arXiv subjects

Ekaterina Dmitrieva

Publications and source records attributed to Ekaterina Dmitrieva.

7 recordsLinked to original sources

arXiv2026

Resonant production of millicharged scalars in $k^2>0$ electromagnetic wave background

We investigate the solution of the Klein-Gordon equation for a charged scalar particle in an electromagnetic plane wave background with $k^2>0$, which can be realized in a medium with a refractive index $n<1$. We reduce the equation of motion to the Mathieu equation, which has resonant solutions that grow exponentially for certain parameter ranges. Physically, this effect leads to anomalous losses for certain beams of electromagnetic fields in a given medium. We set the constraints on the coupling and mass of millicharged particles from the absence of such anomalous losses for optical laser and radio beams.

hep-ph↗

arXiv2026

Sona Technical Report

We introduce Sona, a single-model generative recommender for Yandex Music. In an online A/B test, Sona replaced the entire production cascade, comprising more than 15 candidate generators followed by pre-ranking and ranking models that consume hundreds of features, including signals from large transformer models such as Argus and target-attention scorers, while significantly improving key engagement metrics. The architecture of Sona unifies candidate generation and ranking around a shared user representation. Its encoder transforms the user's chronological sequence of logged engagement events into hidden states consumed by both the autoregressive decoder and the Ranking Module. The next-token-prediction and distillation objectives jointly update the encoder, coupling generation and ranking through the same user state. Neither Sona nor its Teacher Ranker uses hand-engineered features; both operate on logged event fields and learned item representations. In the final Sona configuration, the larger teacher supplies ranking targets during training but is absent from serving, leaving the encoder, decoder, and Ranking Module as a single deployed model. We evaluate Sona in an online A/B experiment using live traffic from My Vibe on smart speakers, one of Yandex Music's largest recommendation surfaces. Relative to the production control, Sona produced statistically significant uplifts of 4.53% in Active Users, the primary metric, 6.30% in Total Listening Time, and 11.42% in Likes. These effects were incremental to improvements retained from preceding deployments. The Active Users uplift was 2.35 times the increment previously delivered by Argus, the strongest model deployed on this surface before Sona. These results show that a single jointly trained model can replace a mature multi-stage recommendation cascade while improving recommendation quality on live traffic.

cs.IR↗

arXiv2026

High-Field EPR/ENDOR of N/Be Centers for Defect Engineering in 6H-SiC

Silicon carbide (SiC) in its various structural modifications is widely used in power semiconductor electronics, operating under extreme conditions of high temperature, high voltage, and intense radiation. The discovery of spin defects (S>0) with unique optical and coherent properties has further positioned SiC as a promising platform for quantum technologies. Here, we investigate a 6H-SiC single crystal co-doped with nitrogen and beryllium at concentrations of 1018 cm-3, using continuous-wave and pulsed electron paramagnetic resonance (EPR) and electron-nuclear double resonance (ENDOR). To enhance spectral resolution, experiments were conducted in the W-band (94 GHz; B = 3.4 T). Pulsed EPR identified nitrogen donors and beryllium acceptors in various lattice positions, allowing for the determination of their phase coherence and spin-lattice relaxation times. ENDOR measurements elucidated the electron-nuclear interactions with the local silicon and carbon environment, including distant coordination spheres. The observed hyperfine structures indicated highly delocalized spin density within the supercell. The TRIPLE resonance spectra verify coupled nuclear spin subspaces from different coordination spheres due to defect spin density. These results demonstrate the feasibility of incorporating dual impurities with distinct functional roles while preserving the crystal lattice`s structural features.

cond-mat.mtrl-sci↗

arXiv2026

Constraints on millicharged particles from thunderstorms on the Solar system planets

We investigate the production of millicharged particles (mCPs) by the Schwinger mechanism in thunderstorms in the atmospheres of different planets in the Solar system. We consider a thundercloud as a giant capacitor that can be discharged in two ways: either by lightnings or by mCP production. Taking into account the observation of lightning strikes, we establish the constraints on the charge and mass of mCPs. We examine two types of cloud configurations: a simple arrangement of two clouds, and a more complex layered structure that gives rise to potential wells. In the latter case, we take into account the effects of Bose enhancement for scalar mCPs, and Pauli blocking for fermionic ones. We use the observational data of planetary atmospheres obtained by satellite missions to establish constraints on the charge and mass of mCP particles. The best constraints came from the observation of thunderstorms in Saturn's atmosphere under an assumption of layered cloud structure: $q > 10^{-24}$ for bosonic mCPs. These constraints for bosons are, to the best of our knowledge, the best in the literature.

hep-ph↗

arXiv2025

HiFi-Stream: Streaming Speech Enhancement with Generative Adversarial Networks

Speech Enhancement techniques have become core technologies in mobile devices and voice software. Still, modern deep learning solutions often require high amount of computational resources what makes their usage on low-resource devices challenging. We present HiFi-Stream, an optimized version of recently published HiFi++ model. Our experiments demonstrate that HiFi-Stream saves most of the qualities of the original model despite its size and computational complexity improved in comparison to the original HiFi++ making it one of the smallest and fastest models available. The model is evaluated in streaming setting where it demonstrates its superior performance in comparison to modern baselines.

cs.SD↗

arXiv2025

Tachyonic and parametric resonances for massive particle production in an intense plane wave background

We investigate the stability of an intensive plane wave of a massless or light field $ϕ$ in a trilinear scalar model $gϕχ^2$ due to the resonant production of massive $χ$ particles in a perturbatively forbidden regime. We apply two methods: first, we solve the Heisenberg equation for the quantum amplitudes of the field $χ$ in an external plane wave background, generalizing the solution of A.Arza. Second, for the light but massive $ϕ$ we perform the relativistic boost to the rest frame of $ϕ$, reducing the problem to the stability of the thoroughly investigated massive condensate. It turns out that the stability properties are significantly different for the cases of massless and light fields. In the first case, one should adopt the Heisenberg equation approach, since the alternative method cannot provide a comprehensive outcome. In the second case, the use of the Mathieu equation provides a more accurate solution, while for the massless case, this approach is not applicable.

hep-ph↗

arXiv2021

A Differentiable Language Model Adversarial Attack on Text Classifiers

Robustness of huge Transformer-based models for natural language processing is an important issue due to their capabilities and wide adoption. One way to understand and improve robustness of these models is an exploration of an adversarial attack scenario: check if a small perturbation of an input can fool a model. Due to the discrete nature of textual data, gradient-based adversarial methods, widely used in computer vision, are not applicable per~se. The standard strategy to overcome this issue is to develop token-level transformations, which do not take the whole sentence into account. In this paper, we propose a new black-box sentence-level attack. Our method fine-tunes a pre-trained language model to generate adversarial examples. A proposed differentiable loss function depends on a substitute classifier score and an approximate edit distance computed via a deep learning model. We show that the proposed attack outperforms competitors on a diverse set of NLP problems for both computed metrics and human evaluation. Moreover, due to the usage of the fine-tuned language model, the generated adversarial examples are hard to detect, thus current models are not robust. Hence, it is difficult to defend from the proposed attack, which is not the case for other attacks.

cs.CL↗