DSA Faculty
API
← к списку преподавателей

Деркач Денис Александрович

Факультет компьютерных наук

Профиль на hse.ru ↗ тел.: 27266
Публикаций
226
Языков
3
Наград
6
Конференций
0
Профиль Публикации (226) Курсы (9)

Профессиональные интересы

физика высоких энергийанализ данных

Должности

  • Заведующий лабораторией — Факультет компьютерных наук, Институт искусственного интеллекта и цифровых наук, Научно-учебная лаборатория методов анализа больших данных
  • Доцент — Факультет компьютерных наук, Департамент больших данных и информационного поиска
  • Научный руководитель образовательной программы — Умные устройства: аппаратная разработка

Био

  • · Начал работать в НИУ ВШЭ в 2015 году.
  • · Научно-педагогический стаж: 7 лет.

Образование

  • 2010 · PhD: Университет Париж XI
  • 2007 · Магистратура: Санкт-Петербургский государственный университет, специальность «Физика», квалификация «Магистр»
  • 2004 · Бакалавриат: Санкт-Петербургский государственный университет, специальность «Физика», квалификация «Бакалавр»

Опыт работы

  • · 2017: н. в

Награды и поощрения

  • · Медаль "Признание - 10 лет успешной работы" НИУ ВШЭ (сентябрь 2025)
  • · Благодарность первого проректора НИУ ВШЭ (август 2024)
  • · Благодарность НИУ ВШЭ (май 2024)
  • · Почетная грамота Министерства науки и высшего образования Российской Федерации (ноябрь 2022)
  • · Благодарность первого проректора НИУ ВШЭ (август 2022)
  • · Благодарность Факультета компьютерных наук НИУ ВШЭ (август 2018)

Гранты и проекты

  • — · на соискание учёной степени кандидата наук

Идентификаторы исследователя

Публикации (226)

Global Optimisation of Black-Box Functions with Generative Models in the Wasserstein Space

2024 · CHAPTER · en

We propose a new uncertainty estimator for gradient-free optimisation of black-box simulators using deep generative surrogate models. Optimisation of these simulators is especially challenging for stochastic simulators and higher dimensions. To address these issues, we utilise a deep generative surrogate approach to model the black box response for the entire parameter space. We then leverage this knowledge to estimate the proposed uncertainty based on the Wasserstein distance - the Wasserstein uncertainty. This approach is employed in a posterior agnostic gradient-free optimisation algorithm that minimises regret over the entire parameter space. A series of tests were conducted to demonstrate that our method is more robust to the shape of both the black box function and the stochastic response of the black box than state-of-the-art methods, such as efficient global optimisation with a deep Gaussian process surrogate.

DOI ↗ PDF ↗

Symbolic expression generation via Variational Auto-Encoder

2023 · ARTICLE · en

Есть много проблем в физике, биологии и других естественных науках, в которых символическая регрессия может дать ценную информацию и открыть новые законы природы. Широко распространенные глубокие нейронные сети не предоставляют интерпретируемых решений. Между тем символические выражения дают нам четкую связь между наблюдениями и целевой переменной. Однако на данный момент нет доминирующего решения для задачи символической регрессии, и мы стремимся уменьшить этот разрыв с помощью нашего алгоритма. В этой работе мы предлагаем новую структуру глубокого обучения для генерации символьных выражений с помощью вариационного автоэнкодера (VAE). Короче говоря, мы предлагаем использовать VAE для создания математических выражений, и наша стратегия обучения заставляет сгенерированные формулы соответствовать заданному набору данных. Наша структура позволяет кодировать априорные знания о формулах в предикаты быстрой проверки, которые ускоряют процесс оптимизации. Мы сравниваем наш метод с современными тестами символьной регрессии и показываем, что наш метод превосходит конкурентов по большинству задач.

DOI ↗ PDF ↗

A full detector description using neural network driven simulation

2023 · ARTICLE · en

The abundance of data arriving in the new runs of the Large Hadron Collider creates tough requirements for the amount of necessary simulated events and thus for the speed of generating such events. Current approaches can suffer from long generation time and lack of important storage resources to preserve the simulated datasets. The development of the new fast generation techniques is thus crucial for the proper functioning of experiments. We present a novel approach to simulate LHCb detector events using generative machine learning algorithms and other statistical tools. The approaches combine the speed and flexibility of neural networks and encapsulates knowledge about the detector in the form of statistical patterns. Whenever possible, the algorithms are trained using real data, which enhances their robustness against differences between real data and simulation. We discuss particularities of neural network detector simulation implementations and corresponding systematic uncertainties.

DOI ↗ PDF ↗

Supernova Light Curves Approximation based on Neural Network Models

2023 · ARTICLE · en

Photometric data-driven classification of supernovae becomes a challenge due to the appearance of real-time processing of big data in astronomy. Recent studies have demonstrated the superior quality of solutions based on various machine learning models. These models learn to classify supernova types using their light curves as inputs. Preprocessing these curves is a crucial step that significantly affects the final quality. In this talk, we study the application of multilayer perceptron (MLP), bayesian neural network (BNN), and normalizing flows (NF) to approximate observations for a single light curve. We use these approximations as inputs for supernovae classification models and demonstrate that the proposed methods outperform the state-of-the-art based on Gaussian processes applying to the Zwicky Transient Facility Bright Transient Survey light curves. MLP demonstrates similar quality as Gaussian processes and speed increase. Normalizing Flows exceeds Gaussian processes in terms of approximation quality as well.

PDF ↗

Toward an understanding of the properties of neural network approaches for supernovae light curve approximation

2023 в печати · ARTICLE · en

The modern time-domain photometric surveys collect a lot of observations of various astronomical objects, and the coming era of large-scale surveys will provide even more information. Most of the objects have never received a spectroscopic follow-up, which is especially crucial for transients e.g. supernovae. In such cases, observed light curves could present an affordable alternative. Time series are actively used for photometric classification and characterization, such as peak and luminosity decline estimation. However, the collected time series are multidimensional, irregularly sampled, contain outliers, and do not have well-defined systematic uncertainties. Machine learning methods help extract useful information from available data in the most efficient way. We consider several light curve approximation methods based on neural networks: Multilayer Perceptrons, Bayesian Neural Networks, and Normalizing Flows, to approximate observations of a single light curve. Tests using both the simulated PLAsTiCC and real Zwicky Transient Facility data samples demonstrate that even few observations are enough to fit networks and achieve better approximation quality than other state-of-the-art methods. We show that the methods described in this work have better computational complexity and work faster than Gaussian Processes. We analyze the performance of the approximation techniques aiming to fill the gaps in the observations of the light curves, and show that the use of appropriate technique increases the accuracy of peak finding and supernova classification. In addition, the study results are organized in a Fulu Python library available on GitHub, which can be easily used by the community.

New UTfit Analysis of the Unitarity Triangle in the Cabibbo-Kobayashi-Maskawa scheme

2023 · ARTICLE · en

Flavour mixing and CP violation as measured in weak decays and mixing of neutral mesons are a fundamental tool to test the Standard Model (SM) and to search for new physics. New analyses performed at the LHC experiment open an unprecedented insight into the Cabibbo-Kobayashi-Maskawa (CKM) metrology and new evidence for rare decays. Important progress has also been achieved in theoretical calculations of several hadronic quantities with a remarkable reduction of the uncertainties. This improvement is essential since previous studies of the Unitarity Triangle did show that possible contributions from new physics, if any, must be tiny and could easily be hidden by theoretical and experimental errors. Thanks to the experimental and theoretical advances, the CKM picture provides very precise SM predictions through global analyses. We present here the results of the latest global SM analysis performed by the UTfit collaboration including all the most updated inputs from experiments, lattice QCD and phenomenological calculations.

DOI ↗

Stokes inversion techniques with neural networks: analysis of uncertainty in parameter estimation

2023 · ARTICLE · en

Magnetic fields are responsible for a multitude of Solar phenomena, including such destructive events as solar flares and coronal mass ejections, with the number of such events rising as we approach the peak of the 11-year solar cycle, in approximately 2025. High-precision spectropolarimetric observations are necessary to understand the variability of the Sun. The field of quantitative inference of magnetic field vectors and related solar atmospheric parameters from such observations has long been investigated. In recent years, very sophisticated codes for spectropolarimetric observations have been developed. Over the past two decades, neural networks have been shown to be a fast and accurate alternative to classic inversion technique methods. However, most of these codes can be used to obtain point estimates of the parameters, so ambiguities, the degeneracies, and the uncertainties of each parameter remain uncovered. In this paper, we provide end-to-end inversion codes based on the simple Milne-Eddington model of the stellar atmosphere and deep neural networks to both parameter estimation and their uncertainty intervals. The proposed framework is designed in such a way that it can be expanded and adapted to other atmospheric models or combinations of them. Additional information can also be incorporated directly into the model. It is demonstrated that the proposed architecture provides high accuracy of results, including a reliable uncertainty estimation, even in the multidimensional case. The models are tested using simulation and real data samples.

DOI ↗

Toward the end-to-end optimization of particle physics instruments with differentiable programming

2023 в печати · ARTICLE · en

The full optimization of the design and operation of instruments whose functioning relies on the interaction of radiation with matter is a super-human task, due to the large dimensionality of the space of possible choices for geometry, detection technology, materials, data-acquisition, and information-extraction techniques, and the interdependence of the related parameters. On the other hand, massive potential gains in performance over standard, “experience-driven” layouts are in principle within our reach if an objective function fully aligned with the final goals of the instrument is maximized through a systematic search of the configuration space. The stochastic nature of the involved quantum processes make the modeling of these systems an intractable problem from a classical statistics point of view, yet the construction of a fully differentiable pipeline and the use of deep learning techniques may allow the simultaneous optimization of all design parameters. In this white paper, we lay down our plans for the design of a modular and versatile modeling tool for the end-to-end optimization of complex instruments for particle physics experiments as well as industrial and medical applications that share the detection of radiation as their basic ingredient. We consider a selected set of use cases to highlight the specific needs of different applications.

DOI ↗

Observation of the B+→Jψη′K+ decay

2023 · ARTICLE · en

The B+ →Jψη′K+ decay is observed for the first time using proton-proton collision data collected by the LHCb experiment at centre-of-mass energies of 7, 8, and 13 TeV, corresponding to a total integrated luminosity of 9 fb−1. The branching fraction of this decay is measured relative to the known branching fraction of the B+ →ψ(2S)K+ decay and found to beBB+→Jψη′K+BB+→ψ2SK+=4.91±0.47±0.29±0.07×10−2

DOI ↗ PDF ↗

Latent Stochastic Differential Equations for Change Point Detection

2023 · ARTICLE · en

Automated analysis of complex systems based on multiple readouts remains a challenge. Change point detection algorithms are aimed to locating abrupt changes in the time series behaviour of a process. In this paper, we present a novel change point detection algorithm based on Latent Neural Stochastic Differential Equations (SDE). Our method learns a non-linear deep learning transformation of the process into a latent space and estimates a SDE that describes its evolution over time. The algorithm uses the likelihood ratio of the learned stochastic processes in different timestamps to find change points of the process. We demonstrate the detection capabilities and performance of our algorithm on synthetic and real-world datasets. The proposed method outperforms the state-of-the-art algorithms on the majority of our experiments.

DOI ↗ PDF ↗

Курсы (9)