?
StochasticRank: Global Optimization of Scale-Free Discrete Functions
P. 9669–9679.
Liudmila Prokhorenkova, Ustimenko A.
In this paper, we introduce a powerful and efficient framework for direct optimization of ranking metrics. The problem is ill-posed due to the discrete structure of the loss, and to deal with that, we introduce two important techniques: stochastic smoothing and novel gradient estimate based on partial integration. We show that classic smoothing approaches may introduce bias and present a universal solution for a proper debiasing. Importantly, we can guarantee global convergence of our method by adopting a recently proposed Stochastic Gradient Langevin Boosting algorithm. Our algorithm is implemented as a part of the CatBoost gradient boosting library and outperforms the existing approaches on several learning-to-rank datasets. In addition to ranking metrics, our framework applies to any scale-free discrete loss function.
Ахмедов Э., Furmanov K. K., Экономика и математические методы 2025 Т. 61 № 1 С. 109–117
A stochastic frontier model is a regression model where an explained variable is either output of a firm or its costs, and unexplained variation of this variable is divided into two components: inefficiency and stochastic shock. These components are modeled by random variables with different families of distributions. The model allows estimation of inefficiency at firm level and ...
Added: March 27, 2025
Zelenkov Y., , in: 2023 Ivannikov ISPRAS Open Conference (ISPRAS).: IEEE, 2023. P. 176–182.
We propose a regression ensemble based on a decomposition that separates the weighted average errors of individual learners and the ambiguity of their estimates. This approach is a modification of Gradient Boosting with a variation of the gradient at each step. That allows ensuring explicitly a diversity of base estimators. In addition, the proposed approach ...
Added: May 1, 2024
Dubinina M., Вестник Государственного социально-гуманитарного университета 2021 № 4(44) С. 68–72
Ranking socially signi cant phenomena has become a wide-spread practice. The Chinese social credit initiative is a unique example of how the concept of social ranking can be used across public and private sector of the economy. Thus, the aim of this study is to show the synergy of traditional cultural attitudes, desire to preserve ...
Added: February 15, 2024
Bukina T. V., Kashin D., Экономический журнал Высшей школы экономики 2024 Т. 28 № 1 С. 81–107
The paper reveals the forecasts for regional inflation based on the regions of the Privolzhskiy Federal District (PFD). The purpose of the study is to determine the model that most accurately predicts regional inflation. The paper compares the tools of machine learning – support vector machines, gradient boosting, and random forest – with econometric models ...
Added: February 13, 2024
Никольский И. М., Furmanov K. K., Прикладная эконометрика 2023 Т. 71 С. 128–142
A technique for assessing the concordance between true inefficiency of decision making units in a basic stochastic frontier model and their JLMS estimates is proposed. An approximate formula for Harrell’s C-index is derived for the case of half-normal distribution of inefficiency component. A simulation study shows that the approximation error is about 0.01 and that ...
Added: September 19, 2023
Nikol’skii I. M., K. K. Furmanov, Computational Mathematics and Modeling 2022 Vol. 33 No. 3 P. 319–329
A stochastic frontier model is considered – a linear regression model linking the firm activity measure (e.g., the production volume of an enterprise) and the resource inputs. This model is used to determine the efficiency of decision-making units (DMUs). A series of stochastic experiments is run to determine the model’s ability to rank the DMUs ...
Added: April 24, 2023
Никольский И. М., Furmanov K. K., Прикладная математика и информатика 2022 № 71 С. 12–18
Рассматривается модель стохастической границы — линейная регрессионная модель, связывающая показатель деятельности субъекта принятия решения (например, объём производства предприятия) с затратами ресурсов и используемая для определения эффективности субъектов. На основании серии стохастических экспериментов определяется способность модели ранжировать субъекты по эффективности — рассчитываются коэффициент ранговой корреляции Спирмена и коэффициент согласованности Харрелла в зависимости от дисперсий двух компонент ...
Added: November 11, 2022
Muratova A., Mitrofanova E., Islam R., , in: Procedia Computer Science: 11th International Young Scientist Conference on Computational ScienceVol. 212.: Elsevier, 2022. P. 358–367.
The article presents a case study on demographic sequences analysis through modern machine learning (ML)
techniques. The studied data contains demographic and socioeconomic events, where the events are presented
as sequences of statuses. The involved demographers are interested in applications of advanced ML techniques
and interpretable patterns for their needs. We show how Shapley value-based explanations can be ...
Added: September 10, 2022
Knizhnikov A., Shvarts E., Ametistova L. et al., Extractive Industries and Society 2021 Vol. 8 No. 3 Article 100937
This article examines the current state and drivers of environmental transparency in the Russian Mining and Metals sector. The study is based upon 2016–19 successive annual rankings calculated transparently, using publicly available information and a third-party-verified ranking system. Ranking results reveal a definite increase in the transparency level of one of the most closed industries ...
Added: February 10, 2022
Glagoleva A., Kachesova I., Kuznetsova E. et al., , in: Modern Global Economic System: Evolutional Development vs. Revolutionary Leap:.: Springer, 2021. P. 457–464.
Added: October 8, 2021
Ustimenko A., Prokhorenkova L., , in: Proceedings of the 38th International Conference on Machine Learning (ICML 2021)Vol. 139.: PMLR, 2021. P. 1–10.
This paper introduces Stochastic Gradient Langevin Boosting (SGLB) - a powerful and efficient machine learning framework that may deal with a wide range of loss functions and has provable generalization guarantees. The method is based on a special form of the Langevin diffusion equation specifically designed for gradient boosting. This allows us to theoretically guarantee ...
Added: August 6, 2021
Малинин А. А., Prokhorenkova L., Ustimenko A., , in: Proceedings of the 9th International Conference on Learning Representations (ICLR 2021). ICLR, 2021.: ICLR, 2021..
Added: August 2, 2021
Ivanov S., Prokhorenkova L., , in: Proceedings of the 9th International Conference on Learning Representations (ICLR 2021). ICLR, 2021.: ICLR, 2021..
Added: August 2, 2021
Antipov E. A., Pokryshevskaya E. B., Journal of Revenue and Pricing Management 2020 No. 19 P. 355–364
Forecasting demand and understanding sales drivers are one of the most important tasks in retail analytics. However, traditionally, linear models and/or models with a small number of predictors have been predominantly used in sales modeling. Taking into account that real-world demand is naturally determined by complex substitution and complementation patterns among a large number of ...
Added: October 31, 2020
Liudmila Prokhorenkova, Gusev G., Vorobev A. et al., , in: Advances in Neural Information Processing Systems 31 (NeurIPS 2018).: Neural Information Processing Systems Foundation, 2018. P. 6638–6648.
This paper presents the key algorithmic techniques behind CatBoost, a new gradient boosting toolkit. Their combination leads to CatBoost outperforming other publicly available boosting implementations in terms of quality on a variety of datasets. Two critical algorithmic advances introduced in CatBoost are the implementation of ordered boosting, a permutation-driven alternative to the classic algorithm, and ...
Added: May 1, 2020
Timur Kadyrov, Ignatov D. I., , in: Proceedings of the Fifth International Workshop on Experimental Economics and Machine Learning (EEML 2019),Perm, Russia, September 26, 2019Vol. 2479.: CEUR Workshop Proceedings, 2019. P. 77–88.
A multichannel attribution model based on gradient boost- ing over trees is proposed, which was compared with the state of the art models: bagged logistic regression, Markov chains approach, shapely value. Experiments on digital advertising datasets showed that the pro- posed model is better than the solutions considered by ROC AUC metric. In addition, the ...
Added: January 20, 2020
Kuralenok I., Ershov V., Лабутин И. Н., , in: Advances in Neural Information Processing Systems 32 (NeurIPS 2019).: [б.и.], 2019. P. 1–10.
In this work, we introduce a new decision tree ensemble representation framework: instead of using a graph model we transform each tree into a well-known polynomial form. We apply the new representation to three tasks: theoretical analysis, model reduction, and interpretation. The polynomial form of a tree ensemble allows a straightforward interpretation of the original ...
Added: December 27, 2019
Mirkin B., Fenner T., Journal of Classification 2019 Vol. 36 No. 2 P. 350–367
Ranking is an important part of several areas of contemporary research, including social sciences, decision theory, data analysis and information retrieval. The goal of this project is to align developments in quantitative social sciences and decision theory with the current thought in computer science, including a few novel results. Specifically, we consider binary preference relations, ...
Added: November 1, 2019
Bulychev A., Сомов О. Д., В кн.: Информатика, управление и системный анализ: Труды V Всероссийской научной конференции молодых ученых с международным участием.: Ростов н/Д: Ростовский государственный экономический университет "РИНХ", 2018. С. 94–102.
In the process of developing an information system for logistics transportation, there is a need to determine the initial rating of the new carrier within the parent company. The presence of the rating helps to more accurately carry out the formation of orders and build forecasts of its interaction with the parent company in the ...
Added: September 3, 2019
Shvindt A., Моделирование, оптимизация и информационные технологии 2017 Т. 5 № 4 С. 1–18
The article reviews models and procedures for processing and evaluation of monitoring results, including student participation, focused on intellectual support of administrative managerial decisions when developing of conditions and corresponding resources for the achievement of applicable regulatory requirements for the quality of university education. The first stage of processing is normalization of factors which characterize ...
Added: August 19, 2019
Kulagin M., Sidorenko V., , in: Proceedings of the Third International Scientific Conference “Intelligent Information Technologies for Industry” (IITI’18) Volume 2Vol. 2.: Springer, 2019. P. 308–316.
This article describes modern methods of data processing regarding the task of assessing activities of transportation employees. The main purpose was to find dependencies in data and construct an algorithm for predicting the probability of transport safety violation by employee. The research was conducted for locomotive drivers. The following algorithms were used: neural networks, gradient boosting over decision trees and ...
Added: January 19, 2019