Pavel Braslavski
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Pavel Braslavski has a total of 72 co-authors. Here are the top co-authors by number of shared publications:
- Dmitry I. Ignatov
National Research University Higher School of Economics, Moscow, Russia
4 shared publications - Leonid Boytsov
Carnegie Mellon University Pittsburgh Pittsburgh PA PA USA
4 shared publications - Minxing Wang
Laboratory for Models and Methods of Computational Pragmatics, School of Data Analysis and AI, Faculty of Computer Science, HSE University, 11 Pokrovskiy Boulevard, Moscow 109028, Russia
4 shared publications - Pavel Efimov
ITMO University Saint Petersburg Russia
4 shared publications Alexander Beloborodov
Ural Federal University
2 shared publicationsArtem Kuznetsov
Ural Federal University
2 shared publications
Latest Publications
- Open Access
- 17 October 2026
- Published by Elsevier in SSRN Electronic Journal
When Wall Street Sleeps: Sparsity-Induced Multifractality and Regime Transitions in 24/7 Tokenized Equity Markets
Financial event processes—on-chain transfers, order arrivals, transaction counts—are often strongly zero-inflated: long silences punctuated by bursts. On such series, apparent multifractality can arise from the marginal distribution and the arrangement of zeros rather than from dependence among active events, and conventional shuffle-based attribution cannot separate these sources. We develop an operational decomposition of multifractal width for sparse event processes—separating marginal-distribution, zero-run, and active-event contributions via shuffle and run-preserving surrogates—and validate it on synthetic processes with known ground truth. As an empirical laboratory we use 24/7 tokenized equity markets: minute-level on-chain event panels for the Top-100 tokenized stocks issued by Ondo (over 70% of issuance) on BNB Smart Chain (83% of on-chain tokenized-equity volume), analyzing the Top-30 instruments across regular, extended, overnight, and weekend sessions. The market splits into two dynamical regimes. Weekday off-hours activity retains the in-hours persistence regime, and overnight prices are incorporated into the next open almost one-for-one. On weekends, a dramatic amplification of raw multifractal width inverts under surrogate correction: it is accounted for primarily by the marginal distribution of sparse activity, with no detectable positive contribution from active-event ordering; a staged 24/7 rollout inside the sample window provides quasi-experimental evidence consistent with this sparsity channel. Price-side validation shows that activity structure and price anchoring are separate information channels: de-anchoring is invisible to activity-side features. Continuous trading access thus does not imply a homogeneous continuous market: the regime of a 24/7 market is jointly determined by reference-price availability and primary-market activity.
- 17 October 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessWhen Wall Street Sleeps: Sparsity-Induced Multifractality and Regime Transitions in 24/7 Tokenized Equity Markets
Financial event processes—on-chain transfers, order arrivals, transaction counts—are often strongly zero-inflated: long silences punctuated by bursts. On such series, apparent multifractality can arise from the marginal distribution and the arrangement of zeros rather than from dependence among active events, and conventional shuffle-based attribution cannot separate these sources. We develop an operational decomposition of multifractal width for sparse event processes—separating marginal-distribution, zero-run, and active-event contributions via shuffle and run-preserving surrogates—and validate it on synthetic processes with known ground truth. As an empirical laboratory we use 24/7 tokenized equity markets: minute-level on-chain event panels for the Top-100 tokenized stocks issued by Ondo (over 70% of issuance) on BNB Smart Chain (83% of on-chain tokenized-equity volume), analyzing the Top-30 instruments across regular, extended, overnight, and weekend sessions. The market splits into two dynamical regimes. Weekday off-hours activity retains the in-hours persistence regime, and overnight prices are incorporated into the next open almost one-for-one. On weekends, a dramatic amplification of raw multifractal width inverts under surrogate correction: it is accounted for primarily by the marginal distribution of sparse activity, with no detectable positive contribution from active-event ordering; a staged 24/7 rollout inside the sample window provides quasi-experimental evidence consistent with this sparsity channel. Price-side validation shows that activity structure and price anchoring are separate information channels: de-anchoring is invisible to activity-side features. Continuous trading access thus does not imply a homogeneous continuous market: the regime of a 24/7 market is jointly determined by reference-price availability and primary-market activity.
- Open Access
- 19 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 5209-5214
FactOWL: A Cost-Efficient Tool for Long-Form Factuality Evaluation
- 19 July 2026
- Published by Association for Computing Machinery (ACM)
- p. 5209-5214
Open AccessFactOWL: A Cost-Efficient Tool for Long-Form Factuality Evaluation
- Open Access
- 13 April 2026
- Published by MDPI AG in Mathematics
- Vol. 14 (8) , 1286
What Drives Multi-Chain Crypto Forecasting: Model Choice, Feature Selection, and Transferability
Increasingly shaped by heterogeneous on-chain activity rather than a single shared market process, this study investigates 7-day-ahead forecasting using 147 market and on-chain indicators across eight major blockchain ecosystems from October 2023 to April 2025. We benchmark statistical, deep-learning, and foundation-model baselines under multiple feature-selection pipelines using both error metrics and Diebold–Mariano tests. TiRex achieves the best average MAPE () in a univariate setting without additional optimized covariates, while TFT remains slightly weaker even under its best feature-input configuration (MAPE: ; versus TiRex), suggesting a persistent practical advantage for TiRex. Importantly, TiRex’s zero-shot nature confers a substantial efficiency edge: by bypassing feature selection, it delivers comparable accuracy at a fraction of the computational cost. At the same time, feature selection materially affects many model families, with Boruta chosen in roughly 71.7% of best configurations. Taken together, the evidence supports a selective-feature principle: robust forecasting depends on validated, chain-specific features rather than larger feature sets. Feature-importance and overlap analyses further indicate a mixed structure of transferability, where broad market proxies provide baseline context while chain-specific variables drive marginal gains. Overall, this study highlights that effective multi-chain forecasting is primarily a feature selection problem under statistical uncertainty, while also showing that zero-shot designs like TiRex can achieve state-of-the-art accuracy with unmatched efficiency, offering practical implications for building leaner, more robust trading systems.- 13 April 2026
- Published by MDPI AG in Mathematics
- Vol. 14 (8) , 1286
Open AccessWhat Drives Multi-Chain Crypto Forecasting: Model Choice, Feature Selection, and Transferability
Increasingly shaped by heterogeneous on-chain activity rather than a single shared market process, this study investigates 7-day-ahead forecasting using 147 market and on-chain indicators across eight major blockchain ecosystems from October 2023 to April 2025. We benchmark statistical, deep-learning, and foundation-model baselines under multiple feature-selection pipelines using both error metrics and Diebold–Mariano tests. TiRex achieves the best average MAPE () in a univariate setting without additional optimized covariates, while TFT remains slightly weaker even under its best feature-input configuration (MAPE: ; versus TiRex), suggesting a persistent practical advantage for TiRex. Importantly, TiRex’s zero-shot nature confers a substantial efficiency edge: by bypassing feature selection, it delivers comparable accuracy at a fraction of the computational cost. At the same time, feature selection materially affects many model families, with Boruta chosen in roughly 71.7% of best configurations. Taken together, the evidence supports a selective-feature principle: robust forecasting depends on validated, chain-specific features rather than larger feature sets. Feature-importance and overlap analyses further indicate a mixed structure of transferability, where broad market proxies provide baseline context while chain-specific variables drive marginal gains. Overall, this study highlights that effective multi-chain forecasting is primarily a feature selection problem under statistical uncertainty, while also showing that zero-shot designs like TiRex can achieve state-of-the-art accuracy with unmatched efficiency, offering practical implications for building leaner, more robust trading systems. - Open Access
- 10 September 2025
- Published by MDPI AG in Forecasting
- Vol. 7 (3) , 48
TimeGPT’s Potential in Cryptocurrency Forecasting: Efficiency, Accuracy, and Economic Value
Accurate and efficient cryptocurrency price prediction is vital for investors in the volatile crypto market. This study comprehensively evaluates nine models—including baseline, zero-shot, and deep learning architectures—on 21 major cryptocurrencies using daily and hourly data. Our multi-dimensional evaluation assesses models based on prediction accuracy (MAE, RMSE, MAPE), speed, statistical significance (Diebold–Mariano test), and economic value (Sharpe Ratio). Our research found that the optimally fine-tuned TimeGPT model (without variables) demonstrated superior performance across both Daily and Hourly datasets, with its statistical leadership confirmed by the Diebold–Mariano test. Fine-tuned Chronos excelled in daily predictions, while TFT was a close second to TimeGPT for hourly forecasts. Crucially, zero-shot models like TimeGPT and Chronos were tens of times faster than traditional deep learning models, offering high accuracy with superior computational efficiency. A key finding from our economic analysis is that a model’s effectiveness is highly dependent on market characteristics. For instance, TimeGPT with variables showed exceptional profitability in the volatile ETH market, whereas the zero-shot Chronos model was the top performer for the cyclical BTC market. This also highlights that variables have asset-specific effects with TimeGPT: improving predictions for ICP, LTC, OP, and DOT, but hindering UNI, ATOM, BCH, and ARB. Recognizing that prior research has overemphasized prediction accuracy, this study provides a more holistic and practical standard for model evaluation by integrating speed, statistical significance, and economic value. Our findings collectively underscore TimeGPT’s immense potential as a leading solution for cryptocurrency forecasting, offering a top-tier balance of accuracy and efficiency. This multi-dimensional approach provides critical, theoretical, and practical guidance for investment decisions and risk management, proving especially valuable in real-time trading scenarios.
- 10 September 2025
- Published by MDPI AG in Forecasting
- Vol. 7 (3) , 48
Open AccessTimeGPT’s Potential in Cryptocurrency Forecasting: Efficiency, Accuracy, and Economic Value
Accurate and efficient cryptocurrency price prediction is vital for investors in the volatile crypto market. This study comprehensively evaluates nine models—including baseline, zero-shot, and deep learning architectures—on 21 major cryptocurrencies using daily and hourly data. Our multi-dimensional evaluation assesses models based on prediction accuracy (MAE, RMSE, MAPE), speed, statistical significance (Diebold–Mariano test), and economic value (Sharpe Ratio). Our research found that the optimally fine-tuned TimeGPT model (without variables) demonstrated superior performance across both Daily and Hourly datasets, with its statistical leadership confirmed by the Diebold–Mariano test. Fine-tuned Chronos excelled in daily predictions, while TFT was a close second to TimeGPT for hourly forecasts. Crucially, zero-shot models like TimeGPT and Chronos were tens of times faster than traditional deep learning models, offering high accuracy with superior computational efficiency. A key finding from our economic analysis is that a model’s effectiveness is highly dependent on market characteristics. For instance, TimeGPT with variables showed exceptional profitability in the volatile ETH market, whereas the zero-shot Chronos model was the top performer for the cyclical BTC market. This also highlights that variables have asset-specific effects with TimeGPT: improving predictions for ICP, LTC, OP, and DOT, but hindering UNI, ATOM, BCH, and ARB. Recognizing that prior research has overemphasized prediction accuracy, this study provides a more holistic and practical standard for model evaluation by integrating speed, statistical significance, and economic value. Our findings collectively underscore TimeGPT’s immense potential as a leading solution for cryptocurrency forecasting, offering a top-tier balance of accuracy and efficiency. This multi-dimensional approach provides critical, theoretical, and practical guidance for investment decisions and risk management, proving especially valuable in real-time trading scenarios.
- Open Access
- 13 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 35478-35489
Bitcoin Ordinals: Bitcoin Price and Transaction Fee Rate Predictions
Ordinals, a method for creating unique digital assets on the Bitcoin blockchain, has significantly impacted the blockchain over the past year, yet there is a notable lack of research on it. This study is the first to demonstrate that Bitcoin Ordinals-related data are crucial features for predicting Bitcoin transaction fee rates and prices. Our main contributions are threefold. 1) Dataset Construction: We construct a dataset that includes Bitcoin chain data, Ordinals index data, and Ordinals market data, as well as a dataset excluding Ordinals-related data. Our findings reveal that the fluctuation in the number of Ordinals inscriptions tends to correlate with market activity. When the Ordinals market is active, the share of Ordinals inscribed fees and the average Bitcoin transaction fee rate remain high. We argue that the upgrades of SegWit and Taproot drove the creation and development of Bitcoin Ordinals. Combined with users’ interest in Ordinals, this in turn affected the Bitcoin blockchain and its price; 2) Prediction: Using three metrics (MAE, RMSE, and MAPE) and the TemporalFusionTransformer model as a baseline, our comparative experiments show that Bitcoin Ordinals-related data is essential for predicting Bitcoin transaction fee rates and prices. This finding aids investors and participants in the Bitcoin Ordinals market in avoiding losses and leveraging congestion-related arbitrage opportunities, thus enabling more accurate decision-making in the cryptocurrency market; 3) Chronos Model: Additionally, the fine-tuned Chronos model achieves metrics comparable to or better than those of the TemporalFusionTransformer for shorter time intervals, especially in low-noise environments. With its outstanding zero-shot prediction performance, fast execution, and easy cloud deployment, the Chronos model allows investors and market participants to quickly obtain high-quality predictions without requiring complex data features.
- 13 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 35478-35489
Open AccessBitcoin Ordinals: Bitcoin Price and Transaction Fee Rate Predictions
Ordinals, a method for creating unique digital assets on the Bitcoin blockchain, has significantly impacted the blockchain over the past year, yet there is a notable lack of research on it. This study is the first to demonstrate that Bitcoin Ordinals-related data are crucial features for predicting Bitcoin transaction fee rates and prices. Our main contributions are threefold. 1) Dataset Construction: We construct a dataset that includes Bitcoin chain data, Ordinals index data, and Ordinals market data, as well as a dataset excluding Ordinals-related data. Our findings reveal that the fluctuation in the number of Ordinals inscriptions tends to correlate with market activity. When the Ordinals market is active, the share of Ordinals inscribed fees and the average Bitcoin transaction fee rate remain high. We argue that the upgrades of SegWit and Taproot drove the creation and development of Bitcoin Ordinals. Combined with users’ interest in Ordinals, this in turn affected the Bitcoin blockchain and its price; 2) Prediction: Using three metrics (MAE, RMSE, and MAPE) and the TemporalFusionTransformer model as a baseline, our comparative experiments show that Bitcoin Ordinals-related data is essential for predicting Bitcoin transaction fee rates and prices. This finding aids investors and participants in the Bitcoin Ordinals market in avoiding losses and leveraging congestion-related arbitrage opportunities, thus enabling more accurate decision-making in the cryptocurrency market; 3) Chronos Model: Additionally, the fine-tuned Chronos model achieves metrics comparable to or better than those of the TemporalFusionTransformer for shorter time intervals, especially in low-noise environments. With its outstanding zero-shot prediction performance, fast execution, and easy cloud deployment, the Chronos model allows investors and market participants to quickly obtain high-quality predictions without requiring complex data features.
- Open Access
KazQAD: Kazakh Open-Domain Question Answering Dataset
We introduce KazQAD -- a Kazakh open-domain question answering (ODQA) dataset -- that can be used in both reading comprehension and full ODQA settings, as well as for information retrieval experiments. KazQAD contains just under 6,000 unique questions with extracted short answers and nearly 12,000 passage-level relevance judgements. We use a combination of machine translation, Wikipedia search, and in-house manual annotation to ensure annotation efficiency and data quality. The questions come from two sources: translated items from the Natural Questions (NQ) dataset (only for training) and the original Kazakh Unified National Testing (UNT) exam (for development and testing). The accompanying text corpus contains more than 800,000 passages from the Kazakh Wikipedia. As a supplementary dataset, we release around 61,000 question-passage-answer triples from the NQ dataset that have been machine-translated into Kazakh. We develop baseline retrievers and readers that achieve reasonable scores in retrieval (NDCG@10 = 0.389 MRR = 0.382), reading comprehension (EM = 38.5 F1 = 54.2), and full ODQA (EM = 17.8 F1 = 28.7) settings. Nevertheless, these results are substantially lower than state-of-the-art results for English QA collections, and we think that there should still be ample room for improvement. We also show that the current OpenAI's ChatGPTv3.5 is not able to answer KazQAD test questions in the closed-book setting with acceptable quality. The dataset is freely available under the Creative Commons licence (CC BY-SA) at https://github.com/IS2AI/KazQAD.
Open AccessKazQAD: Kazakh Open-Domain Question Answering Dataset
We introduce KazQAD -- a Kazakh open-domain question answering (ODQA) dataset -- that can be used in both reading comprehension and full ODQA settings, as well as for information retrieval experiments. KazQAD contains just under 6,000 unique questions with extracted short answers and nearly 12,000 passage-level relevance judgements. We use a combination of machine translation, Wikipedia search, and in-house manual annotation to ensure annotation efficiency and data quality. The questions come from two sources: translated items from the Natural Questions (NQ) dataset (only for training) and the original Kazakh Unified National Testing (UNT) exam (for development and testing). The accompanying text corpus contains more than 800,000 passages from the Kazakh Wikipedia. As a supplementary dataset, we release around 61,000 question-passage-answer triples from the NQ dataset that have been machine-translated into Kazakh. We develop baseline retrievers and readers that achieve reasonable scores in retrieval (NDCG@10 = 0.389 MRR = 0.382), reading comprehension (EM = 38.5 F1 = 54.2), and full ODQA (EM = 17.8 F1 = 28.7) settings. Nevertheless, these results are substantially lower than state-of-the-art results for English QA collections, and we think that there should still be ample room for improvement. We also show that the current OpenAI's ChatGPTv3.5 is not able to answer KazQAD test questions in the closed-book setting with acceptable quality. The dataset is freely available under the Creative Commons licence (CC BY-SA) at https://github.com/IS2AI/KazQAD.
- 12 March 2024
- Published by Springer Nature
- p. 185-196
Needle in a Haystack: Finding Suitable Idioms Based on Text Descriptions
- 12 March 2024
- Published by Springer Nature
- p. 185-196
Needle in a Haystack: Finding Suitable Idioms Based on Text Descriptions
- 1 February 2024
- Published by Pleiades Publishing Ltd in Programming and Computer Software
- Vol. 50 (1) , 85-89
Calculating Similarity of Javadoc Comments
Code comments are an essential part of software documentation. Many software projects suffer from the problem of low-quality comments that are often produced by copy-paste. In case of similar methods, classes, etc. copy-pasted comments with minor modifications are justified. However, in many cases this approach leads to degraded documentation quality and, subsequently, to problematic maintenance and development of the project. In this study, we address the problem of near-duplicate code comments detection, which can potentially improve software documentation. We have conducted a thorough evaluation of traditional string similarity metrics and modern machine learning methods. In our experiment, we use a collection of Javadoc comments from four industrial open-source Java projects. We have found out that LCS (Longest Common Subsequence) is the best similarity algorithm taking into account both quality (Precision 94%, Recall 74%) and performance.
- 1 February 2024
- Published by Pleiades Publishing Ltd in Programming and Computer Software
- Vol. 50 (1) , 85-89
Calculating Similarity of Javadoc Comments
Code comments are an essential part of software documentation. Many software projects suffer from the problem of low-quality comments that are often produced by copy-paste. In case of similar methods, classes, etc. copy-pasted comments with minor modifications are justified. However, in many cases this approach leads to degraded documentation quality and, subsequently, to problematic maintenance and development of the project. In this study, we address the problem of near-duplicate code comments detection, which can potentially improve software documentation. We have conducted a thorough evaluation of traditional string similarity metrics and modern machine learning methods. In our experiment, we use a collection of Javadoc comments from four industrial open-source Java projects. We have found out that LCS (Longest Common Subsequence) is the best similarity algorithm taking into account both quality (Precision 94%, Recall 74%) and performance.
- Open Access
The Impact of Cross-Lingual Adjustment of Contextual Word Representations on Zero-Shot Transfer
Large multilingual language models such as mBERT or XLM-R enable zero-shot cross-lingual transfer in various IR and NLP tasks. Cao et al. (2020) proposed a data- and compute-efficient method for cross-lingual adjustment of mBERT that uses a small parallel corpus to make embeddings of related words across languages similar to each other. They showed it to be effective in NLI for five European languages. In contrast we experiment with a typologically diverse set of languages (Spanish, Russian, Vietnamese, and Hindi) and extend their original implementations to new tasks (XSR, NER, and QA) and an additional training regime (continual learning). Our study reproduced gains in NLI for four languages, showed improved NER, XSR, and cross-lingual QA results in three languages (though some cross-lingual QA gains were not statistically significant), while mono-lingual QA performance never improved and sometimes degraded. Analysis of distances between contextualized embeddings of related and unrelated words (across languages) showed that fine-tuning leads to "forgetting" some of the cross-lingual alignment information. Based on this observation, we further improved NLI performance using continual learning.
Open AccessThe Impact of Cross-Lingual Adjustment of Contextual Word Representations on Zero-Shot Transfer
Large multilingual language models such as mBERT or XLM-R enable zero-shot cross-lingual transfer in various IR and NLP tasks. Cao et al. (2020) proposed a data- and compute-efficient method for cross-lingual adjustment of mBERT that uses a small parallel corpus to make embeddings of related words across languages similar to each other. They showed it to be effective in NLI for five European languages. In contrast we experiment with a typologically diverse set of languages (Spanish, Russian, Vietnamese, and Hindi) and extend their original implementations to new tasks (XSR, NER, and QA) and an additional training regime (continual learning). Our study reproduced gains in NLI for four languages, showed improved NER, XSR, and cross-lingual QA results in three languages (though some cross-lingual QA gains were not statistically significant), while mono-lingual QA performance never improved and sometimes degraded. Analysis of distances between contextualized embeddings of related and unrelated words (across languages) showed that fine-tuning leads to "forgetting" some of the cross-lingual alignment information. Based on this observation, we further improved NLI performance using continual learning.
- 17 March 2023
- Published by Springer Nature
- p. 571-579
Consumer Health Question Answering Using Off-the-Shelf Components
- 17 March 2023
- Published by Springer Nature
- p. 571-579
Consumer Health Question Answering Using Off-the-Shelf Components