of F-Fanions and S-Fanions, whose updates are convex combinations of the updates of Fanions and Normalized SGD or SignSGD, respectively. The most promising algorithms in these families are F-Muon and S-Muon. By conducting an extensive empirical study of all three algorithm families across a wide range of tasks and settings, we demonstrate that F-Muon and S-Muon consistently match Muon's performance, while outperforming Muon on a synthetic smooth convex problem.

math.OC↗

arXiv2026

Receipt Replay OOD: A Small Benchmark for Screen Replay Detection Under Domain Shift

Public datasets such as DLC-2021, SynID, and KID34K have significantly contributed to research on presentation attack detection for identity documents, including screen replay attacks. However, evaluation of out-of-domain (OOD) robustness remains insufficiently explored, especially under realistic domain shifts. In this work, we introduce Receipt Replay OOD, a small out-of-domain benchmark for screen replay detection. Receipts share several characteristics with identity documents, including planar geometry, curved corners, wear-and-tear artifacts, and text or logo patterns, while avoiding personally identifiable information constraints commonly associated with identity documents. We evaluate document replay detection models under cross-domain conditions and demonstrate the impact of domain shift on generalization performance. The dataset is publicly available.

cs.CV↗

arXiv2026

Kinetic-based macro-modeling of the solar wind at large heliocentric distances: Kappa electrons at the exobase

Recent evidence from Parker Solar Probe on the suprathermal electrons with Kappa-type velocity distributions in the outer corona has revived interest in the kinetic-based macro-modelling of the solar (SW), aiming to explain its properties. Invoked in kinetic modelling of nonequilibrium plasmas, standard Kappa distributions (SKDs) have been adjusted to the regularized Kappa distributions (RKDs) to fix the inconsistencies of SKD and develop consistent fluid modelling of space plasmas. We propose a new analysis of these properties at large heliocentric distances based on the existence of RKD electrons at the exobase. This new semi-analytic formalism is inspired by the methodology proposed initially by Meyer-Vernet and Issautier (1998), https://doi.org/10.1029/98ja02853. Compared to SKDs, the results for RKDs have extended applicability, since all moments can be defined and calculated consistently for all values of the $κ$ parameter, even lower than the critical ones (e.g., $κ_c=3/2$ imposed to the second-order moment) of SKDs. However, the excess energy of the more energetic suprathermal electrons associated with low values of $κ\lesssim 3/2$, is regulated by the RKD-specific cutoff parameter $α< 1$. The estimates for, e.g., the temperature and bulk velocity of the SW, remain at realistic values even for small $3/2 < κ\lesssim 2$, which would otherwise exceed specific observations. One can thus model a higher abundance of suprathermal electrons at the exobase (e.g., $κ\leqslant 3/2$), which is plausible for the sources of energetic events (flares and coronal mass ejections), and also in the astrospheres of stars with coronas hotter than the Sun's.

astro-ph.SR↗

arXiv2025

Can Generative Models Actually Forge Realistic Identity Documents?

Generative image models have recently shown significant progress in image realism, leading to public concerns about their potential misuse for document forgery. This paper explores whether contemporary open-source and publicly accessible diffusion-based generative models can produce identity document forgeries that could realistically bypass human or automated verification systems. We evaluate text-to-image and image-to-image generation pipelines using multiple publicly available generative model families, including Stable Diffusion, Qwen, Flux, Nano-Banana, and others. The findings indicate that while current generative models can simulate surface-level document aesthetics, they fail to reproduce structural and forensic authenticity. Consequently, the risk of generative identity document deepfakes achieving forensic-level authenticity may be overestimated, underscoring the value of collaboration between machine learning practitioners and document-forensics experts in realistic risk assessment.

cs.CV↗

arXiv2024

Embedded coherent structures from MHD to sub-ion scales in turbulent solar wind at 0.17 au

We study intermittent coherent structures in solar wind magnetic turbulence from MHD to kinetic plasma scales using Parker Solar Probe data during its first perihelion (at 0.17 au), when the satellite was in the Alfvénic slow wind of 340 km/s. The coherent structures are energetic events localized in time and covering wide range of scales. We detect them using Morlet wavelets. For the first time, we apply a multi-scale analyses in physical space to study these structures. At MHD scales within the inertial range, times scales $τ\in (1, 10^{2} )$ s, we find (i) current sheets including switchback boundaries and (ii) Alfvén vortices. Within these events, there are embedded structures at smaller scales: typically Alfvén vortices at ion scales, $τ\in (0.08, 1)$ s, and a compressible vortices at sub-ion scales, $τ\in (8,80)$ ms. The number of coherent structures grows toward smaller scales: we observe about $\sim 200$ events during 5 h time interval at MHD scales, $\sim 10^{3}$ events ai ion scales and $\sim 10^{4}$ events at sub-ion scales. In general, there are multiple structures of ion and sub-ion scales embedded within one MHD structure. There are also examples of ion and sub-ion scales structures outside of MHD structures. To quantify the relative importance of different type of structures, we do a statistical comparison of the observed structures with the expectations of models of the current sheets and vortices. This comparison is based on amplitude anisotropy of magnetic fluctuations within the structures. The results show the dominance of Alfvén vortices at all scales in contrast to the widespread view of dominance of current sheets. This means that Alfvén vortices are important building blocs of solar wind turbulence.

astro-ph.SR↗