DSA Faculty
API
← к списку преподавателей

Никитин Илья Сергеевич

Факультет компьютерных наук

Профиль на hse.ru ↗ тел.: 23303 | +79304095166
Публикаций
4
Языков
1
Наград
2
Конференций
0
Профиль Публикации (4) Курсы (6)

Профессиональные интересы

машинное обучение

Должности

  • Преподаватель — Факультет компьютерных наук, Департамент больших данных и информационного поиска
  • Преподаватель — Факультет физики
  • Академический руководитель образовательной программы — Машинное обучение в цифровом продукте

Био

  • · Начал работать в НИУ ВШЭ в 2023 году.
  • · Научно-педагогический стаж: 2 года.

Образование

  • 2025 · Магистратура: Национальный исследовательский университет "Высшая школа экономики", специальность «Физика», квалификация «Магистр»
  • 2023 · Бакалавриат: Национальный исследовательский университет "Высшая школа экономики", специальность «Физика», квалификация «Бакалавр»

Опыт работы

  • · 2025: : приглашённый преподаватель НИУ ВШЭ

Награды и поощрения

  • · Благодарность департамента больших данных и информационного поиска НИУ ВШЭ (февраль 2026)
  • · Надбавка за публикацию в журнале из Списка А (и приравненном к нему научном издании) (2025–2026)

Идентификаторы исследователя

Публикации (4)

Deep-learning-based Identification of Solar Magnetic Tornadoes and Their Spatial Properties during Solar Minimum and Maximum

2026 · ARTICLE · en

Solar magnetic tornadoes are dynamic, spiral-shaped plasma structures characterized by helical magnetic fields and rotating plasma flows in the solar atmosphere. They play a significant role in the transport of energy and mass within the solar environment. Identifying and analyzing solar magnetic tornadoes is challenging due to their transient nature and complex morphology and the large volume of associated observational data. We propose two automated methods for detecting these magnetoplasma structures using modern deep learning techniques. Our models search for twisted prominences in the solar corona visible at the solar limb. Our approach involves analyzing the Solar Dynamics Observatory Atmospheric Imaging Assembly 171 Å images using convolutional and recurrent neural networks. By applying established techniques, the methods proposed can detect previously unknown magnetic tornadoes alongside those documented in the literature. The models are trained on 10,294 instances, which corresponds to detection of ∼100 tornadoes with high precision and recall. Identification of 1,476,885 new instances is performed. The resulting database allows for the first comparative analysis of magnetic tornadoes’ spatial distributions across solar cycle phases. We find that tornadoes can serve as tracers of environments prone to reconnection. During solar minimum, these structures occur at the boundaries of coronal holes with strong current sheets and at the edges of polar conic current sheets. During solar maximum, they appear at the footpoints of magnetic loops and are associated with polarity inversion lines.

DOI ↗

Mean Pairwise Distances in Rouse Polymer Subject to Fast Loop Extrusion

2025 · ARTICLE · en

We consider a model of a Rouse polymer extended by the mechanism of active loop extrusion. The model is based on a kinetic equation that is valid provided that the extrusion rate is high enough and the resulting loop ensemble is sufficiently sparse. Within the one-loop approximation of diagrammatic calculations, a semi-analytical method for determining the mean square physical distance between a pair of chain beads as a function of the contour distance between them is developed. The model is based on a kinetic equation that is valid provided that the extrusion rate is high enough and the resulting loop ensemble is sufficiently sparse. Within the framework of the one-loop approximation of diagrammatic calculations, a semi-analytical method for determining the mean square of the physical distance between a pair of chain sections as a function of the contour distance between them is developed. The mean square of the physical distance and its logarithmic derivative as functions of the contour separation are plotted for different values of the equilibrium degree. The results are compared with the case of frozen disorder of sparse loops.

DOI ↗ PDF ↗

Constructing efficient strategies for the process optimization by restart

2024 · ARTICLE · en

Optimization of the mean completion time of random processes by restart is a subject of active theoretical research in statistical physics and has long found practical application in computer science. Meanwhile, one of the key issues remains largely unsolved: how to construct a restart strategy for a process whose detailed statistics are unknown to ensure that the expected completion time will reduce? Addressing this query here we propose several constructive criteria for the effectiveness of various protocols of noninstantaneous restart in the mean completion time problem and in the success probability problem. Being expressed in terms of a small number of easily estimated statistical characteristics of the original process (MAD, median completion time, low-order statistical moments of completion time), these criteria allow informed restart decision based on partial information.

DOI ↗ PDF ↗

Inferring Parameters and Reconstruction of Two-dimensional Turbulent Flows with Physics-informed Neural Networks

2024 · ARTICLE · en

Obtaining system parameters and reconstructing the full flow state from limited velocity observations using conventional fluid dynamics solvers can be prohibitively expensive. Here we employ machine learning algorithms to overcome the challenge. As an example, we consider a moderately turbulent fluid flow, excited by a stationary force and described by a two-dimensional Navier–Stokes equation with linear bottom friction. Using dense in time, spatially sparse and probably noisy velocity data, we reconstruct the spatially dense velocity field, infer the pressure and driving force up to a harmonic function and its gradient, respectively, and determine the unknown fluid viscosity and friction coefficient. Both the root-mean-square errors of the reconstructions and their energy spectra are addressed.We study the dependence of these metrics on the degree of sparsity and noise in the velocity measurements. Our approach involves training a physics-informed neural network by minimizing the loss function, which penalizes deviations from the provided data and violations of the governing equations. The suggested technique extracts additional information from velocity measurements, potentially enhancing the capabilities of particle image/tracking velocimetry.

DOI ↗ PDF ↗

Курсы (6)