Oleg Bakhteev | Publications

Main publications

2023

ICAART
Deep Learning Model Selection With Parametric Complexity Control

Olga Grebenkova, Oleg Bakhteev, Vadim Strijov

2023

Abs HTML PDF Bibtex

The paper is devoted to deep learning model complexity. It is estimated by Bayesian inference and based on a computational budget. The idea of the proposed method is to represent deep learning model parameters in the form of hypernetwork output. A hypernetwork is a supplementary model which generates parameters of the selected model. This paper considers the minimum description length from a Bayesian point of view. We introduce prior distributions of deep learning model parameters to control the model complexity. The paper analyzes and compares three types of regularization to define the parameter distribution. It infers and generalizes the model evidence as a criterion that depends on the required model complexity. Finally, it analyzes this method in the computational experiments on the Wine, MNIST, and CIFAR-10 datasets.
```
@conference{icaart23,
  abbr = {ICAART},
  author = {Grebenkova, Olga and Bakhteev, Oleg and Strijov, Vadim},
  html = {https://www.scitepress.org/PublicationsDetail.aspx?ID=lRGAknY5+Ow=&t=1},
  title = {Deep Learning Model Selection With Parametric Complexity Control},
  booktitle = {Proceedings of the 15th International Conference on Agents and Artificial Intelligence - Volume 2: ICAART,},
  year = {2023},
  pages = {65-74},
  publisher = {SciTePress},
  organization = {INSTICC},
  pdf = {https://www.scitepress.org/Papers/2023/116269/116269.pdf},
  doi = {10.5220/0011626900003393},
  isbn = {978-989-758-623-1},
  issn = {2184-433X}
}
```

2022

AIST
Neural architecture search with structure complexity control

Konstantin Yakovlev, Olga Grebenkova, Oleg Bakhteev, Vadim Strijov

In Recent Trends in Analysis of Images, Social Networks and Texts, 2022

Abs HTML PDF Bibtex

The paper investigates the problem of deep learning model selection. We propose a method of a neural architecture search with respect to the desired model complexity called DARTS-CC. An amount of parameters in the model is considered as a model complexity. The proposed method is based on a differential architecture search algorithm (DARTS). Instead of optimizing structural parameters of the architecture, we consider them as a function depending on the complexity parameter. It enables us to obtain multiple architectures at one optimization procedure and select the architecture based on our computation budget. To evaluate the performance of the proposed algorithm, we conduct experiments on the Fashion-MNIST and CIFAR-10 datasets and compare the resulting architecture with architectures obtained by other neural architecture search methods.
```
@inproceedings{yakovlev2021neural,
  abbr = {AIST},
  title = {Neural architecture search with structure complexity control},
  author = {Yakovlev, Konstantin and Grebenkova, Olga and Bakhteev, Oleg and Strijov, Vadim},
  booktitle = {Recent Trends in Analysis of Images, Social Networks and Texts},
  year = {2022},
  pages = {207--219},
  pdf = {https://yahootechpulse.easychair.org/publications/preprint_download/H5MC},
  html = {https://link.springer.com/book/10.1007/978-3-031-15168-2}
}
```

RFBR

Digital Education Platform [In Russian]

Цифровая платформа образования

Oleg Bakhteev et al.

Russian Foundation for Basic Research Journal, 2022

HTML Bibtex

@article{бахтеев2022цифровая,
  abbr_en = {RFBR},
  abbr = {РФФИ},
  title_en = {Digital Education Platform},
  title = {Цифровая платформа образования},
  author = {Бахтеев, Олег и др.},
  author_en = {Oleg Bakhteev et al.},
  journal = {Вестник Российского фонда фундаментальных исследований},
  journal_en = {Russian Foundation for Basic Research Journal},
  volume = {113},
  number = {1},
  pages = {87--104},
  year = {2022},
  html = {https://www.rfbr.ru/rffi/ru/bulletin/o_2128693},
  sitetag = {publ}
}

AiT
Gradient methods of metaparameter optimization in knowledge distillation

Maria Gorphinich, Oleg Bakhteev, Vadim Strijov

Automation and Remote Control, 2022

Abs Bibtex

The paper investigates the problem of knowledge distillation in deep learning models. Knowledge distillation is considered as a meta-parameter optimization problem in which information from a model of a more complex structure, a teacher model, is transferred to a model of a simpler structure, called a student model. The paper proposes a generalization of the distillation problem for the case of optimization of metaparameters by gradient methods. Metaparameters are the parameters of the distillation optimization problem. The loss function for such a task is a sum of the classification term and the cross-entropy between the logits of the student model and the teacher model. Assigning optimal metaparameters to the distillation loss function is a computationally difficult task. The properties of the optimization problem are investigated in order to predict the trajectory of metaparameters optimization. An analysis of the trajectory of the gradient optimization of metaparameters is carried out and their values are predicted using linear approximation. The proposed approach is illustrated using a computational experiment on CIFAR-10 and Fashion-MNIST datasets, as well as on synthetic data.
```
@article{gorphinich2022gradient,
  abbr_en = {AiT},
  abbr = {АиТ},
  bilingual = {True},
  title_en = {Gradient methods of metaparameter optimization in knowledge distillation},
  title = {Градиентные методы оптимизации метапараметров в задаче дистилляции знаний},
  author_en = {Maria Gorphinich and Oleg Bakhteev and Vadim Strijov},
  author = {М., Горпинич and Бахтеев, О. Ю. and Стрижов, В. В.},
  journal_en = {Automation and Remote Control},
  journal = {Автоматика и телемеханика},
  number = {10},
  pages = {67--79},
  year = {2022}
}
```
ENAI
Cross-language plagiarism detection: a case study of European universities academic works

Oleg Bakhteev et al.

2022

Abs HTML Supp Bibtex

The chapter investigates the problem of cross-lingual plagiarism in academic works of European universities. Although the possibly massive problem of incorrect text reuse, most text reuse detection systems generally focus only on the monolingual plagiarism text reuse: when both the analysed document and source of text reuse are written in one language. In this chapter, we analyse a more difficult setting: when the languages of the analysed document and reused language are different. For this problem solution, we present a system of cross-lingual text reuse detection. The system composes the methods of statistical machine translation and deep learning methods based on the contextualized word embeddings, such as BERT and its multilingual version, LaBSE. To analyse the efficiency of the proposed method, we conduct experiments both on the synthetic dataset generated using machine translation systems and on the real dataset of academic graduation theses. We experimented on the collection of 10202 documents and found 103 documents with a significant amount of cross-lingual text reuse. Although these results are preliminary and should be verified further, they confirm the massiveness of this problem in academic science.
```
@conference{bakhteev2022enai,
  abbr = {ENAI},
  title = {Cross-language plagiarism detection: a case study of European universities academic works},
  author = {Bakhteev, Oleg et al.},
  author_en = {Oleg Bakhteev et al.},
  booktitle = {Academic Integrity: Broadening Practices, Technologies, and the Role of Students},
  pages = {143--161},
  year = {2022},
  publisher = {Springer},
  supp = {https://academicintegrity.eu/conference/proceedings/2021/book_of_abstracts2021.pdf},
  html = {https://link.springer.com/chapter/10.1007/978-3-031-16976-2_9}
}
```
ISP RAS
Anti-Distillation: Knowledge Transfer from a Simple Model to the Complex One

Kseniia Petrushina, Oleg Bakhteev, Andrey Grabovoy, Vadim Strijov

In 2022 Ivannikov Ispras Open Conference (ISPRAS), 2022

Abs HTML Bibtex

The paper considers the problem of adapting the model to new data with a large amount of information. We propose to build a more complex model using the parameters of a simple one. We take into account not only the accuracy of the prediction on the original samples but also the adaptability to new data and the robustness of the obtained solution. The work is devoted to developing the method that allows adapting the pre-trained model to a more heterogeneous dataset. In the computational experiment, we analyse the quality of predictions and model robustness on Fashion-MNIST dataset.
```
@inproceedings{petrushina2022isp,
  abbr = {ISP RAS},
  html = {https://ieeexplore.ieee.org/document/10076855},
  author = {Petrushina, Kseniia and Bakhteev, Oleg and Grabovoy, Andrey and Strijov, Vadim},
  booktitle = {2022 Ivannikov Ispras Open Conference (ISPRAS)},
  title = {Anti-Distillation: Knowledge Transfer from a Simple Model to the Complex One},
  year = {2022},
  volume = {},
  number = {},
  pages = {65-69},
  doi = {10.1109/ISPRAS57371.2022.10076855}
}
```

2021

IiEP
Methods of cross-lingual text reuse detection in large textual collections [In Russian]

Методы обнаружения переводных заимствований в больших текстовых коллекциях

Rita Kuznetsova, Oleg Bakhteev, Yury Chekhovich

Informatics and Applications, 2021

Abs HTML PDF Bibtex

The paper investigates the cross-lingual text reuse detection problem. The paper proposes a monolingualapproach to this problem: to translate the suspicious document into the language of the collection for the furthermonolingual analysis. One of the major requirements for theproposed method is robustness to the machinetranslation ambiguity. The further document analysis is divided into two steps. At the first step, the authors retrievedocuments-candidates which are likely to be the source of the text reuse. For the robustness, the authors proposeto retrieve the documents using word clusters that are constructed using distributional semantics. At the secondstep, the authors compare the suspicious document with candidates using sentence embeddings that are obtainedby deep learning neural networks. The experiment was conducted for the “English–Russian” language pair bothon the synthetic data and on the articles included in the Russian Science Citation Index.
```
@article{кузнецова2021методы,
  abbr_en = {IiEP},
  abbr = {ИИП},
  title_en = {Methods of cross-lingual text reuse detection in large textual collections},
  title = {Методы обнаружения переводных заимствований в больших текстовых коллекциях},
  author_en = {Rita Kuznetsova and Oleg Bakhteev and Yury Chekhovich},
  author = {Кузнецова, Р. В. and Бахтеев, О. Ю. and Чехович, Ю. В.},
  journal = {Информатика и её применения},
  journal_en = {Informatics and Applications},
  volume = {15},
  number = {1},
  pages = {30--41},
  year = {2021},
  html = {http://www.ipiran.ru/journal/issues/article/19922264210105.html},
  publisher = {Российская академия наук, Отделение нанотехнологий и информационных технологий},
  pdf = {http://www.ipiran.ru/journal/issues/2021_15_01/Vol15_Issue1.pdf},
  sitetag = {publ}
}
```
IiEP
Variational deep learning model optimization with complexity control [In Russian]

Вариационная оптимизация модели глубокого обучения с контролем сложности

Olga Grebenkova, Oleg Bakhteev, Vadim Strijov

Informatics and Applications, 2021

Abs HTML PDF Bibtex

This paper investigates the problem of deep learning model optimization. The authors propose a methodto control model complexity. The minimum description length is interpreted as the complexity of the model. It actsas the minimal amount of information that is required to transfer information about the model and the dataset. Theproposed method is based on representation of a deep learning model. The authors propose a form of a hypernetusing the Bayesian inference. A hypernet is a model that generates parameters of an optimal model. The authorsintroduce probabilistic assumptions about the distribution of parameters of the deep learning model. The papersuggests maximizing the evidence lower bound of the Bayesian model validity. The authors consider the evidencebound as a conditional value that depends on the required model complexity. The authors analyze this method incomputational experiments on the MNIST dataset.
```
@article{гребенькова2021вариационная,
  abbr_en = {IiEP},
  abbr = {ИИП},
  title_en = {Variational deep learning model optimization with complexity control},
  title = {Вариационная оптимизация модели глубокого обучения с контролем сложности},
  author_en = {Olga Grebenkova and Oleg Bakhteev and Vadim Strijov},
  author = {Гребенькова, О. С. and Бахтеев, О. Ю. and Стрижов, В. В.},
  journal = {Информатика и её применения},
  journal_en = {Informatics and Applications},
  volume = {15},
  number = {1},
  pages = {42--49},
  year = {2021},
  html = {http://www.ipiran.ru/journal/issues/article/19922264210106.html},
  publisher = {Российская академия наук, Отделение нанотехнологий и информационных технологий},
  pdf = {http://www.ipiran.ru/journal/issues/2021_15_01/Vol15_Issue1.pdf},
  sitetag = {publ}
}
```
Dialogue
Near-duplicate handwritten document detection without text recognition

Oleg Bakhteev et al.

In Computational Linguistics and Intellectual Technologies, 2021

Abs PDF Supp Bibtex

The paper presents a novel method for near-duplicate detection in handwritten document collections of schoolessays. A large amount of online resources with available academic essays currently makes it possible to cheatand reuse them during high school final exams. Despite the importance of the problem, at the moment there isno automatic method for near-duplicate detection for handwritten documents, such as school essays. The schoolessay is represented as a sequence of scanned images of handwritten essay text. Despite advances in recognitionof handwritten printed text, the use of these methods for the current task is a challenge. The proposed method ofnear-duplicate detection does not require detailed markup text, which makes it possible to use it in a large numberof tasks related to the information extraction in zero-shot regime, i.e. without any specific resources written in theprocessed language. The paper presents a method based on series analysis. The image is segmented into words.The text is characterized by a sequence of features, which are invariant to the author’s writing style: normalizedlengths of the segmented words. These features can be used for both handwritten and machine-readable texts. Thecomputational experiment is conducted on IAM dataset of English handwritten texts and the dataset of real imagesof handwritten school essays.
```
@inproceedings{bakhteev2021handwritten,
  abbr = {Диалог-21},
  abbr_en = {Dialogue},
  title = {Near-duplicate handwritten document detection without text recognition},
  author = {Bakhteev, Oleg et al.},
  author_en = {Oleg Bakhteev et al.},
  booktitle = {Computational Linguistics and Intellectual Technologies},
  year = {2021},
  pdf = {http://www.dialog-21.ru/media/5237/bakhteevoplusetal028.pdf},
  supp = {http://bit.ly/ap_handwritten}
}
```
PAN
Hate Speech Spreader Detection using Contextualized Word Embeddings

Evgeny Finogeev et al.

In CLEF (Working Notes), 2021

Abs HTML PDF Bibtex

The paper presents a method of hate speech spreaders recognition developed for the task of “Profiling Hate Speech Spreaders on Twitter” at the PAN@CLEF Conference 2021. Hate speech is an increasing problem nowadays. Due to the spread of the internet and the rise of social media, people with hostile views towards certain groups post hateful messages on social media resources. In this paper, we present a model to detect hate speech spreaders based on their Twitter posts. We aggregate contextualized embeddings of single tweets to form a vector representation for every user and employ classification methods to find users spreading hate speech. We analyze different embedding models based on BERT architecture for the problem. The submitted model achieves 67 pecent in terms of accuracy for the English part of the dataset and 83 percent for the Spanish part of the dataset
```
@inproceedings{finogeev2021author,
  abbr = {PAN},
  title = {Hate Speech Spreader Detection using Contextualized Word Embeddings},
  author_en = {Evgeny Finogeev et al.},
  author = {Finogeev, Evgeny et al.},
  booktitle = {CLEF (Working Notes)},
  year = {2021},
  pdf = {http://ceur-ws.org/Vol-2936/paper-167.pdf},
  html = {https://pan.webis.de/publications.html}
}
```
ISP RAS
The automatic approach for scientific papers dating

Andrey Grabovoy, Oleg Bakhteev, Yury Chekhovich

In Ivannikov Ispras Open Conference (ISPRAS), 2021

Abs HTML Supp Bibtex

The paper investigates automatic methods of document date estimation. The document date is essential metadata for many applied tasks, such as event detection or internet search. Methods for estimating the dates of newspaper articles are well known. In this paper, we propose to predict the publishing year for scientific papers. We present a new dataset based on open access articles from eLIBRARY.RU. It contains articles written in Russian and English. The article introduces a method of dating documents based on the raw document text content. The computational experiment compares different solutions for document date estimation. We also analyze how prediction depends on the topic for different methods.
```
@inproceedings{grabovoy2021automatic,
  abbr = {ISP RAS},
  title = {The automatic approach for scientific papers dating},
  author = {Grabovoy, Andrey and Bakhteev, Oleg and Chekhovich, Yury},
  booktitle = {Ivannikov Ispras Open Conference (ISPRAS)},
  year = {2021},
  supp = {https://www.youtube.com/watch?v=cOM82kHRwkM},
  html = {https://ieeexplore.ieee.org/abstract/document/9708686}
}
```

2020

IiEP
Ordering the set of neural network parameters [In Russian]

Введение отношения порядка на множестве параметров аппроксимирующих моделей

Andrey Grabovoy, Oleg Bakhteev, Vadim Strijov

Informatics and Applications, 2020

Abs PDF Bibtex

This paper investigates a method for setting order on a set of the model parameters. It considers linear models and neural networks. The set is ordered by the covariance matrix of the gradients. It is proposed to use a given order to freeze the model parameters during the optimization procedure. It is assumed that, after few iterations of the optimization algorithm, most of the model parameters can be frozen without significant loss of the model quality. It reduces the dimensionality of the optimization problem. This method is analyzed in the computational experiment on the real data. The proposed order is compared with the random order on the set of the model parameters.
```
@article{грабовой2020введение,
  abbr_en = {IiEP},
  abbr = {ИИП},
  title_en = {Ordering the set of neural network parameters},
  title = {Введение отношения порядка на множестве параметров аппроксимирующих моделей},
  author_en = {Andrey Grabovoy and Oleg Bakhteev and Vadim Strijov},
  author = {Грабовой, А. В. and Бахтеев, О. Ю. and Стрижов, В. В.},
  journal = {Информатика и её применения},
  journal_en = {Informatics and Applications},
  volume = {14},
  number = {2},
  pages = {58--65},
  year = {2020},
  publisher = {Российская академия наук, Отделение нанотехнологий и информационных технологий},
  pdf = {http://www.mathnet.ru/links/54db4af77e607146a4babf4c801c1fc0/ia662.pdf},
  sitetag = {publ}
}
```
PAN
Fake news spreader detection using neural tweet aggregation

Oleg Bakhteev, Aleksandr Ogaltsov, Petr Ostroukhov

In CLEF (Working Notes), 2020

Abs HTML PDF Bibtex

The paper describes the neural networks-based approach for ProfilingFake News Spreaders on Twitter task at PAN 2020. The problem is reduced to thebinary classification with a set of tweets of the user as an object to classify andclass labels corresponding to users that are likely to spread fake news and ordinaryusers. To deal with a set of tweets we employ two neural network architectures:either based on recurrent or convolutional neural networks. We try aggregate thewhole information obtained from the tweets to decide whether the user can spreadfake news or not. We also present an ensemble of models that consists of a neuralnetwork and a classification model that works on each tweet separately.
```
@inproceedings{bakhteev2020fake,
  abbr = {PAN},
  title = {Fake news spreader detection using neural tweet aggregation},
  author = {Bakhteev, Oleg and Ogaltsov, Aleksandr and Ostroukhov, Petr},
  booktitle = {CLEF (Working Notes)},
  year = {2020},
  pdf = {https://pan.webis.de/downloads/publications/papers/bakhteev_2020.pdf},
  html = {https://pan.webis.de/publications.html}
}
```

2019

IiEP
Estimation of the relevance of the neural network parameters [In Russian]

Определение релевантности параметров нейросети

Andrey Grabovoy, Oleg Bakhteev, Vadim Strijov

Informatics and Applications, 2019

Abs PDF Bibtex

The paper investigates a method for optimizing the structure of a neural network. It is assumed that the number of neural network parameters can be reduced without significant loss of quality and without significant increase in the variance of the loss function. The paper proposes a method for automatic estimation of the relevance of parameters to prune a neural network. This method analyzes the covariance matrix of the posteriori distribution of the model parameters and removes the least relevant and multicorrelate parameters. It uses the Belsly method to search for multicorrelation in the neural network. The proposed method was tested on the Boston Housing data set, the Wine data set, and synthetic data.
```
@article{грабовой2019определение,
  abbr_en = {IiEP},
  abbr = {ИИП},
  title_en = {Estimation of the relevance of the neural network parameters},
  title = {Определение релевантности параметров нейросети},
  author_en = {Andrey Grabovoy and Oleg Bakhteev and Vadim Strijov},
  author = {Грабовой, Андрей Валериевич and Бахтеев, Олег Юрьевич and Стрижов, Вадим Викторович},
  journal = {Информатика и её применения},
  journal_en = {Informatics and Applications},
  volume = {13},
  number = {2},
  pages = {62--70},
  year = {2019},
  publisher = {Российская академия наук, Отделение нанотехнологий и информационных технологий},
  pdf = {http://www.mathnet.ru/links/c0d38d0bc4d68901f3ca135fd3050ea6/ia594.pdf}
}
```
AOR
Comprehensive analysis of gradient-based hyperparameter optimization algorithms

Oleg Bakhteev, Vadim Strijov

Annals of Operations Research, 2019

Abs HTML Bibtex

The paper investigates hyperparameter optimization problem. Hyperparameters are the parameters of model parameter distribution. The adequate choice of hyperparameter values prevents model overfit and allows it to obtain higher predictive performance. Neural network models with large amount of hyperparameters are analyzed. The hyperparameter optimization for models is computationally expensive. The paper proposes modifications of various gradient-based methods to simultaneously optimize many hyperparameters. The paper compares the experiment results with the random search. The main impact of the paper is hyperparameter optimization algorithms analysis for the models with high amount of parameters. To select precise and stable models the authors suggest to use two model selection criteria: cross-validation and evidence lower bound. The experiments show that the models optimized using the evidence lower bound give higher error rate than the models obtained using cross-validation. These models also show greater stability when data is noisy. The evidence lower bound usage is preferable when the model tends to overfit or when the cross-validation is computationally expensive. The algorithms are evaluated on regression and classification datasets.
```
@article{bakhteev2019comprehensive,
  abbr = {AOR},
  title = {Comprehensive analysis of gradient-based hyperparameter optimization algorithms},
  author_en = {Oleg Bakhteev and Vadim Strijov},
  author = {Bakhteev, Oleg and Strijov, Vadim},
  journal = {Annals of Operations Research},
  pages = {1--15},
  year = {2019},
  publisher = {Springer},
  html = {https://link.springer.com/article/10.1007%2Fs10479-019-03286-z}
}
```

2018

AiT
Deep Learning Model Selection of Suboptimal Complexity

Oleg Bakhteev, Vadim Strijov

Automation and Remote Control, 2018

Abs HTML Bibtex

We consider the problem of model selection for deep learning models of suboptimal complexity. The complexity of a model is understood as the minimum description length of the combination of the sample and the classification or regression model. Suboptimal complexity is understood as an approximate estimate of the minimum description length, obtained with Bayesian inference and variational methods. We introduce probabilistic assumptions about the distribution of parameters. Based on Bayesian inference, we propose the likelihood function of the model. To obtain an estimate for the likelihood, we apply variational methods with gradient optimization algorithms. We perform a computational experiment on several samples.
```
@article{бахтеев2018выбор,
  bilingual = {True},
  abbr_en = {AiT},
  abbr = {АиТ},
  title_en = {Deep Learning Model Selection of Suboptimal Complexity},
  title = {Выбор моделей глубокого обучения субоптимальной сложности},
  author_en = {Oleg Bakhteev and Vadim Strijov},
  author = {Бахтеев, О. Ю. and Стрижов, В. В.},
  journal_en = {Automation and Remote Control},
  journal = {Автоматика и телемеханика},
  number = {8},
  pages = {129--147},
  year = {2018},
  publisher = {Российская академия наук, Отделение энергетики, машиностроения, механики и~…},
  html_en = {https://link.springer.com/article/10.1134/S000511791808009X},
  html = {http://www.mathnet.ru/php/archive.phtml?wshow=paper&jrnid=at&paperid=14742&option_lang=rus}
}
```
IiEP
Optimal recurrent neural network model in paraphrase detection [In Russian]

Выбор оптимальной модели рекуррентной сети в задачах поиска парафраза

Anton Smerdov, Oleg Bakhteev, Vadim Strijov

Informatics and Applications, 2018

Abs PDF Bibtex

This paper addresses the problem of optimal recurrent neural network selection. It asserts the neural network evidence lower bound as the optimal criterion for selection. It investigates variational inference methods to approximate the posterior distribution of the network parameters. As a particular case, the normal distribution of the parameters with different types of the covariance matrix is investigated. The authors propose a method of pruning parameters with the highest probability density in zero to increase the model marginal likelihood. As an illustrative example, a computational experiment of multiclass classification on the SemEval 2015 dataset was carried out.
```
@article{смердов2018выбор,
  abbr_en = {IiEP},
  abbr = {ИИП},
  title_en = {Optimal recurrent neural network model in paraphrase detection},
  title = {Выбор оптимальной модели рекуррентной сети в задачах поиска парафраза},
  author_en = {Anton Smerdov and Oleg Bakhteev and Vadim Strijov},
  author = {Смердов, А. Н. and Бахтеев, О. Ю. and Стрижов, В. В.},
  journal_en = {Informatics and Applications},
  journal = {Информатика и её применения},
  volume = {12},
  number = {4},
  pages = {63--69},
  year = {2018},
  publisher = {Российская академия наук, Отделение нанотехнологий и информационных технологий},
  pdf = {http://www.mathnet.ru/links/c8c61f174a24a3081b46138945dffbd5/ia564.pdf}
}
```

IiEP

Automatic metadata extraction from scientific PDF documents [In Russian]

Автоматическое извлечение метаданных из научных PDF-документов

Aleksandr Ogaltsov, Oleg Bakhteev

Informatics and Applications, 2018

Abs PDF Bibtex

The authors investigate the task of metadata extraction. The authors consider scientific PDF documents in Russian. One of the features of PDF is a rich layout. It is difficult to extract metadata due to this fact. The authors propose a method based on considering blocks from pdf-parser as objects in a machine learning task. The feature space is constructed not only of text statistics but also of formatting and positioning features of the block. The authors performed computational experiments and compared their approach with the baseline.

@article{огальцов2018автоматическое,
  abbr_en = {IiEP},
  abbr = {ИИП},
  title_en = {Automatic metadata extraction from scientific PDF documents},
  title = {Автоматическое извлечение метаданных из научных PDF-документов},
  author_en = {Aleksandr Ogaltsov and Oleg Bakhteev},
  author = {Огальцов, А. В. and Бахтеев, О. Ю.},
  journal = {Информатика и её применения},
  journal_en = {Informatics and Applications},
  volume = {12},
  number = {2},
  pages = {75--82},
  year = {2018},
  publisher = {Российская академия наук, Отделение нанотехнологий и информационных технологий},
  pdf = {http://www.mathnet.ru/links/da4e7dd191a181dd48c98361a4ae4523/ia535.pdf}
}

2017

PAN
Author Masking using Sequence-to-Sequence Models

Oleg Bakhteev, Andrey Khazov

In CLEF (Working Notes), 2017

Abs HTML PDF Bibtex

The paper describes the approach adopted for Author Masking Taskat PAN 2017. For the purpose of masking the original author, we use the combi-nation of methods based either on deep learning approach or traditional methodsof obfuscation. We obtain sample of obfuscated sentences from original one andchoose best of them using language model. We try to change both the content andlength of original sentence preserving its meaning.
```
@inproceedings{bakhteev2017author,
  abbr = {PAN},
  title = {Author Masking using Sequence-to-Sequence Models},
  author = {Bakhteev, Oleg and Khazov, Andrey},
  booktitle = {CLEF (Working Notes)},
  year = {2017},
  pdf = {https://pan.webis.de/downloads/publications/papers/bakhteev_2017.pdf},
  html = {https://pan.webis.de/publications.html}
}
```
A method for artificial and unscientific texts detection in a large document collection [In Russian]

Об одном методе детектирования искусственных и ненаучных текстов в обширной коллекции документов

Oleg Bakhteev, Rita Kuznetsova, Alexey Romanov, Yury Chekhovich

Russian Digital Libraries Journal, 2017

Abs PDF Bibtex

In this paper, we propose a method of machine-generated and non-scientific text detection in a collection of scientific papers. The method is based on lexical and morphological analysis of the document examined with the help of language modeling. This technique enables estimation of probability that the text belongs to the class of scientific documents. Experimental evidence shows feasibility of the approach.
```
@article{бахтеев2017одном,
  title_en = {A method for artificial and unscientific texts detection in a large document collection},
  title = {Об одном методе детектирования искусственных и ненаучных текстов в обширной коллекции документов},
  author_en = {Oleg Bakhteev and Rita Kuznetsova and Alexey Romanov and Yury Chekhovich},
  author = {Бахтеев, О. Ю. and Кузнецова, Р. В. and Романов, А. В. and Чехович, Ю. В.},
  journal_en = {Russian Digital Libraries Journal},
  journal = {Электронные библиотеки},
  volume = {20},
  number = {5},
  pages = {298--304},
  year = {2017},
  pdf = {https://elbib.ru/article/download/431/513}
}
```
Plagiarism in scientific papers: challenges in translation detection [In Russian]

Плагиат в научных статьях: трудности обнаружения перевода

Yury Chekhovich, Rita Kuznetsova, Oleg Bakhteev

Universitetskaya kniga [University book], 2017

Abs HTML Bibtex

В течение нескольких последних лет в российской научно-образовательной системе произошли существенные изменения. Был принят и вступил в силу ряд нормативных документов, обязывающих организации обеспечивать проверки квалификационных работ (диссертаций и дипломов) на наличие заимствований. Активно используются технические средства для их обнаружения, например система «Антиплагиат». Подобные действия вынуждают недобросовестных учёных и учащихся искать новые пути использования заимствованного текста
```
@article{чехович2017плагиат,
  title_en = {Plagiarism in scientific papers: challenges in translation detection},
  title = {Плагиат в научных статьях: трудности обнаружения перевода},
  author_en = {Yury Chekhovich and Rita Kuznetsova and Oleg Bakhteev},
  author = {Чехович, Ю. В. and Кузнецова, Р. В. and Бахтеев, О. Ю.},
  journal_en = {Universitetskaya kniga [University book]},
  journal = {Университетская книга},
  number = {9},
  pages = {66--67},
  year = {2017},
  publisher = {Общество с ограниченной ответственностью" ИД Университетская книга"},
  html = {https://elibrary.ru/item.asp?id=30519902}
}
```

2016

Dialogue
Machine-translated text detection in a collection of Russian Scientific papers

Alexey Romanov, Rita Kuznetsova, Oleg Bakhteev, Anton Khritankov

Dialogue-21, 2016

Abs PDF Bibtex

In this paper, we propose a method of machine-translated text detection. By "machine-translated" texts, we mean, principally, the output of statistical machine translation systems. We focus on syntactic correctness and se-mantic consistency of sentences that constitute a text. More specifically, we make an attempt of detecting a certain phenomenon typically occur-ring in machine-translated documents. This phenomenon comprises the cases when small parts of the sentence, correctly translated, are combined together in an improper way. The proposed method is based on a super-vised approach with a number of handcrafted features. First, we construct N-gram language models on a set of authentic scientific papers and on a set of machine-generated texts and assess the probability of each sentence according to these models. In addition, we propose N-gram language models on part-of-speech tag sequences corresponding to the texts given. Furthermore, we explore the effectiveness of features obtained from two trained word2vec (CBOW and skip-gram) models. We assess quality of the method on a sample of Russian scientific papers, and English scientific doc-uments machine-translated into Russian. Preliminary results demonstrate feasibility of the approach.
```
@article{romanov2016machine,
  abbr = {Диалог-21},
  abbr_en = {Dialogue},
  title = {Machine-translated text detection in a collection of Russian Scientific papers},
  author = {Romanov, Alexey and Kuznetsova, Rita and Bakhteev, Oleg and Khritankov, Anton},
  journal = {Диалог-21},
  journal_en = {Dialogue-21},
  pages = {2},
  year = {2016},
  pdf = {http://www.dialog-21.ru/media/3478/dialogue2016.pdf}
}
```
Systems and means of deep learning classification problems [In Russian]

Системы и средства глубокого обучения в задачах классификации

Oleg Bakhteev, Maria Popova, Vadim Strijov

Sistemy i Sredstva Informatiki [Systems and Means of Informatics], 2016

Abs PDF Bibtex

The paper provides a guidance on deep learning net construction and optimization using graphics processing unit. The paper proposes to use GPU-instances on the cloud platform Amazon Web Services. The problem of time series classification is considered. The paper proposes to use a deep learning net, i. e., a multilevel superposition of models, belonging to the following classes: restricted Boltzman machines, autoencoders, and neural nets with softmax-function in output. The proposed method was tested on a dataset containing time segments from mobile phone accelerometer. The analysis of relation between classification error, dataset size, and superposition parameter amount has been conducted.
```
@article{бахтеев2016системы,
  title_en = {Systems and means of deep learning classification problems},
  title = {Системы и средства глубокого обучения в задачах классификации},
  author_en = {Oleg Bakhteev and Maria Popova and Vadim Strijov},
  author = {Бахтеев, О. Ю. and Попова, М. С. and Стрижов, В. В.},
  journal_en = {Sistemy i Sredstva Informatiki [Systems and Means of Informatics]},
  journal = {Системы и средства информатики},
  volume = {26},
  number = {2},
  pages = {4--22},
  year = {2016},
  publisher = {Российская академия наук, Отделение нанотехнологий и информационных технологий},
  pdf = {http://www.mathnet.ru/links/d6c87310460869ba502323ef3b49aa26/ssi459.pdf}
}
```

2015

JMLDA
Panel matrix and ranking model recovery using mixed-scale measured data

Oleg Bakhteev

Journal of machine learning and data analysis, 2015

Abs PDF Bibtex

A decision-making problem is solved in the field of operational research education. The paperpresents a method for recovery of changes in ratings of student employees. These ratings arebased on interviews at the information technology (IT) training center. A dataset consistingof expert estimates for assessments for different years and overall rating for these studentsis considered. The scales of the expert estimates vary from year to year, but the scale of therating remains stable. One should recover the time-independent ranking model. The problem isstated as the object–feature–year panel matrix recovery. It is a map from student descriptions(or their generalized portraits) to expected ratings for all years. Also, a stability of the rankingmodel produced by the panel matrix is studied. A new method ofpanel matrix recovery issuggested. It is based on a solution of multidimensional assignment problem. To constructa ranking model, an ordinal classification algorithm with partially ordered feature sets and analgorithm based on support vector machine have been used. The problem is illustrated by thedataset containing the expert assessment of the student interviews at the IT center.
```
@article{bakhteev2015panel,
  abbr = {JMLDA},
  title = {Panel matrix and ranking model recovery using mixed-scale measured data},
  author = {Bakhteev, Oleg},
  author_en = {Oleg Bakhteev},
  journal_en = {Journal of machine learning and data analysis},
  journal = {Машинное обучение и анализ данных},
  volume = {72},
  number = {7},
  pages = {1958},
  year = {2015},
  pdf = {http://jmlda.org/papers/doc/2015/no14/Bakhteev2015Recovery.pdf}
}
```
JMLDA
Handling missing values in mixed-scale datasetswith large amount of missing values [In Russian]

Восстановление пропущенных значений в разнородных шкалах с большим числом пропусков

Oleg Bakhteev

Journal of machine learning and data analysis, 2015

Abs HTML Bibtex

The paper investigates the problem of missing values handling in datasets with large amount of missing values. The one of the problem of missing values filling methods is their instability. The order of missing values filling can seriously change the efficiency of the method. The paper considers the case when the dataset has significant amount of features with discrete scales with low cardinality. There are some different methods of missing values handling. The paper focuses on the filling missing values using the metric properties of the dataset. The paper proposes some definitions and statements in order to formalize the problem of instability. The method using k-nearest neighbours is considered. The paper considers a variation of the method that uses already filled missing values as values of nearest neighbours for new fill. Also some theoretical aspects of this method implementation are considered. In order to analyse the behaviour and efficiency of the considered method 2 experiments were conducted. The results were compared with other missing values filling techniques such as filling with decision trees and filling with average value of the scale. The proposed mathematical framework can be used for further research of missing values filling methods.
```
@article{бахтеев2015восстановление,
  abbr = {JMLDA},
  title_en = {Handling missing values in mixed-scale datasetswith large amount of missing values},
  title = {Восстановление пропущенных значений в разнородных шкалах с большим числом пропусков},
  author = {Бахтеев, О. Ю.},
  author_en = {Oleg Bakhteev},
  journal_en = {Journal of machine learning and data analysis},
  journal = {Машинное обучение и анализ данных},
  volume = {1},
  number = {11},
  pages = {1484--1499},
  year = {2015},
  language = {ru},
  html = {https://www.elibrary.ru/item.asp?id=24931577}
}
```
AINL
A monolingual approach to detection of text reuse in Russian-English collection

Oleg Bakhteev, Rita Kuznetsova, Alexey Romanov, Anton Khritankov

In 2015 Artificial Intelligence and Natural Language and Information Extraction, Social Media and Web Search FRUCT Conference (AINL-ISMW FRUCT), 2015

Abs HTML PDF Bibtex

In this paper we develop a method for cross-lingual (Russian and English) text reuse detection. The method is based on the monolingual approach - translation of texts into one language and reduction to the text similarity problem. We split texts into non-overlapping fragments and compare fragments to each other by means of different metrics - BLEU(1-2), METEOR, cosine similarity between bag-of-words representations of each snippet, and cosine similarity between vectors obtained from doc2vec-trained model. We explore the impact of choice of metric on the quality of text reuse detection. We assess quality of the method on a sample of a hundred scientific documents, originally in Russian, machine translated into English. Preliminary findings demonstrate feasibility of the approach.
```
@inproceedings{bakhteev2015monolingual,
  abbr = {AINL},
  title = {A monolingual approach to detection of text reuse in Russian-English collection},
  author = {Bakhteev, Oleg and Kuznetsova, Rita and Romanov, Alexey and Khritankov, Anton},
  booktitle = {2015 Artificial Intelligence and Natural Language and Information Extraction, Social Media and Web Search FRUCT Conference (AINL-ISMW FRUCT)},
  pages = {3--10},
  year = {2015},
  organization = {IEEE},
  html = {https://ieeexplore.ieee.org/abstract/document/7382960},
  pdf = {https://www.fruct.org/publications/ainl-fruct/files/Bak.pdf}
}
```

Talks

2024

Polmeth
After the enfranchisement: Legislative elites’ reactions to women’s inclusion

Laurence Brandeberger, Jorge M. Fernandes, Sophia Schlosser, Oleg Bakteev, Luis Salamanca

2024

Speaker: Laurence Brandeberger
California, USA
HTML PDF Bibtex
```
@conference{brandeberger2024embed2discover,
  abbr = {Polmeth},
  presenter = {Laurence Brandeberger},
  presenter_en = {Laurence Brandeberger},
  title = {After the enfranchisement: Legislative elites' reactions to women's inclusion},
  author = {Brandeberger, Laurence and Fernandes, Jorge M. and Schlosser, Sophia and Bakteev, Oleg and Salamanca, Luis},
  year = {2024},
  note = {California, USA},
  html = {https://polmeth.ucr.edu/},
  pdf = {https://files.osf.io/v1/resources/gsmqf/providers/osfstorage/668c516cede75d063e693a81?action=download&direct&version=2}
}
```

2023

ICAART
Deep Learning Model Selection With Parametric Complexity Control

Olga Grebenkova, Oleg Bakhteev, Vadim Strijov

2023

Speaker: Olga Grebenkova
Online presentation
Abs HTML Bibtex

The paper is devoted to deep learning model complexity. It is estimated by Bayesian inference and based on a computational budget. The idea of the proposed method is to represent deep learning model parameters in the form of hypernetwork output. A hypernetwork is a supplementary model which generates parameters of the selected model. This paper considers the minimum description length from a Bayesian point of view. We introduce prior distributions of deep learning model parameters to control the model complexity. The paper analyzes and compares three types of regularization to define the parameter distribution. It infers and generalizes the model evidence as a criterion that depends on the required model complexity. Finally, it analyzes this method in the computational experiments on the Wine, MNIST, and CIFAR-10 datasets.
```
@conference{grebenkova2023icaart,
  presenter = {Гребенькова О.С.},
  presenter_en = {Olga Grebenkova},
  author = {Grebenkova, Olga and Bakhteev, Oleg and Strijov, Vadim},
  abbr = {ICAART},
  title = {Deep Learning Model Selection With Parametric Complexity Control},
  booktitle = {International Conference on Agents and Artificial Intelligence (ICAART 2023)},
  note = {Online presentation},
  html = {https://www.insticc.org/node/TechnicalProgram/icaart/2023/presentationDetails/116269},
  year = {2023}
}
```

2022

IDP
Concordant neural architecture search on multi-domain data [In Russian]

Поиск согласованных нейросетевых моделей в задаче мультидоменного обучения

Yakovlev, Konstantin, Bakhteev, Oleg, Strijov, Vadim

2022

Speaker: Konstantin Yakovlev
Moscow
Abs Bibtex

The paper investigates the problem of selection the structure of a deep learning model for multi-domain data. The model is an directed acyclic graph in which the edges are nonlinear functions differentiable by parameters. The structure of the model is determined by a set of vectors that define a categorical distribution on the edges of the graph. It is proposed to consider the problem as a multimodeling problem: a separate structure is optimized for each domain. Two regularization methods are considered: structural and regularization of the space of hidden representations of the model. Structural regularization is based on the approximation of probability distributions, which defines the structure. Regularization of the space of hidden representations of the model is based on the approximation of hidden representations obtained under the action of nonlinear functions. The paper shows that the proposed optimization problem makes it possible to find a compromise between the complexity of the final model and its performance. The basic experiment is performed on a MNIST dataset. Considered models trained on its domain, a model trained on all domains, as well as a proposed multimodel trained on all domains with proposed regularizations. It is shown that the proposed model has the best performance.
```
@conference{yakovlevl2022idp,
  presenter = {Яковлев К. Д.},
  presenter_en = {Konstantin Yakovlev},
  author_en = {Yakovlev, Konstantin and Bakhteev, Oleg and Strijov, Vadim},
  author = {Яковлев, К. Д. and Бахтеев, О. Ю. and В., Стрижов В.},
  abbr_en = {IDP},
  abbr = {ИОИ},
  title = {Поиск согласованных нейросетевых моделей в задаче мультидоменного обучения},
  title_en = {Concordant neural architecture search on multi-domain data},
  booktitle_en = {Intelligent Data Processing: Theory and Applications},
  booktitle = {Интеллектуализация обработки информаци},
  note = {Москва},
  note_en = {Moscow},
  year = {2022}
}
```
IDP
Methods of near-duplicate handwritten document search in large collections of texts [In Russian]

Методы поиска почти-дубликатов рукописных документов в больших коллекциях текстов

Oleg Bakhteev et al.

2022

Speaker: Anrey Grabovoy
Moscow
Abs Supp Bibtex

This paper considers the problem for searching of near-duplicate of school essays in large data collections. The problem is conditioned by the availability of school essay collections, which can be used for writing essays and essays by school students. The relevance of the task is confirmed by the worksdevoted to the analysis of violations in writing academic tests, as well as the partial transition of school education to remote mode. The near-duplicate search problem is considered as an information search problem, where an essay is matched with a reused text from the collection. In the framework of the problem, an essay is represented as a set of scanned images of handwritten text, while a collection of essays is represented by machine-readable texts. This paper compares two approaches to near-duplicate search: a deep learning-based search and an approach based on text length sequence analysis. The deep learning-based search uses a neural network model that is optimized in a supervised manner. This approach consists of two stages. The first stage performs Optical Character Recognition (OCR). The second step is bigram search in the collection. The second method involves treating the text in scans of school essays as a sequence of features extracted from the text. The method extracts words from the image without further recognition. Based on the extracted words, a sequence of normalized word lengths is constructed, which is invariant for handwritten and machine-readable variants of the text. This paper compares the quality of the search with two datasets: a dataset of essays written on special forms corresponding to the forms of the state exam, and a dataset of essays written on various types of forms and notebook sheets. A subsample of the Taiga open corpus is used as a search collection. The results of the experiment demonstrate the applicability of both methods to the considered problem. It is shown that the deep learning-based approach is more robust to data heterogeneity, and in particular, it copes better with the heterogeneity of the essay layout, as well as the general quality of scans. At the same time, with good image scanning quality and the use of standardized essay forms, both methods show an acceptable search quality.
```
@conference{bakhteev2022idp,
  presenter = {Грабовой А. В.},
  presenter_en = {Anrey Grabovoy},
  abbr_en = {IDP},
  abbr = {ИОИ},
  title = {Методы поиска почти-дубликатов рукописных документов в больших коллекциях текстов},
  title_en = {Methods of near-duplicate handwritten document search in large collections of texts},
  author = {Бахтеев, О. Ю. и др.},
  author_en = {Oleg Bakhteev et al.},
  booktitle_en = {Intelligent Data Processing: Theory and Applications},
  booktitle = {Интеллектуализация обработки информаци},
  note = {Москва},
  note_en = {Moscow},
  supp = {http://www.machinelearning.ru/wiki/images/5/54/Ioi-ap-grabovoy-2022.pdf},
  year = {2022}
}
```
ISP
RuGECToR: rule based error correction model for Russian language [In Russian]

RuGECToR: нейросетевая модель на основе правил для исправления грамматических ошибок на русском языке

Ildar Khabutdinov et al.

2022

Speaker: Ildar Khabutdinov
Moscow
Abs Supp Bibtex

Grammatical Error Correction is one of the core Natural Language Processing tasks. At the moment, the open source state-of-the-art Sequence Tagger for English is GECToR model. For the Russian language, this problem does not have solutions with the same good results due to the lack of labeled datasets, therefore we decided to contribute to the aforementioned task. In this research, we described the process of creating a synthetic dataset and training a model on it. We adapt GECToR architecture for Russian language and call it RuGECToR. We use this architecture because, unlike Sequence-to-Sequence approach, it is easy to interpret and does not require a lot of training data. Our goal was to train the model in such a way that it does not adapt to a specific sample, but generalizes the mophorological properties of the language. The presented model achieve an F0.5 of 82.5 on synthetic data and an F0.5 of 22.2 on RULEC dataset which was not in the training set.
```
@conference{khabutdinov2022isp,
  presenter = {Хабутдинов И. А.},
  presenter_en = {Ildar Khabutdinov},
  abbr = {ИСП},
  abbr_en = {ISP},
  title = {RuGECToR: нейросетевая модель на основе правил для исправления грамматических ошибок на русском языке},
  title_en = {RuGECToR: rule based error correction model for Russian language},
  author = {Хабутдинов И. А., и др.},
  author_en = {Ildar Khabutdinov et al.},
  booktitle = {Ivannikov ISPRAS Open Conference},
  supp = {https://youtu.be/f1DN4M8HWf4?t=13737},
  note = {Москва},
  note_en = {Moscow},
  year = {2022}
}
```
ISP
Anti-Distillation: Knowledge Transfer from a Simple Model to the Complex One [In Russian]

Антидистилляция: перенос знаний от простой модели к более сложной

Petrushina, Kseniia, Bakhteev, Oleg, Grabovoy, Andrey abd Strijov, Vadim

2022

Speaker: Kseniia Petrushina
Moscow
Abs Supp Bibtex

The paper considers the problem of adapting the model to new data with a large amount of information. We propose to build a more complex model using the parameters of a simple one. We take into account not only the accuracy of the prediction on the original samples but also the adaptability to new data and the robustness of the obtained solution. The work is devoted to developing the method that allows adapting the pre-trained model to a more heterogeneous dataset. In the computational experiment, we analyse the quality of predictions and model robustness on Fashion-MNIST dataset.
```
@conference{petrushina2022isp,
  presenter = {Петрушина К. Е.},
  presenter_en = {Kseniia Petrushina},
  abbr = {ИСП},
  abbr_en = {ISP},
  title = {Антидистилляция: перенос знаний от простой модели к более сложной},
  title_en = {Anti-Distillation: Knowledge Transfer from a Simple Model to the Complex One},
  author_en = {Petrushina, Kseniia and Bakhteev, Oleg and Grabovoy, Andrey  abd Strijov, Vadim},
  author = {Петрушина, К. Е. and Бахтеев, О. Ю. and Грабовой, А. В. and Стрижов, В. В.},
  booktitle = {Ivannikov ISPRAS Open Conference},
  supp = {https://youtu.be/RtttuJMnyzo?t=14545},
  note = {Москва},
  note_en = {Moscow},
  year = {2022}
}
```

ENAI

Image reuse detection in large-scale document scientific collection

Oleg Bakhteev et al.

2022

Speaker: Mariam Kaprielova

Online conference

PDF Bibtex

@conference{bakhteev2022enai,
  presenter = {Каприелова М. С.},
  presenter_en = {Mariam Kaprielova},
  abbr = {ENAI},
  title = {Image reuse detection in large-scale document scientific collection},
  author = {Bakhteev, Oleg et al.},
  author_en = {Oleg Bakhteev et al.},
  booktitle = {European Conference on Academic Integrityand Plagiarism},
  note = {Online conference},
  year = {2022},
  pdf = {https://academicintegrity.eu/conference/wp-content/files/2022/Book_of_Abstracts_2022.pdf}
}

2021

MMPR

Image reuse detection in large-scale document scientific collection [In Russian]

Поиск заимствованных изображений в больших коллекциях научных документов

Evgeny Finogeev et al.

2021

Speaker: Evgeny Finogeev

Online conference

Abs PDF Supp Bibtex

We consider the problem of information retrieval of reused images in extensive collections of scientific documents. The problem’s relevance is due to the presence of precedents for reusing images from other sources in the field of medicine and biology. Thus, it is shown that up to 4 percents of misconducted images are found in scientific articles on biology and medicine. At the same time, until now, there were no automatic means to search for reused images.

@conference{финогеев2021,
  presenter = {Финогеев Е. Л.},
  presenter_en = {Evgeny Finogeev},
  abbr = {ММРО},
  abbr_en = {MMPR},
  title_en = {Image reuse detection in large-scale document scientific collection},
  title = {Поиск заимствованных изображений в больших коллекциях научных документов},
  author_en = {Evgeny Finogeev et al.},
  author = {Финогеев, Е. Л. and и, др.},
  booktitle_en = {Mathematical methods of pattern recognition},
  booktitle = {Математические методы распознавания образов},
  note_en = {Online conference},
  note = {Онлайн-конференция},
  year = {2021},
  supp = {https://youtu.be/JEPhbyW3Eak},
  pdf = {http://www.machinelearning.ru/wiki/images/0/02/Mmpr_2021.pdf}
}

MMPR
Multi-task Learning in the Problem of Rubrication of Scientific Documents [In Russian]

Многозадачное обучение в задаче рубрикации научных документов

Oleg Shevchenko et al.

2021

Speaker: Oleg Shevchenko
Online conference
Abs PDF Supp Bibtex

The paper is devoted to solving the problem of rubricating scientific publications, theses, dissertations, reports and other scientifically intensive texts. Rubrication is traditionally posed as a problem of multiclass classification of texts, where a rubric in classifier used is considered as a class. In this paper, we describes the application of the multi-task learning approach to the problem of rubricating scientific documents. The decision-making stage for various rubricators is divided into several classification tasks. Each rubricator has its classification task within the multi-task approach. As rubricators of scientific documents can be used, for example, rubrics of different levels of the State Rubricator of Scientific and Technical Information (SRSTI) or the classifier of the Organization for Economic Cooperation and Development (OECD). Pretrained thematic vectors are used as a feature description of documents, while the document’s thematic vector describes the probability that the document belongs to a particular topic.
```
@conference{шевченко2021,
  presenter = {Шевченко О. В.},
  presenter_en = {Oleg Shevchenko},
  abbr = {ММРО},
  abbr_en = {MMPR},
  title_en = {Multi-task Learning in the Problem of Rubrication of Scientific Documents},
  title = {Многозадачное обучение в задаче рубрикации научных документов},
  author_en = {Oleg Shevchenko et al.},
  author = {Шевченко, О. В. and и, др.},
  booktitle_en = {Mathematical methods of pattern recognition},
  booktitle = {Математические методы распознавания образов},
  note_en = {Online conference},
  note = {Онлайн-конференция},
  year = {2021},
  supp = {https://youtu.be/UhhZ7O5XhDA},
  pdf = {http://www.machinelearning.ru/wiki/images/0/02/Mmpr_2021.pdf}
}
```
MMPR
Model selection using Bayesian hypernetworks [In Russian]

Порождение моделей заданной сложности с использованием байесовских гиперсетей

Oleg Grebenkova, Oleg Bakhteev, Vadim Strijov

2021

Speaker: Olga Grevenkova
Online conference
Abs PDF Supp Bibtex

The paper considers the problem of optimizing a deep learning model. Optimal model selection is the fundamental problem of deep learning, since this family of models has an excessive number of parameters implicitly. The paper assumes that the complexity of the model is given. We consider the model complexity as a coefficient assigned during model final training depending on the operational requirements.
```
@conference{гребенькова2021порождение,
  presenter = {Гребенькова О. С.},
  presenter_en = {Olga Grevenkova},
  abbr = {ММРО},
  abbr_en = {MMPR},
  title_en = {Model selection using Bayesian hypernetworks},
  title = {Порождение моделей заданной сложности с использованием байесовских гиперсетей},
  author_en = {Oleg Grebenkova and Oleg Bakhteev and Vadim Strijov},
  author = {Гребенькова, О. С. and Бахтеев, О. Ю. and Стрижов, В. В.},
  booktitle_en = {Mathematical methods of pattern recognition},
  booktitle = {Математические методы распознавания образов},
  note_en = {Online conference},
  note = {Онлайн-конференция},
  year = {2021},
  supp = {https://youtu.be/lsE88wqmwmc},
  pdf = {http://www.machinelearning.ru/wiki/images/0/02/Mmpr_2021.pdf}
}
```
MMPR
Gradient-based metaparameter optimization in knowledge distillation task [In Russian]

Градиентные методы оптимизации метапараметров в задаче дистилляции знаний

Mariya Gorpinich, Oleg Bakhteev, Vadim Strijov

2021

Speaker: Mariya Gorpinich
Online conference
Abs PDF Supp Bibtex

The paper investigates the knowledge distillation problem for deep learning models. Knowledge distillation is a model parameter optimization problem that takes into account not only information contained in the initial dataset but also the information contained in the model with a more complex structure that is called the teacher model. The loss function has the initial dataset likelihood term and distillation dataset likelihood term. Metaparameters of this problem are the parameters of the optimization problem, namely, the coefficient of the distillation term in loss function and a temperature factor. The student model parameters are optimized by transferring the knowledge of a teacher model. The knowledge distillation problem with metaparameter optimization is a bi-level problem. Metaparameters are optimized on the first level and model parameters are optimized on the second. Gradient methods are used to solve the bi-level problem. To make the solution less computationally expensive we alternate gradient optimization steps and optimization trajectory prediction with linear models.
```
@conference{горпинич2021,
  presenter = {Горпинич М.},
  presenter_en = {Mariya Gorpinich},
  abbr = {ММРО},
  abbr_en = {MMPR},
  title_en = {Gradient-based metaparameter optimization in knowledge distillation task},
  title = {Градиентные методы оптимизации метапараметров в задаче дистилляции знаний},
  author_en = {Mariya Gorpinich and Oleg Bakhteev and Vadim Strijov},
  author = {Горпинич, М. and Бахтеев, О. Ю. and Стрижов, В. В.},
  booktitle_en = {Mathematical methods of pattern recognition},
  booktitle = {Математические методы распознавания образов},
  note_en = {Online conference},
  note = {Онлайн-конференция},
  year = {2021},
  supp = {https://youtu.be/0VztZVv7a08},
  pdf = {http://www.machinelearning.ru/wiki/images/0/02/Mmpr_2021.pdf}
}
```
MMPR
Differentiable architecture search with model complexity control [In Russian]

Дифференцируемый алгоритм поиска архитектуры с контролем сложности

Konstantin Yakovlev et al.

2021

Speaker: Konstantin Yakovlev
Online conference
Abs PDF Supp Bibtex

The paper investigates the problem of optimizing a deep learning model with the control of complexity of the architecture. The proposed method is based on a differentiable architecture search algorithm (DARTS). The structural parameters are generated by a hypernetwork, depending on a regularization parameter. The hypernetwork is a model that generates parameters of the optimized model. The structure of the model is presented in the form of a directed acyclic graph, where the edges correspond to nonlinear maps on the objects, and the vertices are intermediate representations of the objects under the action of these maps. The structural parameters corresponding to a certain edge are taken from Gumbel-Softmax distribution.
```
@conference{яковлев2021,
  presenter = {Яковлев К. Д.},
  presenter_en = {Konstantin Yakovlev},
  abbr = {ММРО},
  abbr_en = {MMPR},
  title_en = {Differentiable architecture search with model complexity control},
  title = {Дифференцируемый алгоритм поиска архитектуры с контролем сложности},
  author_en = {Konstantin Yakovlev et al.},
  author = {Яковлев, К. Д. and и, др.},
  booktitle_en = {Mathematical methods of pattern recognition},
  booktitle = {Математические методы распознавания образов},
  note_en = {Online conference},
  note = {Онлайн-конференция},
  year = {2021},
  supp = {https://youtu.be/khJGw_zffZc},
  pdf = {http://www.machinelearning.ru/wiki/images/0/02/Mmpr_2021.pdf}
}
```
ISP RAS
The automatic approach for scientific papers dating

Andrey Grabovoy, Oleg Bakhteev, Yury Chekhovich

2021

Speaker: Andrey Grabovoy
Moscow, Russia
Abs Supp Bibtex

The paper investigates automatic methods of document date estimation. The document date is essential metadata for many applied tasks, such as event detection or internet search. Methods for estimating the dates of newspaper articles are well known. In this paper, we propose to predict the publishing year for scientific papers. We present a new dataset based on open access articles from eLIBRARY.RU. It contains articles written in Russian and English. The article introduces a method of dating documents based on the raw document text content. The computational experiment compares different solutions for document date estimation. We also analyze how prediction depends on the topic for different methods.
```
@conference{grabovoy2021automatic,
  abbr = {ISP RAS},
  presenter = {Грабовой А. В.},
  presenter_en = {Andrey Grabovoy},
  title = {The automatic approach for scientific papers dating},
  author = {Grabovoy, Andrey and Bakhteev, Oleg and Chekhovich, Yury},
  booktitle = {Ivannikov ISPRAS Open Conference},
  year = {2021},
  note = {Москва},
  note_en = {Moscow, Russia},
  supp = {https://www.youtube.com/watch?v=cOM82kHRwkM}
}
```

MIPT

Metaparameter optimization in knowledge distatillation task [In Russian]

Оптимизация метапараметров в задаче дистилляции знаний

Mariya Gorpinich, Oleg Bakhteev, Vadim Strijov

2021

Speaker: Mariya Gorpinich

Moscow, Russia

Supp Bibtex

@conference{яковлев2021мфти,
  presenter = {Горпинич М.},
  presenter_en = {Mariya Gorpinich},
  abbr = {МФТИ},
  abbr_en = {MIPT},
  title = {Оптимизация метапараметров в задаче дистилляции знаний},
  title_en = {Metaparameter optimization in knowledge distatillation task},
  booktitle_en = {Moscow Institute of Physics and Technology 64th student conference},
  author_en = {Mariya Gorpinich and Oleg Bakhteev and Vadim Strijov},
  author = {Горпинич, М. and Бахтеев, О. Ю. and Стрижов, В. В.},
  booktitle = {64-я конференция МФТИ},
  note = {Москва},
  note_en = {Moscow, Russia},
  year = {2021},
  supp = {https://youtu.be/rmxjhGqQCbg}
}

MIPT

Deep learning neural architecture selection with complexity control [In Russian]

Выбор архитектуры модели с контролем сложности

Konstantin Yakovlev et al.

2021

Speaker: Konstantin Yakovlev

Moscow, Russia

Supp Bibtex

@conference{горпинич2021мфти,
  presenter = {Яковлев К. Д.},
  presenter_en = {Konstantin Yakovlev},
  abbr = {МФТИ},
  abbr_en = {MIPT},
  title = {Выбор архитектуры модели с контролем сложности},
  title_en = {Deep learning neural architecture selection with complexity control},
  booktitle_en = {Moscow Institute of Physics and Technology 64th student conference},
  author_en = {Konstantin Yakovlev et al.},
  author = {Яковлев, К. Д. and и, др.},
  booktitle = {64-я конференция МФТИ},
  note = {Москва},
  note_en = {Moscow, Russia},
  year = {2021},
  supp = {https://youtu.be/rmxjhGqQCbg}
}

Dialogue
Near-duplicate handwritten document detection without text recognition

Oleg Bakhteev et al.

2021

Speaker: Oleg Bakhteev
Online conference
Abs Supp Bibtex

The paper presents a novel method for near-duplicate detection in handwritten document collections of schoolessays. A large amount of online resources with available academic essays currently makes it possible to cheatand reuse them during high school final exams. Despite the importance of the problem, at the moment there isno automatic method for near-duplicate detection for handwritten documents, such as school essays. The schoolessay is represented as a sequence of scanned images of handwritten essay text. Despite advances in recognitionof handwritten printed text, the use of these methods for the current task is a challenge. The proposed method ofnear-duplicate detection does not require detailed markup text, which makes it possible to use it in a large numberof tasks related to the information extraction in zero-shot regime, i.e. without any specific resources written in theprocessed language. The paper presents a method based on series analysis. The image is segmented into words.The text is characterized by a sequence of features, which are invariant to the author’s writing style: normalizedlengths of the segmented words. These features can be used for both handwritten and machine-readable texts. Thecomputational experiment is conducted on IAM dataset of English handwritten texts and the dataset of real imagesof handwritten school essays.
```
@conference{bakhteev2021handwrittentalk,
  presenter = {Бахтеев О. Ю.},
  presenter_en = {Oleg Bakhteev},
  abbr = {Диалог-21},
  abbr_en = {Dialogue},
  title = {Near-duplicate handwritten document detection without text recognition},
  author = {Bakhteev, Oleg et al.},
  author_en = {Oleg Bakhteev et al.},
  booktitle = {Dialogue, Conference on Computational Linguistics},
  booktitle_en = {Диалог, Конференция по компьютерной лингвистике},
  note = {Онлайн-конференция},
  note_en = {Online conference},
  year = {2021},
  supp = {http://bit.ly/ap_handwritten}
}
```

ENAI

Cross-language plagiarism detection: a case study of European universities academic works

Oleg Bakhteev et al.

2021

Speaker: Oleg Bakhteev

Online conference

PDF Bibtex

@conference{bakhteev2021enai,
  presenter = {Бахтеев О. Ю.},
  presenter_en = {Oleg Bakhteev},
  abbr = {ENAI},
  title = {Cross-language plagiarism detection: a case study of European universities academic works},
  author = {Bakhteev, Oleg et al.},
  author_en = {Oleg Bakhteev et al.},
  booktitle = {European Conference on Academic Integrityand Plagiarism},
  note = {Online conference},
  year = {2021},
  pdf = {https://academicintegrity.eu/conference/proceedings/2021/book_of_abstracts2021.pdf}
}

2020

MIPT

Deep learning model generation using Bayesian hypernetworks [In Russian]

Порождение моделей глубокого обучения с использованием байесовских гиперсетей

Oleg Grebenkova, Oleg Bakhteev, Vadim Strijov

2020

Speaker: Olga Grebenkova

Moscow, Russia

PDF Bibtex

@conference{гребенькова2020мфти,
  presenter = {Гребенькова О. С.},
  presenter_en = {Olga Grebenkova},
  abbr = {МФТИ},
  abbr_en = {MIPT},
  title = {Порождение моделей глубокого обучения с использованием байесовских гиперсетей},
  title_en = {Deep learning model generation using Bayesian hypernetworks},
  booktitle_en = {Moscow Institute of Physics and Technology 63th student conference},
  author_en = {Oleg Grebenkova and Oleg Bakhteev and Vadim Strijov},
  author = {Гребенькова, О. С. and Бахтеев, О. Ю. and Стрижов, В. В.},
  booktitle = {63-я конференция МФТИ},
  note = {Москва},
  note_en = {Moscow, Russia},
  year = {2020},
  pdf = {https://mipt.ru/science/5top100/education/courseproposal/%D0%A4%D0%9F%D0%9C%D0%98.pdf}
}

IDP
Near-duplicate detection in handwritten school essays [In Russian]

Поиск почти-дубликатов в рукописных текстах школьных сочинений

Oleg Bakhteev et al.

2020

Speaker: Oleg Bakhteev
Online conference
Abs PDF Supp Bibtex

The paper is devoted to the problem of near-duplicate detection in collections of school essays. Due to the large amount of online resources with available school essays the problem of text reuse detection is actual. This problem is important for the education system, which was noted in a number of works, devoted to the analysis of cheating in writing essays and passing academic tests. Despite the importance of the problem, at the moment there are no automatic methods for text reuse detection available for the school essays analysis. The task of near-duplicate detection is considered as an information search problem. It is assumed that authors of essays usually use only one text as a source of text reuse. The essay is represented as a sequence of scanned images of handwritten essay text. Traditional search engines for near-duplicate detection consider the text as an object to analyse. Despite advances in recognition of printed text, as well as handwritten text written using touch devices, the use of these methods for current task is a challenge. Unlike the methods used for the recognition of printed text, the methods of handwritten text recognition have a rather low quality, which does not allow them to be used to search for the near-duplicates in the essays texts. The major problem in school essays text recognition is the inability of the text recognition system to adapt to a large variety of handwriting styles. Since the essays are written by different people, there is no opportunity to fine-tune to the specific author writing style, which also complicates the task. Another problem of near-duplicate detection methods based on text recognition is the requirements of the markup — the correspondence between the image region and the text in it. Such a markup can rarely be found in the public domain. The proposed method of near-duplicate detection does not require detailed markup text, which makes it possible to use it in a large number of tasks related to the information extraction from the images of handwritten text.
```
@conference{бахтеев2020иои,
  presenter = {Бахтеев О. Ю.},
  presenter_en = {Oleg Bakhteev},
  abbr = {ИОИ},
  abbr_en = {IDP},
  title_en = {Near-duplicate detection in handwritten school essays},
  title = {Поиск почти-дубликатов в рукописных текстах школьных сочинений },
  author_en = {Oleg Bakhteev et al.},
  author = {Бахтеев, О. Ю. and и, др.},
  booktitle_en = {Intelligent Data Processing: Theory and Applications},
  booktitle = {Интеллектуализация обработки информаци},
  note_en = {Online conference},
  note = {Онлайн-конференция},
  year = {2020},
  supp = {https://www.youtube.com/watch?v=ZuT6xs_mTvI},
  pdf = {http://www.machinelearning.ru/wiki/images/3/31/Idp20.pdf}
}
```

2018

IDP

Bayesian deep learning optimal model structure selection [In Russian]

Байесовский выбор наиболее правдоподобной структуры модели глубокого обучения

Oleg Bakhteev

2018

Speaker: Oleg Bakhteev

Gaeta, Italy

PDF Supp Bibtex

@conference{бахтеев2018иои,
  presenter = {Бахтеев О. Ю.},
  presenter_en = {Oleg Bakhteev},
  abbr = {ИОИ},
  abbr_en = {IDP},
  title_en = {Bayesian deep learning optimal model structure selection},
  title = {Байесовский выбор наиболее правдоподобной структуры модели глубокого обучения},
  booktitle_en = {Intelligent Data Processing: Theory and Applications},
  author_en = {Oleg Bakhteev},
  author = {Бахтеев, О. Ю.},
  booktitle = {Интеллектуализация обработки информаци},
  note = {Gaeta, Italy},
  year = {2018},
  pdf = {http://www.machinelearning.ru/wiki/images/e/e2/IDP18.pdf},
  supp = {http://www.machinelearning.ru/wiki/images/1/10/BakhteevIDP18.pdf}
}

IDP

Estimation of the relevance of the neural network parameters [In Russian]

Определене релевантности параметров нейросети

Andrey Grabovoy, Oleg Bakhteev, Vadim Strijov

2018

Speaker: Andrey Grabovoy

Gaeta, Italy

PDF Supp Bibtex

@conference{грабовой2018иои,
  presenter = {Грабовой А. В.},
  presenter_en = {Andrey Grabovoy},
  abbr = {ИОИ},
  abbr_en = {IDP},
  title_en = {Estimation of the relevance of the neural network parameters},
  title = {Определене релевантности параметров нейросети},
  booktitle_en = {Intelligent Data Processing: Theory and Applications},
  author_en = {Andrey Grabovoy and Oleg Bakhteev and Vadim Strijov},
  author = {Грабовой, А. В. and Бахтеев, О. Ю. and Стрижов, В. В.},
  booktitle = {Интеллектуализация обработки информаци},
  note = {Gaeta, Italy},
  year = {2018},
  pdf = {http://www.machinelearning.ru/wiki/images/e/e2/IDP18.pdf},
  supp = {http://www.machinelearning.ru/wiki/images/e/e0/Grabovoy.pdf}
}

2017

CLEF
Author Masking using Sequence-to-Sequence Models

Oleg Bakhteev, Andrey Khazov

2017

Speaker: Oleg Bakhteev
Dublin, Ireland
Abs HTML PDF Bibtex

The paper describes the approach adopted for Author Masking Taskat PAN 2017. For the purpose of masking the original author, we use the combi-nation of methods based either on deep learning approach or traditional methodsof obfuscation. We obtain sample of obfuscated sentences from original one andchoose best of them using language model. We try to change both the content andlength of original sentence preserving its meaning.
```
@conference{bakhteev2017author,
  presenter = {Бахтеев О. Ю.},
  presenter_en = {Oleg Bakhteev},
  abbr = {CLEF},
  title = {Author Masking using Sequence-to-Sequence Models},
  author = {Bakhteev, Oleg and Khazov, Andrey},
  booktitle = {CLEF},
  note = {Dublin, Ireland},
  year = {2017},
  pdf = {https://pan.webis.de/downloads/publications/papers/bakhteev_2017.pdf},
  html = {https://pan.webis.de/publications.html}
}
```

MMPR

Gradient-based methods of deep learning hyperparameters optimization [In Russian]

Градиентные методы оптимизации гиперпараметров моделей глубокого обучения

Oleg Bakhteev

2017

Speaker: Oleg Bakhteev

Taganrog, Russia

HTML Bibtex

@conference{бахтеев2017ммро1,
  presenter = {Бахтеев О. Ю.},
  presenter_en = {Oleg Bakhteev},
  abbr = {ММРО},
  abbr_en = {MMPR},
  title = {Градиентные методы оптимизации гиперпараметров моделей глубокого обучения},
  title_en = {Gradient-based methods of deep learning hyperparameters optimization},
  author = {Бахтеев, О. Ю.},
  author_en = {Oleg Bakhteev},
  booktitle_en = {Mathematical methods of pattern recognition},
  booktitle = {Математические методы распознавания образов},
  note = {Таганрог},
  note_en = {Taganrog, Russia},
  year = {2017},
  html = {https://www.elibrary.ru/item.asp?id=32284754}
}

MMPR

Cross-lingual text reuse detection in scientific papers included in RSCI [In Russian]

Детектирование переводных заимствований в текстах научных статей из журналов, входящих в РИНЦ

Oleg Bakhteev, Rita Kuznetsova

2017

Speaker: Oleg Bakhteev

Taganrog, Russia

HTML Bibtex

@conference{бахтеев2017ммро2,
  presenter = {Бахтеев О. Ю.},
  presenter_en = {Oleg Bakhteev},
  abbr = {ММРО},
  abbr_en = {MMPR},
  title = {Детектирование переводных заимствований в текстах научных статей из журналов, входящих в РИНЦ},
  title_en = {Cross-lingual text reuse detection in scientific papers included in RSCI},
  booktitle_en = {Mathematical methods of pattern recognition},
  author_en = {Oleg Bakhteev and Rita Kuznetsova},
  author = {Бахтеев, О. Ю. and Кузнецова, Р. В.},
  booktitle = {Математические методы распознавания образов},
  note = {Таганрог},
  note_en = {Taganrog, Russia},
  year = {2017},
  html = {https://www.elibrary.ru/item.asp?id=32284813}
}

2015

AINL
A monolingual approach to detection of text reuse in Russian-English collection

Oleg Bakhteev, Rita Kuznetsova, Alexey Romanov, Anton Khritankov

2015

Speaker: Oleg Bakhteev
Saint Petersburg, Russia
Abs HTML PDF Bibtex

In this paper we develop a method for cross-lingual (Russian and English) text reuse detection. The method is based on the monolingual approach - translation of texts into one language and reduction to the text similarity problem. We split texts into non-overlapping fragments and compare fragments to each other by means of different metrics - BLEU(1-2), METEOR, cosine similarity between bag-of-words representations of each snippet, and cosine similarity between vectors obtained from doc2vec-trained model. We explore the impact of choice of metric on the quality of text reuse detection. We assess quality of the method on a sample of a hundred scientific documents, originally in Russian, machine translated into English. Preliminary findings demonstrate feasibility of the approach.
```
@conference{bakhteev2015monolingual,
  presenter = {Бахтеев О. Ю.},
  presenter_en = {Oleg Bakhteev},
  abbr = {AINL},
  title = {A monolingual approach to detection of text reuse in Russian-English collection},
  author = {Bakhteev, Oleg and Kuznetsova, Rita and Romanov, Alexey and Khritankov, Anton},
  booktitle = {2015 Artificial Intelligence and Natural Language and Information Extraction, Social Media and Web Search FRUCT Conference (AINL-ISMW FRUCT)},
  note = {Saint Petersburg},
  note_en = {Saint Petersburg, Russia},
  pages = {3--10},
  year = {2015},
  organization = {IEEE},
  html = {https://ieeexplore.ieee.org/abstract/document/7382960},
  pdf = {https://www.fruct.org/publications/ainl-fruct/files/Bak.pdf}
}
```

2014

MIPT

Panel matrix and ranking model recovery using mixed-scale measured data [In Russian]

Восстановление панельной матрицы и ранжирующей модели по метризованной выборке в разнородных шкалах

Oleg Bakhteev

2014

Speaker: Oleg Bakhteev

Moscow, Russia

Bibtex

@conference{бахтеев2014восстановление,
  presenter = {Бахтеев О. Ю.},
  presenter_en = {Oleg Bakhteev},
  abbr = {МФТИ},
  abbr_en = {MIPT},
  title = {Восстановление панельной матрицы и ранжирующей модели по метризованной выборке в разнородных шкалах},
  title_en = {Panel matrix and ranking model recovery using mixed-scale measured data},
  booktitle_en = {Moscow Institute of Physics and Technology 57th student conference},
  author_en = {Oleg Bakhteev},
  author = {Бахтеев, О. Ю.},
  booktitle = {57-я конференция МФТИ},
  note = {Москва},
  note_en = {Moscow, Russia},
  year = {2014}
}

Workshops and poster sessions

2021

Automated architecture search with model complexity control

Konstantin Yakovlev et al.

2021

Speaker: Konstantin Yakovlev
ECMLPKDD Workshop on Automating Data Science
Abs HTML Bibtex

The paper investigates the problem of deep learning model selection. The authors propose a method of a neural architecture search with respect to its desired complexity. As a complexity, we consider a number of parameters that use selected architecture. The method is based on a differential architecture search algorithm (DARTS). Instead of optimizing structural parameters of the architecture, we consider them as a function depending on the complexity parameter. To evaluate the quality of the proposed algorithm we conduct experiments on a Fashion-MNIST dataset and compare the resulting architecture with DARTS method
```
@unpublished{yakovlev2021autods,
  author = {Yakovlev, Konstantin et al.},
  author_en = {Konstantin Yakovlev et al.},
  presenter = {Яковлев К. Д.},
  presenter_en = {Konstantin Yakovlev},
  title = {Automated architecture search with model complexity control},
  note = {ECMLPKDD Workshop on Automating Data Science},
  year = {2021},
  html = {https://sites.google.com/view/autods}
}
```

Multi-Modeling and Deep Learning Model Selection

Oleg Bakhteev, Vadim Strijov

2021

Speaker: Oleg Bakhteev

MIPT-UGA AI workshop

HTML Supp Bibtex

@unpublished{bakhteev2021uga,
  author = {Bakhteev, Oleg and Strijov, Vadim},
  presenter = {Бахтеев О. Ю.},
  presenter_en = {Oleg Bakhteev},
  title = {Multi-Modeling and Deep Learning Model Selection},
  note = {MIPT-UGA AI workshop},
  year = {2021},
  html = {https://sites.google.com/view/mipt-uga-ai-workshop/program},
  supp = {https://www.youtube.com/watch?v=LU6-6O5KHyA}
}

2019

NeurIPS
CrossLang: The System of Cross-lingual Plagiarism Detection

Oleg Bakhteev, Aleksandr Ogaltsov, Andrey Khazov, Kamil Safin, Rita Kuznetsova

2019
NeurIPS, Workshop on Systems for ML
HTML Bibtex
```
@unpublished{bakhteev2019nips2,
  abbr = {NeurIPS},
  author = {Bakhteev, Oleg and Ogaltsov, Aleksandr and Khazov, Andrey and Safin, Kamil and Kuznetsova, Rita},
  title = {CrossLang: The System of Cross-lingual Plagiarism Detection},
  note = {NeurIPS, Workshop on Systems for ML},
  year = {2019},
  html = {https://nips.cc/Conferences/2019/ScheduleMultitrack?event=13183}
}
```
NeurIPS
CrossLang: The System of Cross-lingual Plagiarism Detection

Oleg Bakhteev, Aleksandr Ogaltsov, Andrey Khazov, Kamil Safin, Rita Kuznetsova

2019

Speaker: Aleksandr Ogaltsov
NeurIPS, Document Intelligence workshop
HTML Bibtex
```
@unpublished{bakhteev2019nips1,
  abbr = {NeurIPS},
  presenter = {Огальцов А. В.},
  presenter_en = {Aleksandr Ogaltsov},
  author = {Bakhteev, Oleg and Ogaltsov, Aleksandr and Khazov, Andrey and Safin, Kamil and Kuznetsova, Rita},
  title = {CrossLang: The System of Cross-lingual Plagiarism Detection},
  note = {NeurIPS, Document Intelligence workshop},
  year = {2019},
  html = {https://nips.cc/Conferences/2019/ScheduleMultitrack?event=13178}
}
```
KDD
CrossLang: The System of Cross-lingual Plagiarism Detection

Oleg Bakhteev, Aleksandr Ogaltsov, Andrey Khazov, Kamil Safin, Rita Kuznetsova

2019

Speaker: Kamil Safin
KDD, Truth Discovery and Fact Checking: Theory and Practice workshop
HTML Bibtex
```
@unpublished{bakhteev2019kdd2,
  abbr = {KDD},
  presenter = {Сафин К. Ф.},
  presenter_en = {Kamil Safin},
  author = {Bakhteev, Oleg and Ogaltsov, Aleksandr and Khazov, Andrey and Safin, Kamil and Kuznetsova, Rita},
  title = {CrossLang: The System of Cross-lingual Plagiarism Detection},
  note = {KDD, Truth Discovery and Fact Checking: Theory and Practice workshop},
  year = {2019},
  html = {https://truth-discovery-kdd2019.github.io/}
}
```
KDD
CrossLang: The System of Cross-lingual Plagiarism Detection

Oleg Bakhteev, Aleksandr Ogaltsov, Andrey Khazov, Kamil Safin, Rita Kuznetsova

2019

Speaker: Kamil Safin
KDD, Workshop on Deep Learning for Education (DL4Ed)
HTML PDF Bibtex
```
@unpublished{bakhteev2019kdd1,
  abbr = {KDD},
  presenter = {Сафин К. Ф.},
  presenter_en = {Kamil Safin},
  author = {Bakhteev, Oleg and Ogaltsov, Aleksandr and Khazov, Andrey and Safin, Kamil and Kuznetsova, Rita},
  title = {CrossLang: The System of Cross-lingual Plagiarism Detection},
  note = {KDD, Workshop on Deep Learning for Education (DL4Ed)},
  year = {2019},
  html = {http://ml4ed.cc/2019-kdd-workshop/},
  pdf = {http://ml4ed.cc/attachments/Bakhteev.pdf}
}
```

2018

NeurIPS

Variational Bi-domain Triplet Autoencoder

Rita Kuznetsova, Oleg Bakhteev

2018

Speaker: Rita Kuznetsova

NeurIPS, Visually Grounded Interaction and Language workshop

HTML Bibtex

@unpublished{kuznetsova2018nips3,
  abbr = {NeurIPS},
  presenter = {Кузнецова М. В.},
  presenter_en = {Rita Kuznetsova},
  author = {Kuznetsova, Rita and Bakhteev, Oleg},
  title = {Variational Bi-domain Triplet Autoencoder},
  note = {NeurIPS, Visually Grounded Interaction and Language workshop},
  year = {2018},
  html = {https://nips2018vigil.github.io/}
}

NeurIPS

Variational Bi-domain Triplet Autoencoder

Rita Kuznetsova, Oleg Bakhteev

2018

Speaker: Rita Kuznetsova

NeurIPS, Visually Grounded Interaction and Language workshop

HTML Bibtex

@unpublished{kuznetsova2018nips2,
  abbr = {NeurIPS},
  presenter = {Кузнецова М. В.},
  presenter_en = {Rita Kuznetsova},
  author = {Kuznetsova, Rita and Bakhteev, Oleg},
  title = {Variational Bi-domain Triplet Autoencoder},
  note = {NeurIPS, Visually Grounded Interaction and Language workshop},
  year = {2018},
  html = {https://nips2018vigil.github.io/}
}

NeurIPS

Variational Bi-domain Triplet Autoencoder

Rita Kuznetsova, Oleg Bakhteev

2018

Speaker: Rita Kuznetsova

NeurIPS, Relational Representation Learning workshop

HTML Bibtex

@unpublished{kuznetsova2018nips1,
  abbr = {NeurIPS},
  presenter = {Кузнецова М. В.},
  presenter_en = {Rita Kuznetsova},
  author = {Kuznetsova, Rita and Bakhteev, Oleg},
  title = {Variational Bi-domain Triplet Autoencoder},
  note = {NeurIPS, Relational Representation Learning workshop},
  year = {2018},
  html = {https://nips.cc/Conferences/2018/Schedule?showEvent=10927}
}

KDD

Variational Bi-domain Triplet Autoencoder

Rita Kuznetsova, Oleg Bakhteev

2018

KDD Deep learning day workshop

PDF Bibtex

@unpublished{kuznetsova2018kdd2,
  abbr = {KDD},
  author = {Kuznetsova, Rita and Bakhteev, Oleg},
  title = {Variational Bi-domain Triplet Autoencoder},
  note = {KDD Deep learning day workshop},
  year = {2018},
  pdf = {https://www.kdd.org/kdd2018/files/deep-learning-day/DLDay18_paper_22.pdf}
}

KDD
ParaPlagDet: The system of paraphrased plagiarism detection

Rita Kuznetsova, Oleg Bakhteev, Andrey Khazov, Aleksandr Ogaltsov

2018

Speaker: Rita Kuznetsova
KDD BigScholar workshop
HTML Bibtex
```
@unpublished{kuznetsova2018kdd1,
  abbr = {KDD},
  presenter = {Кузнецова М. В.},
  presenter_en = {Rita Kuznetsova},
  author = {Kuznetsova, Rita and Bakhteev, Oleg and Khazov, Andrey and Ogaltsov, Aleksandr},
  title = {ParaPlagDet: The system of paraphrased plagiarism detection},
  note = {KDD BigScholar workshop},
  year = {2018},
  html = {http://thealphalab.org/bigscholar/2018/Program.htm}
}
```

Machine learning methods for fiscal data analysis [In Russian]

Методы машинного обучения для анализа фискальных данных

Oleg Bakhteev

2018

Workshop at Educational Сenter "Sirius", Russia

Bibtex

@unpublished{сириус2018,
  author = {Бахтеев, О. Ю.},
  author_en = {Oleg Bakhteev},
  title = {Методы машинного обучения для анализа фискальных данных},
  title_en = {Machine learning methods for fiscal data analysis},
  note_en = {Workshop at Educational Сenter "Sirius", Russia},
  note = {Семинар школы Сириус},
  year = {2018}
}

2015

RuSSIR
Explicit Semantic Analysis for Cross-Language Retrieval in Case of Russian-English Translation

Oleg Bakhteev, Alexey Romanov, Rita Kuznetsova

2015
Russian summer school in information retrieval, poster session
HTML Bibtex
```
@unpublished{bakhteev2015CLESA,
  abbr = {RuSSIR},
  author = {Bakhteev, Oleg and Romanov, Alexey and Kuznetsova, Rita},
  title = {Explicit Semantic Analysis for Cross-Language Retrieval in Case of Russian-English Translation},
  note = {Russir, постерная сессия},
  note_en = {Russian summer school in information retrieval, poster session},
  year = {2015},
  html = {http://romip.ru/russir2015/section.php?id=229}
}
```

Other

2024

Medium
Godot, HTML5, and Neural Networks: Yet Another Way to Reinvent the Wheel

Oleg Bakhteev

2024
Blog-post
Abs HTML Bibtex

Godot is an awesome and a very rapidly growing game engine allowing thousands of developers to implement their ideas to life — ranging from simple pet-projects to quite mature commercial games. However, like any emerging technology, Godot still has areas that need improvement. One such area is the integration of deep learning and neural networks.
```
@unpublished{embed2discover,
  abbr = {Medium},
  author = {Bakhteev, Oleg},
  title = {Godot, HTML5, and Neural Networks: Yet Another Way to Reinvent the Wheel},
  html = {https://medium.com/@bahleg/45461d56d0b5},
  note_en = {Blog-post},
  note = {Блог пост},
  year = {2024}
}
```
OSF
Introducing embed2discover: A tool for semi-automated, dictionary-based content-analysis

Laurence Brandenberger, Oleg Bakhteev, Jorge M. Fernandez, Sophia Schlosser, Luis Salamanca

2024

PDF Bibtex
```
@unpublished{embed2discoves,
  abbr = {OSF},
  author = {Brandenberger, Laurence and Bakhteev, Oleg and Fernandez, Jorge M. and Schlosser, Sophia and Salamanca, Luis},
  title = {Introducing embed2discover: A tool for semi-automated, dictionary-based content-analysis},
  pdf = {https://files.osf.io/v1/resources/gsmqf/providers/osfstorage/668c516cede75d063e693a81?action=download&direct&version=2},
  year = {2024}
}
```

Student

Predictive Models for Motor Pattern Recognition

Johannes Kjær

2024

Teaching assistants at student project at EPFL: Oleg Bakhteev and Leonid Iosipoi. Professor: Guillaume Obozinski

Bibtex

@phdthesis{kjaerproject,
  abbr = {Student},
  author = {Kjær, Johannes},
  title = {Predictive Models for Motor Pattern Recognition},
  note = {Teaching assistants at student project at EPFL: Oleg Bakhteev and Leonid Iosipoi. Professor: Guillaume Obozinski},
  year = {2024}
}

Student

Representation learning for motor pattern recognition during Inertial Measurement Unit

Daria Yakovchuk

2024

Teaching assistants at student project at EPFL: Oleg Bakhteev and Leonid Iosipoi. Professor: Guillaume Obozinski

Bibtex

@phdthesis{yakovhukproject,
  abbr = {Student},
  author = {Yakovchuk, Daria},
  title = {Representation learning for motor pattern recognition during Inertial Measurement Unit},
  note = {Teaching assistants at student project at EPFL: Oleg Bakhteev and Leonid Iosipoi. Professor: Guillaume Obozinski},
  year = {2024}
}

2023

Project

EvolvingDemocraSci: Advancing parliamentary data analysis

2023

A project page at Swiss Data science center

HTML Bibtex

@unpublished{evolvingdemocrasciproject,
  abbr = {Project},
  title = {EvolvingDemocraSci: Advancing parliamentary data analysis},
  note = {A project page at Swiss Data science center},
  html = {https://www.datascience.ch/projects/evolvingdemocrasci},
  year = {2023}
}

Student

LLM4SciLit - Large Language Models for Information Retrieval in Scientific Literature

Tommaso Martorella

2023

Teaching assistant at student project at EPFL: Oleg Bakhteev. Professor: Guillaume Obozinski

Bibtex

@phdthesis{martorellaproject,
  abbr = {Student},
  author = {Martorella, Tommaso},
  title = {LLM4SciLit - Large Language Models for Information Retrieval in Scientific Literature},
  note = {Teaching assistant at student project at EPFL: Oleg Bakhteev. Professor: Guillaume Obozinski},
  year = {2023}
}

Project

STIMO: Personalized epidural electrical stimulation of the lumbar spinal cord for clinically applicable therapy to restore mobility after paralyzing spinal cord injury,

2023

A project page at Swiss Data science center

HTML Bibtex

@unpublished{stimoproject,
  abbr = {Project},
  title = {STIMO: Personalized epidural electrical stimulation of the lumbar spinal cord for clinically applicable therapy to restore mobility after paralyzing spinal cord injury,},
  note = {A project page at Swiss Data science center},
  html = {https://www.datascience.ch/projects/stimo},
  year = {2023}
}

2022

Project

Software for air quality assessment and calculation of atmospheric transport processes [In Russian]

Программное обеспечение для оценки качества атмосферного воздуха и расчетов процессов переноса в атмосфере

2022

A project page at Skoltech

HTML Bibtex

@unpublished{rppproject,
  abbr = {Project},
  title = {Программное обеспечение для оценки качества атмосферного воздуха и расчетов процессов переноса в атмосфере},
  title_en = {Software for air quality assessment and calculation of atmospheric transport processes},
  note_en = {A project page at Skoltech},
  note = {Страница проекта на сайте Сколтеха},
  html = {https://raic.skoltech.ru/po_rpp},
  year = {2022}
}

BSc

Metaparameter optimization in knowledge distillation problem, Bachelor thesis, MIPT [In Russian]

Оптимизация метапараметров в задаче дистилляции знаний, бакалаврский диплом, МФТИ

Maria Gorpinich

2022

Scientific adviser: Oleg Bakhteev, PhD

Supp Bibtex

@phdthesis{gorpinichbsc,
  abbr = {BSc},
  author_en = {Maria Gorpinich},
  author = {Мария, Горпинич},
  title_en = {Metaparameter optimization in knowledge distillation problem, Bachelor thesis, MIPT},
  title = {Оптимизация метапараметров в задаче дистилляции знаний, бакалаврский диплом, МФТИ},
  note_en = {Scientific adviser: Oleg Bakhteev, PhD},
  note = {Научный руководитель: к.ф.-м.н.  Бахтеев О.Ю.},
  supp = {https://youtu.be/mmAacGSUvPQ?t=823},
  year = {2022}
}

BSc

Concordant neural model selection with complexity control, Bachelor thesis at MIPT [In Russian]

Выбор согласованных нейросетевых моделей с контролем сложности, бакалаврский диплом, МФТИ

Konstantin Yakovlev

2022

Scientific adviser: Oleg Bakhteev, PhD

Supp Bibtex

@phdthesis{yakovlevbsc,
  abbr = {BSc},
  author_en = {Konstantin Yakovlev},
  author = {Константин, Яковлев},
  title_en = {Concordant neural model selection with complexity control, Bachelor thesis at MIPT},
  title = {Выбор согласованных нейросетевых моделей с контролем сложности, бакалаврский диплом, МФТИ},
  note_en = {Scientific adviser: Oleg Bakhteev, PhD},
  note = {Научный руководитель: к.ф.-м.н.  Бахтеев О.Ю.},
  supp = {https://youtu.be/mmAacGSUvPQ},
  year = {2022}
}

MSc
Bayesian neural architecture model selectionб Master thesis at MIPT [In Russian]

Байесовский выбор архитектуры нейросетевой модели, диплом магистра, МФТИ

Anton Sotnikov

2022
Scientific adviser: Oleg Bakhteev, PhD
Abs Bibtex

В работе исследуется задача байесовского выбора архитектуры нейросетевой модели. Предлагается метод, оценивающий апостериорное совместное распределение структуры и параметров модели с помощью вариационной нижней оценки обоснованности. Вводятся априорные предположения о распределении параметров и структуры модели. Показано, что байесовский подход регуляризует норму гессиана оптимизируемой функции ошибки по структуре модели. Для оценки качества и устойчивости предложенного метода проводится вычислительный эксперимент на выборке Fashion-MNIST. Показано, что предлагаемый метод является более устойчивым в сравнении с базовыми градиентными методами выбора структуры.
```
@phdthesis{sotnikovmc,
  abbr = {MSc},
  author_en = {Anton Sotnikov},
  author = {Антон, Сотников},
  title_en = {Bayesian neural architecture model selectionб Master thesis at MIPT},
  title = {Байесовский выбор архитектуры нейросетевой модели, диплом магистра, МФТИ},
  note_en = {Scientific adviser: Oleg Bakhteev, PhD},
  note = {Научный руководитель: к.ф.-м.н.  Бахтеев О.Ю.},
  year = {2022}
}
```
MSc
An adversarial method for neural network fine-tuning for transfer learning problem, Master thesis at MIPT

Aleksandr Kolesov

2022
Scientific adviser: Oleg Bakhteev, PhD
Abs PDF Supp Bibtex

Inductive transfer learning aims at adapting a pre-trained neural network to target data without access to a source database. Most modern methods suffer from strong regularization between corresponding parameters. An eﬃcient inductive transfer learning method should take enough information from a pre-trained model, but also learn vital patterns from new data. We present the novel technique of transfer learning that is based on optimal transport theory. However, this method requires deep neural networks(DNN) that are strongly 1-Lipschitz continuity functions in according to the theory. To solve this issue, we develop such networks and prove necessary thereotically guarantees for them. We show, that the regularization of the proposed method computes ground-truth the Wasserstein-1 distance between distributions of parameters, while another penalization strategies compute a upper bound. We demonstrate the abilities of the proposed method and compare the method with relevant transfer learning techniques.
```
@phdthesis{kolesovmsc,
  abbr = {MSc},
  author_en = {Aleksandr Kolesov},
  author = {Александр, Колесов},
  title_en = {An adversarial method for neural network fine-tuning for transfer learning problem, Master thesis at MIPT},
  title = {Состязательный метод дообучения нейронной сети в задаче переноса информации, диплом магистра, МФТИ},
  note_en = {Scientific adviser: Oleg Bakhteev, PhD},
  note = {Научный руководитель: к.ф.-м.н.  Бахтеев О.Ю.},
  pdf = {https://github.com/Intelligent-Systems-Phystech/Kolesov-MS-thesis/raw/main/Thesis_Kolesov.pdf},
  year = {2022},
  supp = {https://youtu.be/f4C9U59krTE?t=6736},
  bilingual = {True}
}
```

2021

BSc
An investigation of gradient-based methods for neural architecture search, Bachelor thesis at MIPT [In Russian]

Исследование методов поиска структур нейронной сети на основе градиентного поиска, бакалаврский диплом, МФТИ

Valeria Sherbakova

2021
Scientific consultant: Oleg Bakhteev, PhD
Abs Bibtex

В настоящее время одним из активно развивающихся направлений глубокого обучения является поиск нейронной архитектуры. Глубокое обучение позволило за последние годы добиться значительного прогресса в решении множества задач, таких как классификация изображений, распознавание речи и машинный перевод. Одним из важнейших аспектов этого прогресса являются новые нейронные архитектуры. Используемые в настоящее время архитектуры в основном разрабатывались вручную людьми, что является трудоемким и подверженным ошибкам процессом. В связи с этим растет интерес к методам поиска автоматизированной нейронной архитектуры. В данной работе исследуются методы сэмплирования структуры в градиентных методах поиска нейронной архитектуры. Современные методы поиска нейронной архитектуры, такие как DARTS и SNAS, интерпретируют структуру глубокой модели как граф с ребрами, соответствующими нелинейным функциям в нем. Для обучения различных моделей с использованием градиентного подхода вводим вероятностную интерпретацию для этих ребер. Рассмотрим различные распределения, такие как распределения Softmax, Gumbel-Softmax и Invertible Gaussian Reparameterization для них. Проанализируем производительность полученных нейронных сетей и их устойчивость к прунингу и атакам противника. Вычислительные эксперименты проводятся с использованием набора данных MNIST.
```
@phdthesis{sherbakovabsc,
  abbr = {BSc},
  author_en = {Valeria Sherbakova},
  author = {Валерия, Щербакова},
  title_en = {An investigation of gradient-based methods for neural architecture search, Bachelor thesis at MIPT},
  title = {Исследование методов поиска структур нейронной сети на основе градиентного поиска, бакалаврский диплом, МФТИ},
  note_en = {Scientific consultant: Oleg Bakhteev, PhD},
  note = {Научный консультант: к.ф.-м.н.  Бахтеев О.Ю.},
  year = {2021}
}
```
BSc
Model generation with complexity control using Bayesian hypernetworks, Bachelor thesis at MIPT [In Russian]

Порождение моделей заданной сложности с использованием байесовских гиперсетей, бакалаврский диплом, МФТИ

Olga Grebenkova

2021
Scientific adviser: Oleg Bakhteev, PhD
Abs Supp Bibtex

The paper is devoted to deep learning model complexity control. The complexity is an amount of the information that is required to transfer information about the model and the dataset. We consider a compexity of the model as a parameter that can be set during model inference step based on computational budget and other operational requirements for the model. In order to control the model complexity we introduce a probabilistic assumptions about the distribution of parameters of the deep learning model. The paper instigates three forms of the regularization that allow to control the model parameter distribution. We consider the model evidence as a conditional value that depends on the required model complexity. The proposed method is based on the representation of deep learning model parameters in the form of hypernetwork output. A hypernetwork is a model that generates parameters of an optimal model. We analyze this method in the computational experiments on the MNIST and CIFAR datasets.
```
@phdthesis{grebenkovabsc,
  abbr = {BSc},
  author_en = {Olga Grebenkova},
  author = {Ольга, Гребенькова},
  title_en = {Model generation with complexity control using Bayesian hypernetworks, Bachelor thesis at MIPT},
  title = {Порождение моделей заданной сложности с использованием байесовских гиперсетей, бакалаврский диплом, МФТИ},
  note_en = {Scientific adviser: Oleg Bakhteev, PhD},
  note = {Научный руководитель: к.ф.-м.н.  Бахтеев О.Ю.},
  supp = {https://youtu.be/Jn7jChIBszw?t=3253},
  year = {2021}
}
```

2020

PhD

Bayesian suboptimal deep learning structure selection, PhD thesis [In Russian]

Байесовский выбор субоптимальной структуры модели глубокого обучения, диссертация к.ф.-м. н.

Oleg Bakhteev

2020

Scientific adviser: Vadim Strijov, DSc

HTML PDF Bibtex

@phdthesis{phdthesis,
  abbr = {PhD},
  author_en = {Oleg Bakhteev},
  author = {Бахтеев, Олег},
  title_en = {Bayesian suboptimal deep learning structure selection, PhD thesis},
  title = {Байесовский выбор субоптимальной структуры модели глубокого обучения, диссертация к.ф.-м. н. },
  html = {http://www.frccsc.ru/diss-council/00207305/diss/list/bahteev_oy},
  pdf = {http://www.frccsc.ru/sites/default/files/docs/ds/002-073-05/diss/26-bahteev/ds05-26-bahteev_main.pdf?28},
  note_en = {Scientific adviser: Vadim Strijov, DSc},
  note = {Научный руководитель: д.ф.-м.н. Стрижов В. В.},
  year = {2020}
}

arXiv
Variational learning across domains with triplet information

Rita Kuznetsova, Oleg Bakhteev, Alexandr Ogaltsov

2020

arXiv Bibtex
```
@unpublished{VBTA,
  abbr = {arXiv},
  author = {Kuznetsova, Rita and Bakhteev, Oleg and Ogaltsov, Alexandr},
  title = {Variational learning across domains with triplet information},
  arxiv = {1806.08672},
  year = {2020}
}
```

Habr

Klingon language tutorial [In Russian]

Самоучитель клингонского

Antiplagiat company

2020

It-blog. Article co-author.

HTML Bibtex

@unpublished{habr_klin,
  abbr = {Хабр},
  author = {Антиплагиат},
  title_en = {Klingon language tutorial},
  title = {Самоучитель клингонского},
  note = {Хабр. Соавтор статьи},
  year = {2020},
  html = {https://habr.com/ru/company/antiplagiat/blog/507848/},
  abbr_en = {Habr},
  author_en = {Antiplagiat company},
  note_en = {It-blog. Article co-author.}
}

2019

Software

Cross-lingual textual reuse detection module for English-Russian language pair [In Russian]

Модуль поиска переводных текстовых заимствований с русского на английский язык

Oleg Bakhteev et al.

2019

Software registration certificate

HTML Bibtex

@unpublished{VBTB,
  abbr = {ПО},
  abbr_en = {Software},
  author = {Бахтеев, О. Ю. and и, др.},
  author_en = {Oleg Bakhteev et al.},
  title = {Модуль поиска переводных текстовых заимствований с русского на английский язык},
  title_en = {Cross-lingual textual reuse detection module for English-Russian language pair},
  note = {Свидетельство о регистрации ПО},
  note_en = {Software registration certificate},
  year = {2019},
  html = {https://www.elibrary.ru/item.asp?id=41532315}
}

2018

Habr

How Antiplagiat detects paraphrased text [In Russian]

«Трое в лодке, нищета и собаки», или как Антиплагиат ищет парафраз

Antiplagiat company

2018

It-blog. Article co-author.

HTML Bibtex

@unpublished{habr_troe,
  abbr = {Хабр},
  author = {Антиплагиат},
  title_en = {How Antiplagiat detects paraphrased text},
  title = {«Трое в лодке, нищета и собаки», или как Антиплагиат ищет парафраз},
  note = {Хабр. Соавтор статьи},
  year = {2018},
  html = {https://habr.com/ru/company/antiplagiat/blog/422941/},
  abbr_en = {Habr},
  author_en = {Antiplagiat company},
  note_en = {It-blog. Article co-author.}
}

Habr

An overview of autoencoders application in text analysis [In Russian]

«Туда и обратно» для нейронных сетей, или обзор применений автокодировщиков в анализе текстов

Antiplagiat company

2018

It-blog. Article co-author.

HTML Bibtex

@unpublished{habr_tuda,
  abbr = {Хабр},
  author = {Антиплагиат},
  title = {«Туда и обратно» для нейронных сетей, или обзор применений автокодировщиков в анализе текстов},
  title_en = {An overview of autoencoders application in text analysis},
  note = {Хабр. Соавтор статьи},
  year = {2018},
  html = {https://habr.com/ru/company/antiplagiat/blog/418173/},
  abbr_en = {Habr},
  author_en = {Antiplagiat company},
  note_en = {It-blog. Article co-author.}
}

Habr

Challenges in translation: how to find a cross-lingual plagiarism from Russian into English [In Russian]

Трудности перевода: как найти плагиат с английского языка в русских научных статьях

Antiplagiat company

2018

It-blog. Article co-author.

HTML Bibtex

@unpublished{habr_tudb,
  abbr = {Хабр},
  author = {Антиплагиат},
  title = {Трудности перевода: как найти плагиат с английского языка в русских научных статьях},
  title_en = {Challenges in translation: how to find a cross-lingual plagiarism from Russian into English},
  note = {Хабр. Соавтор статьи},
  year = {2018},
  html = {https://habr.com/ru/company/antiplagiat/blog/354142/},
  abbr_en = {Habr},
  author_en = {Antiplagiat company},
  note_en = {It-blog. Article co-author.}
}

Datasets and supplementary materials

2023

Dataset

Dataset of handwritten essays, 2021 (reuploaded)

Oleg Bakhteev et al.

2023

Supplementary material for the article "Near-duplicate handwritten document detection without text recognition"

HTML Bibtex

@misc{handwr_supp,
  abbr = {Dataset},
  author = {Bakhteev, Oleg et al.},
  author_en = {Oleg Bakhteev et al.},
  title = {Dataset of handwritten essays, 2021 (reuploaded)},
  year = {2023},
  doi = {10.17632/vs44v8r3nf.1},
  html = {https://data.mendeley.com/datasets/vs44v8r3nf/1},
  note = {Supplementary material for the article "Near-duplicate handwritten document detection without text recognition"}
}

2022

Dataset

Cross-lingual dataset, 2019 (reuploaded)

Oleg et al. Bakhteev

2022

Supplementary material for the article "CrossLang: the system of cross-lingual plagiarism detection"

HTML Bibtex

@misc{neurip_supp,
  abbr = {Dataset},
  author = {Bakhteev, Oleg et al.},
  title = {Cross-lingual dataset, 2019 (reuploaded)},
  year = {2022},
  html = {https://disk.yandex.ru/d/2UI8veyxSOXRdw},
  note = {Supplementary material for the article "CrossLang: the system of cross-lingual plagiarism detection"}
}

Dataset

Results of cross-lingual text reuse detection among European Universities

Oleg Bakhteev et al.

2022

Supplementary material for the article "Cross-language plagiarism detection: a case study of European languages academic works"

HTML Bibtex

@misc{enai_supp1,
  abbr = {Dataset},
  author = {Bakhteev, Oleg et al.},
  author_en = {Oleg Bakhteev et al.},
  title = {Results of cross-lingual text reuse detection among European Universities},
  year = {2022},
  html = {https://data.mendeley.com/datasets/t726dmtx24/2},
  note = {Supplementary material for the article "Cross-language plagiarism detection: a case study of European languages academic works"}
}

Dataset

Synthetic dataset for cross-lingual text reuse detection evaluation

Oleg Bakhteev et al.

2022

Supplementary material for the article "Cross-language plagiarism detection: a case study of European languages academic works"

HTML Bibtex

@misc{enai_supp2,
  abbr = {Dataset},
  author = {Bakhteev, Oleg et al.},
  author_en = {Oleg Bakhteev et al.},
  title = {Synthetic dataset for cross-lingual text reuse detection evaluation},
  year = {2022},
  html = {https://data.mendeley.com/datasets/53j6hyn99g},
  note = {Supplementary material for the article "Cross-language plagiarism detection: a case study of European languages academic works"}
}

2021

Dataset

Bibliography dataset (reuploaded)

Aleksandr Ogaltsov

2021

Not my paper! Supplementary material for the article "Language-Free Regular Expression Search of Document’s References"

HTML Bibtex

@misc{bib_open,
  abbr = {Dataset},
  author = {Ogaltsov, Aleksandr},
  title = {Bibliography dataset (reuploaded)},
  year = {2021},
  html = {https://disk.yandex.ru/d/Djr73ITnkh9eUQ},
  note = {Not my paper! Supplementary material for the article "Language-Free Regular Expression Search of Document’s References"}
}

Dataset
Open access scientific documents from elibrary.ru

Andrey Grabovoy, Oleg Bakhteev, Yury Chekhovich

2021
Supplementary material for the article "The automatic approach for scientific papers dating"
HTML Bibtex
```
@misc{elib_open,
  abbr = {Dataset},
  author = {Grabovoy, Andrey and Bakhteev, Oleg and Chekhovich, Yury},
  title = {Open access scientific documents from elibrary.ru},
  year = {2021},
  html = {https://data.mendeley.com/datasets/95xt9sngzc},
  note = {Supplementary material for the article "The automatic approach for scientific papers dating"}
}
```