Publications

This article is devoted to the issue of developing adaptive learning systems for vocational education and training (VET). Firstly, it justifies the urgency of developing and using personalized adaptive learning in vocational educational organizations. Specific features of the Russian VET system and its students are described, demonstrating a number of arguments for the importance of a search for new digital educational solutions. Secondly, the paper elaborates on the theoretical framework of personalization of vocational education and training, which takes into account the necessity for both skills and knowledge. Finally, the authors present a prototype of an adaptive educational system, which is based on ontologically-controlled management of learning trajectories. The developed software is aimed at improving the effectiveness of the VET material science curriculum.

The paper addresses the questions of data science education of current importance. It aims to introduce and justify the framework that allows flexibly evaluate the processes of a data expedition and a digital media created during it. For these purposes, the authors explore features of digital media artefacts which are specific to data expeditions and are essential to accurate evaluation. The rubrics as a power but hardly formalizable evaluation method in application to digital media artefacts are also discussed. Moreover, the paper documents the experience of rubrics creation according to the suggested framework. The rubrics were successfully adopted to two data-driven journalism courses. The authors also formulate recommendations on data expedition evaluation which should take into consideration structural features of a data expedition, distinctive features of digital media, etc.

Recurrent neural networks have proved to be an effective method for statistical language modeling. However, in practice their memory and run-time complexity are usually too large to be implemented in real-time offline mobile applications. In this paper we consider several compression techniques for recurrent neural networks including Long–Short Term Memory models. We make particular attention to the high-dimensional output problem caused by the very large vocabulary size. We focus on effective compression methods in the context of their exploitation on devices: pruning, quantization, and matrix decomposition approaches (low-rank factorization and tensor train decomposition, in particular). For each model we investigate the trade-off between its size, suitability for fast inference and perplexity. We propose a general pipeline for applying the most suitable methods to compress recurrent neural networks for language modeling. It has been shown in the experimental study with the Penn Treebank (PTB) dataset that the most efficient results in terms of speed and compression–perplexity balance are obtained by matrix decomposition techniques.

 

We present a study on co-authorship network representation based on network embedding together with additional information on topic modeling of research papers and new edge embedding operator. We use the link prediction (LP) model for constructing a recommender system for searching collaborators with similar research interests. Extracting topics for each paper, we construct keywords co-occurrence network and use its embedding for further generalizing author attributes. Standard graph feature engineering and network embedding methods were combined for constructing co-author recommender system formulated as LP problem and prediction of future graph structure. We evaluate our survey on the dataset containing temporal information on National Research University Higher School of Economics over 25 years of research articles indexed in Russian Science Citation Index and Scopus. Our model of network representation shows better performance for stated binary classification tasks on several co-authorship networks.

There are many different methods for computing relevant patterns in sequential data and interpreting the results. In this paper, we compute emerging patterns (EP) in demographic sequences using sequence-based pattern structures, along with different algorithmic so- lutions. The purpose of this method is to meet the following domain requirement: the obtained patterns must be (closed) frequent contiguous prefixes of the input sequences. This is required in order for demogra- phers to fully understand and interpret the results.

Pattern Mining (PM) has a prominent place in Data Science and finds its application in a wide range of domains. To avoid the exponential explosion of patterns different methods have been proposed. They are based on assumptions on interestingness and usually return very different pattern sets. In this paper, we propose to use a compression-based objective as a well-justified and robust interestingness measure. We define the description lengths for datasets and use the Minimum Description Length principle (MDL) to find patterns that ensure the best compression. Our experiments show that the application of MDL to numerical data provides a small and characteristic subset of patterns describing data in a compact way.

Proceedings of Third Workshop "Computational linguistics and language science"

These day adaptivity is the cutting edge of modern education. Technologies are being developed rapidly and bringing new possibilities to educators. Thus, diverse types of adaptive learning environment have appeared during these last decades. Material Science and Engineering Education (MSEE) have a solid formalized foundation, which consists of standards, recommendations and clear rules. Moreover, investigators report on growing role of computer in teaching and learning in MSEE. These brings great perspectives to computer adaptive learning system based on a material science and engineering ontology. This paper aims to justify general pedagogical foundations of adaptivity and to collect requirements to a computer adaptive learning system. As an extra result we introduce the architecture of ontology-based adaptive learning system to MSEE.

Purpose

Favorable outcomes were achieved for children with acute lymphoblastic leukemia (ALL) with the first Russian multicenter trial Moscow–Berlin (ALL-MB) 91. One major component of this regimen included a total of 18 doses of weekly intramuscular (IM) native Escherichia coli-derived asparaginase (E. coli-ASP) at 10000 U/m2 during three consolidation courses. ASP was initially available from Latvia, but had to be purchased from abroad at substantial costs after the collapse of Soviet Union. Therefore, the subsequent trial ALL-MB 2002 aimed at limiting costs to a reasonable extent and also at reducing toxicity by lowering the dose for standard risk (SR−) patients to 5000 U/m2 without jeopardizing efficacy.

Methods

Between April 2002 and November 2006, 774 SR patients were registered in 34 centers across Russia and Belarus, 688 of whom were randomized. In arm ASP-5000 (n = 334), patients received 5000 U/m2 and in arm ASP-10000 (n = 354) 10 000 U/m2 IM.

Results

Probabilities of disease-free survival, overall survival and cumulative incidence of relapse at 10 years were comparable: 79 ± 2%, 86 ± 2% and 17.4 ± 2.1% (ASP-5000) vs. 75 ± 2% and 82 ± 2%, and 17.9 ± 2.0% (ASP-10000), while death in complete remission was significantly lower in arm ASP-5000 (2.7% vs. 6.5%; p = 0.029).

Conclusion

Our findings suggest that weekly 5000 U/m2E. coli-ASP IM during consolidation therapy are equally effective, more cost-efficient and less toxic than 10000 U/m2 for SR patients with childhood ALL.

We investigate the effect of the proportional hazards assumption on prognostic and predictive models of the survival time of patients suffering from amyotrophic lateral sclerosis (ALS). We theoretically compare the underlying model formulations of several variants of survival forests and implementations thereof, including random forests for survival, conditional inference forests, Ranger, and survival forests with L1 splitting, with two novel variants, namely distributional and transformation survival forests. Theoretical considerations explain the low power of log-rank-based splitting in detecting patterns in non-proportional hazards situations in survival trees and corresponding forests. This limitation can potentially be overcome by the alternative split procedures suggested herein. We empirically investigated this effect using simulation experiments and a re-analysis of the PRO-ACT database of ALS survival, giving special emphasis to both prognostic and predictive models.

We investigate the effect of the proportional hazards assumption on prognostic and predictive models of the survival time of patients suffering from amyotrophic lateral sclerosis (ALS). We theoretically compare the underlying model formulations of several variants of survival forests and implementations thereof, including random forests for survival, conditional inference forests, Ranger, and survival forests with L1 splitting, with two novel variants, namely distributional and transformation survival forests. Theoretical considerations explain the low power of log-rank-based splitting in detecting patterns in non-proportional hazards situations in survival trees and corresponding forests. This limitation can potentially be overcome by the alternative split procedures suggested herein. We empirically investigated this effect using simulation experiments and a re-analysis of the PRO-ACT database of ALS survival, giving special emphasis to both prognostic and predictive models.

Previously, the mathematical models (CoMPaS and CoM-III) of primary tumor (PT) growth and secondary distant metastases (sdMTS) growth of breast cancer (BC) considering TNM classification have been presented (Tyuryumina E., Neznanov A.; 2017, 2018). Goal: To detect the earliest diagnostics period of visible sdMTS via CoMPaS and CoM-III.

The paper addresses the questions of data science education of current importance. It aims to introduce and justify the framework that allows flexibly evaluate the processes of a data expedition and a digital media created during it. For these purposes, the authors explore features of digital media artefacts which are specific to data expeditions and are essential to accurate evaluation. The rubrics as a power but hardly formalizable evaluation method in application to digital media artefacts are also discussed. Moreover, the paper documents the experience of rubrics creation according to the suggested framework. The rubrics were successfully adopted to two data-driven journalism courses. The authors also formulate recommendations on data expedition evaluation which should take into consideration structural features of a data expedition, distinctive features of digital media, etc.

Nowadays education has been significantly transformed by the widespread use of digital solutions. This has made it popular to carry out the new approaches to educational processes, which cannot be easily implemented before. Оne of such forms is blended learning, and it becomes possible everywhere thanks to the global distribution and convergence of telecommunication technologies and to the appearance of cloud-based educational software. At the same time digital technologies have brought the apparent profit for the routines automatization and have become the reason of rethinking as in casual teaching and learning processes as in the modern educational management. This paper focuses on how to support educational processes with technologies. It is also answers several significant questions about effective blended learning support, especially at the administration level.

The paper deals with the direct and inverse bifurcation problems for the von Karman equations; the direct bifurcation problem implies that a bifur-cation (branching) structure of the equations is constructed, while the in-verse bifurcation problem is the problem to identify pre-bifurcation states that is to predict buckling. The paper considers the algorithms to solveboth direct and inverse problems; the algorithm for the inverse problem employs the one used to solve the direct problem. The results for cylindri-cal shell subject to external pressure and that one under axial compression are discussed.

In this paper, we provide the solution for RecSys Challenge 2018 by our Avito team, which obtained the 3rd place in main track. The goal of the competition was to recommend music tracks for automatic playlist continuation. As a part of this challenge, Spotify released a large public dataset, which allowed us to train a rather complex algorithm. Our approach consists of two stages: collaborative filtering for candidate selection and gradient boosting for final prediction. The combination of these two models performed well with the playlist and track metadata given.

Logical frameworks allow the specification of deductive systems using the same logical machinery. Linear logical frameworks have been successfully used for the specification of a number of computational, logics and proof systems. Its success relies on the fact that formulas can be distinguished as linear, which behave intuitively as resources, and unbounded, which behave intuitionistically. Commutative subexponentials enhance the expressiveness of linear logic frameworks by allowing the distinction of multiple contexts. These contexts may behave as multisets of formulas or sets of formulas. Motivated by applications in distributed systems and in type-logical grammar, we propose a linear logical framework containing both commutative and non-commutative subexponentials. Non-commutative subexponentials can be used to specify contexts which behave as lists, not multisets, of formulas. In addition, motivated by our applications in type-logical grammar, where the weakenening rule is disallowed, we investigate the proof theory of formulas that can only contract, but not weaken. In fact, our contraction is non-local. We demonstrate that under some conditions such formulas may be treated as unbounded formulas, which behave intuitionistically.