Alexey Zaytsev

Maxim Zhelnin, Dmitry Redko, Daniil Volkov et al.

Sequential recommendations (SR) with transformer-based architectures are widely adopted in real-world applications, where SR models require frequent retraining to adapt to ever-changing user preferences. However, training transformer-based SR models often encounters a high computational cost associated with scoring extensive item catalogs, often exceeding thousands of items. This occurs mainly due to the use of cross-entropy loss, where peak memory scales proportionally to catalog size, batch size, and sequence length. Recognizing this, practitioners in the field of recommendation systems typically address memory consumption by integrating the cross-entropy (CE) loss with negative sampling, thereby reducing the explicit memory demands of the final layer. However, a small number of negative samples would degrade model performance, and as we demonstrate in our work, increasing the number of negative samples and the batch size further improves the model's performance, but rapidly starts to exceed industrial GPUs' size (~40Gb). In this work, we introduce the CCE- method, which offers a GPU-efficient implementation of the CE loss with negative sampling. Our method accelerates training by up to two times while reducing memory consumption by more than 10 times. Leveraging the memory savings afforded by using CCE- for model training, it becomes feasible to enhance its accuracy on datasets with a large item catalog compared to those trained with original PyTorch-implemented loss functions. Finally, we perform an analysis of key memory-related hyperparameters and highlight the necessity of a delicate balance among these factors. We demonstrate that scaling both the number of negative samples and batch size leads to better results rather than maximizing only one of them. To facilitate further adoption of CCE-, we release a Triton kernel that efficiently implements the proposed method.

Gleb Bazhenov, Sergei Ivanov, Maxim Panov et al.

The problem of out-of-distribution detection for graph classification is far from being solved. The existing models tend to be overconfident about OOD examples or completely ignore the detection task. In this work, we consider this problem from the uncertainty estimation perspective and perform the comparison of several recently proposed methods. In our experiment, we find that there is no universal approach for OOD detection, and it is important to consider both graph representations and predictive categorical distribution.

Egor Shvetsov, Dmitry Osin, Alexey Zaytsev et al.

There is a constant need for high-performing and computationally efficient neural network models for image super-resolution: computationally efficient models can be used via low-capacity devices and reduce carbon footprints. One way to obtain such models is to compress models, e.g. quantization. Another way is a neural architecture search that automatically discovers new, more efficient solutions. We propose a novel quantization-aware procedure, the QuantNAS that combines pros of these two approaches. To make QuantNAS work, the procedure looks for quantization-friendly super-resolution models. The approach utilizes entropy regularization, quantization noise, and Adaptive Deviation for Quantization (ADQ) module to enhance the search procedure. The entropy regularization technique prioritizes a single operation within each block of the search space. Adding quantization noise to parameters and activations approximates model degradation after quantization, resulting in a more quantization-friendly architectures. ADQ helps to alleviate problems caused by Batch Norm blocks in super-resolution models. Our experimental results show that the proposed approximations are better for search procedure than direct model quantization. QuantNAS discovers architectures with better PSNR/BitOps trade-off than uniform or mixed precision quantization of fixed architectures. We showcase the effectiveness of our method through its application to two search spaces inspired by the state-of-the-art SR models and RFDN. Thus, anyone can design a proper search space based on an existing architecture and apply our method to obtain better quality and efficiency. The proposed procedure is 30\% faster than direct weight quantization and is more stable.

Elizaveta Kovtun, Galina Boeva, Andrey Shulga et al.

Most user-related data can be represented as a sequence of events associated with a timestamp and a collection of categorical labels. For example, the purchased basket of goods and the time of buying fully characterize the event of the store visit. Anticipation of the label set for the future event called the problem of temporal sets prediction, holds significant value, especially in such high-stakes industries as finance and e-commerce. A fundamental challenge of this task is the joint consideration of the temporal nature of events and label relations within sets. The existing models fail to capture complex time and label dependencies due to ineffective representation of historical information initially. We aim to address this shortcoming by presenting the framework with a specific way to aggregate the observed information into time- and set structure-aware views prior to transferring it into main architecture blocks. Our strong emphasis on input arrangement facilitates the subsequent efficient learning of label interactions. The proposed model is called Label-Attention NETwork, or LANET. We conducted experiments on four different datasets and made a comparison with four established models, including SOTA, in this area. The experimental results suggest that LANET provides significantly better quality than any other model, achieving an improvement up to $65 \%$ in terms of weighted F1 metric compared to the closest competitor. Moreover, we contemplate causal relationships between labels in our work, as well as a thorough study of LANET components' influence on performance. We provide an implementation of LANET to encourage its wider usage.

Ilya Shashkov, Nikita Balabin, Evgeny Burnaev et al.

Transferring a deep neural network trained on one problem to another requires only a small amount of data and little additional computation time. The same behaviour holds for ensembles of deep learning models typically superior to a single model. However, a transfer of deep neural networks ensemble demands relatively high computational expenses. The probability of overfitting also increases. Our approach for the transfer learning of ensembles consists of two steps: (a) shifting weights of encoders of all models in the ensemble by a single shift vector and (b) doing a tiny fine-tuning for each individual model afterwards. This strategy leads to a speed-up of the training process and gives an opportunity to add models to an ensemble with significantly reduced training time using the shift vector. We compare different strategies by computation time, the accuracy of an ensemble, uncertainty estimation and disagreement and conclude that our approach gives competitive results using the same computation complexity in comparison with the traditional approach. Also, our method keeps the ensemble's models' diversity higher.

Vladislav Zhuzhel, Vsevolod Grabar, Galina Boeva et al.

Event sequences often emerge in data mining. Modeling these sequences presents two main challenges: methodological and computational. Methodologically, event sequences are non-uniform and sparse, making traditional models unsuitable. Computationally, the vast amount of data and the significant length of each sequence necessitate complex and efficient models. Existing solutions, such as recurrent and transformer neural networks, rely on parametric intensity functions defined at each moment. These functions are either limited in their ability to represent complex event sequences or notably inefficient. We propose COTIC, a method based on an efficient convolution neural network designed to handle the non-uniform occurrence of events over time. Our paper introduces a continuous convolution layer, allowing a model to capture complex dependencies, including, e.g., the self-excitement effect, with little computational expense. COTIC outperforms existing models in predicting the next event time and type, achieving an average rank of 1.5 compared to 3.714 for the nearest competitor. Furthermore, COTIC`s ability to produce effective embeddings demonstrates its potential for various downstream tasks. Our code is open and available at: https://github.com/VladislavZh/COTIC.

Roman Kail, Kirill Fedyanin, Nikita Muravev et al.

The performance of modern deep learning-based systems dramatically depends on the quality of input objects. For example, face recognition quality would be lower for blurry or corrupted inputs. However, it is hard to predict the influence of input quality on the resulting accuracy in more complex scenarios. We propose an approach for deep metric learning that allows direct estimation of the uncertainty with almost no additional computational cost. The developed \textit{ScaleFace} algorithm uses trainable scale values that modify similarities in the space of embeddings. These input-dependent scale values represent a measure of confidence in the recognition result, thus allowing uncertainty estimation. We provide comprehensive experiments on face recognition tasks that show the superior performance of ScaleFace compared to other uncertainty-aware face recognition approaches. We also extend the results to the task of text-to-image retrieval showing that the proposed approach beats the competitors with significant margin.

Alexander Marusov, Vsevolod Grabar, Yury Maximov et al.

The problem of high-quality drought forecasting up to a year in advance is critical for agriculture planning and insurance. Yet, it is still unsolved with reasonable accuracy due to data complexity and aridity stochasticity. We tackle drought data by introducing an end-to-end approach that adopts a spatio-temporal neural network model with accessible open monthly climate data as the input. Our systematic research employs diverse proposed models and five distinct environmental regions as a testbed to evaluate the efficacy of the Palmer Drought Severity Index (PDSI) prediction. Key aggregated findings are the exceptional performance of a Transformer model, EarthFormer, in making accurate short-term (up to six months) forecasts. At the same time, the Convolutional LSTM excels in longer-term forecasting.

Polina Proscura, Alexey Zaytsev

Ensembling is a popular and effective method for improving machine learning (ML) models. It proves its value not only in classical ML but also for deep learning. Ensembles enhance the quality and trustworthiness of ML solutions, and allow uncertainty estimation. However, they come at a price: training ensembles of deep learning models eat a huge amount of computational resources. A snapshot ensembling collects models in the ensemble along a single training path. As it runs training only one time, the computational time is similar to the training of one model. However, the quality of models along the training path is different: typically, later models are better if no overfitting occurs. So, the models are of varying utility. Our method improves snapshot ensembling by selecting and weighting ensemble members along the training path. It relies on training-time likelihoods without looking at validation sample errors that standard stacking methods do. Experimental evidence for Fashion MNIST, CIFAR-10, and CIFAR-100 datasets demonstrates the superior quality of the proposed weighted ensembles c.t. vanilla ensembling of deep learning models.

Diana Koldasbayeva, Polina Tregubova, Mikhail Gasanov et al.

With the rise of electronic data, particularly Earth observation data, data-based geospatial modelling using machine learning (ML) has gained popularity in environmental research. Accurate geospatial predictions are vital for domain research based on ecosystem monitoring and quality assessment and for policy-making and action planning, considering effective management of natural resources. The accuracy and computation speed of ML has generally proved efficient. However, many questions have yet to be addressed to obtain precise and reproducible results suitable for further use in both research and practice. A better understanding of the ML concepts applicable to geospatial problems enhances the development of data science tools providing transparent information crucial for making decisions on global challenges such as biosphere degradation and climate change. This survey reviews common nuances in geospatial modelling, such as imbalanced data, spatial autocorrelation, prediction errors, model generalisation, domain specificity, and uncertainty estimation. We provide an overview of techniques and popular programming tools to overcome or account for the challenges. We also discuss prospects for geospatial Artificial Intelligence in environmental applications.

Alexandra Bazarova, Andrei Volodichev, Daria Kotova et al.

While retrieval-augmented generation (RAG) significantly improves the factual reliability of LLMs, it does not eliminate hallucinations, so robust uncertainty quantification (UQ) remains essential. In this paper, we reveal that standard entropy-based UQ methods often fail in RAG settings due to a mechanistic paradox. An internal "tug-of-war" inherent to context utilization appears: while induction heads promote grounded responses by copying the correct answer, they collaterally trigger the previously established "entropy neurons". This interaction inflates predictive entropy, causing the model to signal false uncertainty on accurate outputs. To address this, we propose INTRYGUE (Induction-Aware Entropy Gating for Uncertainty Estimation), a mechanistically grounded method that gates predictive entropy based on the activation patterns of induction heads. Evaluated across four RAG benchmarks and six open-source LLMs (4B to 13B parameters), INTRYGUE consistently matches or outperforms a wide range of UQ baselines. Our findings demonstrate that hallucination detection in RAG benefits from combining predictive uncertainty with interpretable, internal signals of context utilization.

Leonid Erlygin, Vladimir Zholobov, Valeriia Baklanova et al.

Machine learning models play a vital role in time series forecasting. These models, however, often overlook an important element: point uncertainty estimates. Incorporating these estimates is crucial for effective risk management, informed model selection, and decision-making.To address this issue, our research introduces a method for uncertainty estimation. We employ a surrogate Gaussian process regression model. It enhances any base regression model with reasonable uncertainty estimates. This approach stands out for its computational efficiency. It only necessitates training one supplementary surrogate and avoids any data-specific assumptions. Furthermore, this method for work requires only the presence of the base model as a black box and its respective training data. The effectiveness of our approach is supported by experimental results. Using various time-series forecasting data, we found that our surrogate model-based technique delivers significantly more accurate confidence intervals. These techniques outperform both bootstrap-based and built-in methods in a medium-data regime. This superiority holds across a range of base model types, including a linear regression, ARIMA, gradient boosting and a neural network.

Leonid Erlygin, Alexey Zaytsev

Accurate uncertainty estimation is essential for building robust and trustworthy recognition systems. In this paper, we consider the open-set text classification (OSTC) task - and uncertainty estimation for it. For OSTC a text sample should be classified as one of the existing classes or rejected as unknown. To account for the different uncertainty types encountered in OSTC, we adapt the Holistic Uncertainty Estimation (HolUE) method for the text domain. Our approach addresses two major causes of prediction errors in text recognition systems: text uncertainty that stems from ill formulated queries and gallery uncertainty that is related the ambiguity of data distribution. By capturing these sources, it becomes possible to predict when the system will make a recognition error. We propose a new OSTC benchmark and conduct extensive experiments on a wide range of data, utilizing the authorship attribution, intent and topic classification datasets. HolUE achieves 40-365% improvement in Prediction Rejection Ratio (PRR) over the quality-based SCF baseline across datasets: 365% on Yahoo Answers (0.79 vs 0.17 at FPIR 0.1), 347% on DBPedia (0.85 vs 0.19), 240% on PAN authorship attribution (0.51 vs 0.15 at FPIR 0.5), and 40% on CLINC150 intent classification (0.73 vs~0.52). We make public our code and protocols https://github.com/Leonid-Erlygin/text_uncertainty.git

Elizaveta Kostenok, Daniil Cherniavskii, Alexey Zaytsev

Transformer-based language models have set new benchmarks across a wide range of NLP tasks, yet reliably estimating the uncertainty of their predictions remains a significant challenge. Existing uncertainty estimation (UE) techniques often fall short in classification tasks, either offering minimal improvements over basic heuristics or relying on costly ensemble models. Moreover, attempts to leverage common embeddings for UE in linear probing scenarios have yielded only modest gains, indicating that alternative model components should be explored. We tackle these limitations by harnessing the geometry of attention maps across multiple heads and layers to assess model confidence. Our approach extracts topological features from attention matrices, providing a low-dimensional, interpretable representation of the model's internal dynamics. Additionally, we introduce topological features to compare attention patterns across heads and layers. Our method significantly outperforms existing UE techniques on benchmarks for acceptability judgments and artificial text detection, offering a more efficient and interpretable solution for uncertainty estimation in large-scale language models.

Alexander Marusov, Alexey Zaytsev

The representation learning problem in the oil & gas industry aims to construct a model that provides a representation based on logging data for a well interval. Previous attempts are mainly supervised and focus on similarity task, which estimates closeness between intervals. We desire to build informative representations without using supervised (labelled) data. One of the possible approaches is self-supervised learning (SSL). In contrast to the supervised paradigm, this one requires little or no labels for the data. Nowadays, most SSL approaches are either contrastive or non-contrastive. Contrastive methods make representations of similar (positive) objects closer and distancing different (negative) ones. Due to possible wrong marking of positive and negative pairs, these methods can provide an inferior performance. Non-contrastive methods don't rely on such labelling and are widespread in computer vision. They learn using only pairs of similar objects that are easier to identify in logging data. We are the first to introduce non-contrastive SSL for well-logging data. In particular, we exploit Bootstrap Your Own Latent (BYOL) and Barlow Twins methods that avoid using negative pairs and focus only on matching positive pairs. The crucial part of these methods is an augmentation strategy. Our augmentation strategies and adaption of BYOL and Barlow Twins together allow us to achieve superior quality on clusterization and mostly the best performance on different classification tasks. Our results prove the usefulness of the proposed non-contrastive self-supervised approaches for representation learning and interval similarity in particular.

Sergey Egorov, Narek Gevorgyan, Alexey Zaytsev

Adopting data-based approaches leads to model improvement in numerous Oil&Gas logging data processing problems. These improvements become even more sound due to new capabilities provided by deep learning. However, usage of deep learning is limited to areas where researchers possess large amounts of high-quality data. We present an approach that provides universal data representations suitable for solutions to different problems for different oil fields with little additional data. Our approach relies on the self-supervised methodology for sequential logging data for intervals from well, so it also doesn't require labelled data from the start. For validation purposes of the received representations, we consider classification and clusterization problems. We as well consider the transfer learning scenario. We found out that using the variational autoencoder leads to the most reliable and accurate models. approach We also found that a researcher only needs a tiny separate data set for the target oil field to solve a specific problem on top of universal representations.

Elizaveta Kovtun, Maksim Makarenko, Natalia Semenova et al.

A graph with semantically attributed nodes are a common data structure in a wide range of domains. It could be interlinked web data or citation networks of scientific publications. The essential problem for such a data type is to determine nodes that carry greater importance than all the others, a task that markedly enhances system monitoring and management. Traditional methods to identify important nodes in networks introduce centrality measures, such as node degree or more complex PageRank. However, they consider only the network structure, neglecting the rich node attributes. Recent methods adopt neural networks capable of handling node features, but they require supervision. This work addresses the identified gap--the absence of approaches that are both unsupervised and attribute-aware--by introducing a Pipeline for Important Node Exploration (PINE). At the core of the proposed framework is an attention-based graph model that incorporates node semantic features in the learning process of identifying the structural graph properties. The PINE's node importance scores leverage the obtained attention distribution. We demonstrate the superior performance of the proposed PINE method on various homogeneous and heterogeneous attributed networks. As an industry-implemented system, PINE tackles the real-world challenge of unsupervised identification of key entities within large-scale enterprise graphs.

Petr Sokerin, Kristian Kuznetsov, Elizaveta Makhneva et al.

Portfolio management is an essential part of investment decision-making. However, traditional methods often fail to deliver reasonable performance. This problem stems from the inability of these methods to account for the unique characteristics of multivariate time series data from stock markets. We present a two-stage method for constructing an investment portfolio of common stocks. The method involves the generation of time series representations followed by their subsequent clustering. Our approach utilizes features based on Topological Data Analysis (TDA) for the generation of representations, allowing us to elucidate the topological structure within the data. Experimental results show that our proposed system outperforms other methods. This superior performance is consistent over different time frames, suggesting the viability of TDA as a powerful tool for portfolio selection.

Polina Proskura, Alexey Zaytsev

In machine learning, ensembles are important tools for improving the model performance. In natural language processing specifically, ensembles boost the performance of a method due to multiple large models available in open source. However, existing approaches mostly rely on simple averaging of predictions by ensembles with equal weights for each model, ignoring differences in the quality and conformity of models. We propose to estimate weights for ensembles of NLP models using not only knowledge of their individual performance but also their similarity to each other. By adopting distance measures based on Topological Data Analysis (TDA), we improve our ensemble. The quality improves for both text classification accuracy and relevant uncertainty estimation.

Alina Ermilova, Elizaveta Kovtun, Dmitry Berestnev et al.

The financial industry relies on deep learning models for making important decisions. This adoption brings new danger, as deep black-box models are known to be vulnerable to adversarial attacks. In computer vision, one can shape the output during inference by performing an adversarial attack called poisoning via introducing a backdoor into the model during training. For sequences of financial transactions of a customer, insertion of a backdoor is harder to perform, as models operate over a more complex discrete space of sequences, and systematic checks for insecurities occur. We provide a method to introduce concealed backdoors, creating vulnerabilities without altering their functionality for uncontaminated data. To achieve this, we replace a clean model with a poisoned one that is aware of the availability of a backdoor and utilize this knowledge. Our most difficult for uncovering attacks include either additional supervised detection step of poisoned data activated during the test or well-hidden model weight modifications. The experimental study provides insights into how these effects vary across different datasets, architectures, and model components. Alternative methods and baselines, such as distillation-type regularization, are also explored but found to be less efficient. Conducted on three open transaction datasets and architectures, including LSTM, CNN, and Transformer, our findings not only illuminate the vulnerabilities in contemporary models but also can drive the construction of more robust systems.

Alina Ermilova, Nikita Baramiia, Valerii Kornilov et al.

Transformer-based neural network architectures achieve state-of-the-art results in different domains, from natural language processing (NLP) to computer vision (CV). The key idea of Transformers, the attention mechanism, has already led to significant breakthroughs in many areas. The attention has found their implementation for time series data as well. However, due to the quadratic complexity of the attention calculation regarding input sequence length, the application of Transformers is limited by high resource demands. Moreover, their modifications for industrial time series need to be robust to missing or noised values, which complicates the expansion of the horizon of their application. To cope with these issues, we introduce the class of efficient Transformers named Regularized Transformers (Reguformers). We implement the regularization technique inspired by the dropout ideas to improve robustness and reduce computational expenses. The focus in our experiments is on oil&gas data, namely, well logs, a prominent example of multivariate time series. The goal is to solve the problems of similarity and representation learning for them. To evaluate our models for such problems, we work with an industry-scale open dataset consisting of well logs of more than 20 wells. The experiments show that all variations of Reguformers outperform the previously developed RNNs, classical Transformer model, and robust modifications of it like Informer and Performer in terms of well-intervals' classification and the quality of the obtained well-intervals' representations. Moreover, the sustainability to missing and incorrect data in our models exceeds that of others by a significant margin. The best result that the Reguformer achieves on well-interval similarity task is the mean PR~AUC score equal to 0.983, which is comparable to the classical Transformer and outperforms the previous models.

Anastasiia Prutianova, Alexey Zaytsev, Chung-Kuei Lee et al.

Existing neural networks are memory-consuming and computationally intensive, making deploying them challenging in resource-constrained environments. However, there are various methods to improve their efficiency. Two such methods are quantization, a well-known approach for network compression, and re-parametrization, an emerging technique designed to improve model performance. Although both techniques have been studied individually, there has been limited research on their simultaneous application. To address this gap, we propose a novel approach called RepQ, which applies quantization to re-parametrized networks. Our method is based on the insight that the test stage weights of an arbitrary re-parametrized layer can be presented as a differentiable function of trainable parameters. We enable quantization-aware training by applying quantization on top of this function. RepQ generalizes well to various re-parametrized models and outperforms the baseline method LSQ quantization scheme in all experiments.

Ilya Kuleshov, Evgenia Romanenkova, Vladislav Zhuzhel et al.

Neural CDEs provide a natural way to process the temporal evolution of irregular time series. The number of function evaluations (NFE) is these systems' natural analog of depth (the number of layers in traditional neural networks). It is usually regulated via solver error tolerance: lower tolerance means higher numerical precision, requiring more integration steps. However, lowering tolerances does not adequately increase the models' expressiveness. We propose a simple yet effective alternative: scaling the integration time horizon to increase NFEs and "deepen`` the model. Increasing the integration interval causes uncontrollable growth in conventional vector fields, so we also propose a way to stabilize the dynamics via Negative Feedback (NF). It ensures provable stability without constraining flexibility. It also implies robustness: we provide theoretical bounds for Neural ODE risk using Gaussian process theory. Experiments on four open datasets demonstrate that our method, DeNOTS, outperforms existing approaches~ -- ~including recent Neural RDEs and state space models,~ -- ~achieving up to $20\%$ improvement in metrics. DeNOTS combines expressiveness, stability, and robustness, enabling reliable modelling in continuous-time domains.

Anna Dmitrienko, Evgenia Romanenkova, Alexey Zaytsev

The change point is a moment of an abrupt alteration in the data distribution. Current methods for change point detection are based on recurrent neural methods suitable for sequential data. However, recent works show that transformers based on attention mechanisms perform better than standard recurrent models for many tasks. The most benefit is noticeable in the case of longer sequences. In this paper, we investigate different attentions for the change point detection task and proposed specific form of attention related to the task at hand. We show that using a special form of attention outperforms state-of-the-art results.

Alexander Stepikin, Evgenia Romanenkova, Alexey Zaytsev

A change points detection aims to catch an abrupt disorder in data distribution. Common approaches assume that there are only two fixed distributions for data: one before and another after a change point. Real-world data are richer than this assumption. There can be multiple different distributions before and after a change. We propose an approach that works in the multiple-distributions scenario. Our approach learn representations for semi-structured data suitable for change point detection, while a common classifiers-based approach fails. Moreover, our model is more robust, when predicting change points. The datasets used for benchmarking are sequences of images with and without change points in them.

Aleksei Khalin, Ekaterina Zaychenkova, Aleksandr Yugay et al.

Artificial intelligence (AI) systems accelerate medical workflows and improve diagnostic accuracy in healthcare, serving as second-opinion systems. However, the unpredictability of AI errors poses a significant challenge, particularly in healthcare contexts, where mistakes can have severe consequences. A widely adopted safeguard is to pair predictions with uncertainty estimation, enabling human experts to focus on high-risk cases while streamlining routine verification. Current uncertainty estimation methods, however, remain limited, particularly in quantifying aleatoric uncertainty, which arises from data ambiguity and noise. To address this, we propose a novel approach that leverages disagreement in expert responses to generate targets for training machine learning models. These targets are used in conjunction with standard data labels to estimate two components of uncertainty separately, as given by the law of total variance, via a two-ensemble approach, as well as its lightweight variant. We validate our method on binary image classification, binary and multi-class image segmentation, and multiple-choice question answering. Our experiments demonstrate that incorporating expert knowledge can enhance uncertainty estimation quality by $9\%$ to $50\%$ depending on the task, making this source of information invaluable for the construction of risk-aware AI systems in healthcare applications.

Leonid Erlygin, Alexey Zaytsev

Accurate uncertainty estimation is a critical challenge in open-set recognition, where a probe biometric sample may belong to an unknown identity. It can be addressed through sample quality estimation via probabilistic embeddings. However, the low variance of probabilistic embedding only partly implies a low identification error probability: an embedding of a sample could be close to several classes in a gallery, thus yielding high uncertainty despite high sample quality. We propose HolUE - a holistic uncertainty estimation method based on a Bayesian probabilistic model; it is aware of two sources of ambiguity in the open-set recognition system: (1) the gallery uncertainty caused by overlapping classes and (2) the uncertainty of embeddings. Challenging open-set recognition datasets, such as IJB-C for the image domain and VoxBlink for the audio domain, serve as a testbed for our method. We also provide a new open-set recognition protocol for the identification of whales and dolphins. In all cases, HolUE better identifies recognition errors than alternative uncertainty estimation methods, including those based solely on sample quality.

Alexey Zaytsev, Maria Kovaleva, Alex Natekin et al.

Banks routinely use neural networks to make decisions. While these models offer higher accuracy, they are susceptible to adversarial attacks, a risk often overlooked in the context of event sequences, particularly sequences of financial transactions, as most works consider computer vision and NLP modalities. We propose a thorough approach to studying these risks: a novel type of competition that allows a realistic and detailed investigation of problems in financial transaction data. The participants directly oppose each other, proposing attacks and defenses -- so they are examined in close-to-real-life conditions. The paper outlines our unique competition structure with direct opposition of participants, presents results for several different top submissions, and analyzes the competition results. We also introduce a new open dataset featuring financial transactions with credit default labels, enhancing the scope for practical research and development.

Aleksandr Yugay, Alexey Zaytsev

High-quality representation of transactional sequences is vital for modern banking applications, including risk management, churn prediction, and personalized customer offers. Different tasks require distinct representation properties: local tasks benefit from capturing the client's current state, while global tasks rely on general behavioral patterns. Previous research has demonstrated that various self-supervised approaches yield representations that better capture either global or local qualities. This study investigates the integration of two self-supervised learning techniques - instance-wise contrastive learning and a generative approach based on restoring masked events in latent space. The combined approach creates representations that balance local and global transactional data characteristics. Experiments conducted on several public datasets, focusing on sequence classification and next-event type prediction, show that the integrated method achieves superior performance compared to individual approaches and demonstrates synergistic effects. These findings suggest that the proposed approach offers a robust framework for advancing event sequences representation learning in the financial sector.

Ilya Kuleshov, Alexander Marusov, Alexey Zaytsev

Probabilistic forecasting of irregularly sampled time series is crucial in domains such as healthcare and finance, yet it remains a formidable challenge. Existing Neural Controlled Differential Equation (Neural CDE) approaches, while effective at modelling continuous dynamics, suffer from slow, inherently sequential computation, which restricts scalability and limits access to global context. We introduce UFO (U-Former ODE), a novel architecture that seamlessly integrates the parallelizable, multiscale feature extraction of U-Nets, the powerful global modelling of Transformers, and the continuous-time dynamics of Neural CDEs. By constructing a fully causal, parallelizable model, UFO achieves a global receptive field while retaining strong sensitivity to local temporal dynamics. Extensive experiments on five standard benchmarks -- covering both regularly and irregularly sampled time series -- demonstrate that UFO consistently outperforms ten state-of-the-art neural baselines in predictive accuracy. Moreover, UFO delivers up to 15$\times$ faster inference compared to conventional Neural CDEs, with consistently strong performance on long and highly multivariate sequences.

Pavel Burnyshev, Elizaveta Kostenok, Alexey Zaytsev

Adversarial attacks expose vulnerabilities of deep learning models by introducing minor perturbations to the input, which lead to substantial alterations in the output. Our research focuses on the impact of such adversarial attacks on sequence-to-sequence (seq2seq) models, specifically machine translation models. We introduce algorithms that incorporate basic text perturbation heuristics and more advanced strategies, such as the gradient-based attack, which utilizes a differentiable approximation of the inherently non-differentiable translation metric. Through our investigation, we provide evidence that machine translation models display robustness displayed robustness against best performed known adversarial attacks, as the degree of perturbation in the output is directly proportional to the perturbation in the input. However, among underdogs, our attacks outperform alternatives, providing the best relative performance. Another strong candidate is an attack based on mixing of individual characters.

Boris Prokhorov, Diana Koldasbayeva, Alexey Zaytsev

In machine learning models, the estimation of errors is often complex due to distribution bias, particularly in spatial data such as those found in environmental studies. We introduce an approach based on the ideas of importance sampling to obtain an unbiased estimate of the target error. By taking into account difference between desirable error and available data, our method reweights errors at each sample point and neutralizes the shift. Importance sampling technique and kernel density estimation were used for reweighteing. We validate the effectiveness of our approach using artificial data that resemble real-world spatial datasets. Our findings demonstrate advantages of the proposed approach for the estimation of the target error, offering a solution to a distribution shift problem. Overall error of predictions dropped from 7% to just 2% and it gets smaller for larger samples.

Theodore Papamarkou, Pierre Alquier, Matthias Bauer et al.

LLMs excel at predictive tasks and complex reasoning tasks, but many high-value deployments rely on decisions under uncertainty, for example, which tool to call, which expert to consult, or how many resources to invest. While the usefulness and feasibility of Bayesian approaches remain unclear for LLM inference, this position paper argues that the control layer of an agentic AI system (that orchestrates LLMs and tools) is a clear case where Bayesian principles should shine. Bayesian decision theory provides a framework for agentic systems that can help to maintain beliefs over task-relevant latent quantities, to update these beliefs from observed agentic and human-AI interactions, and to choose actions. Making LLMs themselves explicitly Bayesian belief-updating engines remains computationally intensive and conceptually nontrivial as a general modeling target. In contrast, this paper argues that coherent decision-making requires Bayesian principles at the orchestration level of the agentic system, not necessarily the LLM agent parameters. This paper articulates practical properties for Bayesian control that fit modern agentic AI systems and human-AI collaboration, and provides concrete examples and design patterns to illustrate how calibrated beliefs and utility-aware policies can improve agentic AI orchestration.

Valeriy Shevchenko, Nikita Belousov, Alexey Vasilev et al.

In the rapidly evolving domain of Recommender Systems (RecSys), new algorithms frequently claim state-of-the-art performance based on evaluations over a limited set of arbitrarily selected datasets. However, this approach may fail to holistically reflect their effectiveness due to the significant impact of dataset characteristics on algorithm performance. Addressing this deficiency, this paper introduces a novel benchmarking methodology to facilitate a fair and robust comparison of RecSys algorithms, thereby advancing evaluation practices. By utilizing a diverse set of $30$ open datasets, including two introduced in this work, and evaluating $11$ collaborative filtering algorithms across $9$ metrics, we critically examine the influence of dataset characteristics on algorithm performance. We further investigate the feasibility of aggregating outcomes from multiple datasets into a unified ranking. Through rigorous experimental analysis, we validate the reliability of our methodology under the variability of datasets, offering a benchmarking strategy that balances quality and computational demands. This methodology enables a fair yet effective means of evaluating RecSys algorithms, providing valuable guidance for future research endeavors.

Alexandra Bazarova, Maria Kovaleva, Ilya Kuleshov et al.

In today's world, banks use artificial intelligence to optimize diverse business processes, aiming to improve customer experience. Most of the customer-related tasks can be categorized into two groups: 1) local ones, which focus on a client's current state, such as transaction forecasting, and 2) global ones, which consider the general customer behaviour, e.g., predicting successful loan repayment. Unfortunately, maintaining separate models for each task is costly. Therefore, to better facilitate information management, we compared eight state-of-the-art unsupervised methods on 11 tasks in search for a one-size-fits-all solution. Contrastive self-supervised learning methods were demonstrated to excel at global problems, while generative techniques were superior at local tasks. We also introduced a novel approach, which enriches the client's representation by incorporating external information gathered from other clients. Our method outperforms classical models, boosting accuracy by up to 20\%.

Egor Serov, Ilya Kuleshov, Alexey Zaytsev

Neural Controlled Differential Equations (Neural CDEs) provide a powerful continuous-time framework for sequence modeling, yet the roughness of the driving control path often restricts their efficiency. Standard splines introduce high-frequency variations that force adaptive solvers to take excessively small steps, driving up the Number of Function Evaluations (NFE). We propose a novel approach to Neural CDE path construction that replaces exact interpolation with Kernel and Gaussian Process (GP) smoothing, enabling explicit control over trajectory regularity. To recover details lost during smoothing, we propose an attention-based Multi-View CDE (MV-CDE) and its convolutional extension (MVC-CDE), which employ learnable queries to inform path reconstruction. This framework allows the model to distribute representational capacity across multiple trajectories, each capturing distinct temporal patterns. Empirical results demonstrate that our method, MVC-CDE with GP, achieves state-of-the-art accuracy while significantly reducing NFEs and total inference time compared to spline-based baselines.

Alexandra Bazarova, Aleksandr Yugay, Andrey Shulga et al.

Hallucination, i.e., generating factually incorrect content, remains a critical challenge for large language models (LLMs). We introduce TOHA, a TOpology-based HAllucination detector in the RAG setting, which leverages a topological divergence metric to quantify the structural properties of graphs induced by attention matrices. Examining the topological divergence between prompt and response subgraphs reveals consistent patterns: higher divergence values in specific attention heads correlate with hallucinated outputs, independent of the dataset. Extensive experiments - including evaluation on question answering and summarization tasks - show that our approach achieves state-of-the-art or competitive results on several benchmarks while requiring minimal annotated data and computational resources. Our findings suggest that analyzing the topological structure of attention matrices can serve as an efficient and robust indicator of factual reliability in LLMs.

Alexandra Bazarova, Evgenia Romanenkova, Alexey Zaytsev

Change point detection (CPD) methods aim to identify abrupt shifts in the distribution of input data streams. Accurate estimators for this task are crucial across various real-world scenarios. Yet, traditional unsupervised CPD techniques face significant limitations, often relying on strong assumptions or suffering from low expressive power due to inherent model simplicity. In contrast, representation learning methods overcome these drawbacks by offering flexibility and the ability to capture the full complexity of the data without imposing restrictive assumptions. However, these approaches are still emerging in the CPD field and lack robust theoretical foundations to ensure their reliability. Our work addresses this gap by integrating the expressive power of representation learning with the groundedness of traditional CPD techniques. We adopt spectral normalization (SN) for deep representation learning in CPD tasks and prove that the embeddings after SN are highly informative for CPD. Our method significantly outperforms current state-of-the-art methods during the comprehensive evaluation via three standard CPD datasets.

Rodion Oblovatny, Alexandra Bazarova, Alexey Zaytsev

We present a novel approach for detecting hallucinations in large language models (LLMs) by analyzing the probabilistic divergence between prompt and response hidden-state distributions. Counterintuitively, we find that hallucinated responses exhibit smaller deviations from their prompts compared to grounded responses, suggesting that hallucinations often arise from superficial rephrasing rather than substantive reasoning. Leveraging this insight, we propose a model-intrinsic detection method that uses distributional distances as principled hallucination scores, eliminating the need for external knowledge or auxiliary models. To enhance sensitivity, we employ deep learnable kernels that automatically adapt to capture nuanced geometric differences between distributions. Our approach outperforms existing baselines, demonstrating state-of-the-art performance on several benchmarks. The method remains competitive even without kernel training, offering a robust, scalable solution for hallucination detection.

Petr Sychev, Andrey Goncharov, Daniil Vyazhev et al.

Uncertainty estimation is crucial for evaluating Large Language Models (LLMs), particularly in high-stakes domains where incorrect answers result in significant consequences. Numerous approaches consider this problem, while focusing on a specific type of uncertainty, ignoring others. We investigate what estimates, specifically token-wise entropy and model-as-judge (MASJ), would work for multiple-choice question-answering tasks for different question topics. Our experiments consider three LLMs: Phi-4, Mistral, and Qwen of different sizes from 1.5B to 72B and $14$ topics. While MASJ performs similarly to a random error predictor, the response entropy predicts model error in knowledge-dependent domains and serves as an effective indicator of question difficulty: for biology ROC AUC is $0.73$. This correlation vanishes for the reasoning-dependent domain: for math questions ROC-AUC is $0.55$. More principally, we found out that the entropy measure required a reasoning amount. Thus, data-uncertainty related entropy should be integrated within uncertainty estimates frameworks, while MASJ requires refinement. Moreover, existing MMLU-Pro samples are biased, and should balance required amount of reasoning for different subdomains to provide a more fair assessment of LLMs performance.

Diana Koldasbayeva, Alexey Zaytsev

Species Distribution Models (SDMs) often suffer from spatial autocorrelation (SAC), leading to biased performance estimates. We tested cross-validation (CV) strategies - random splits, spatial blocking with varied distances, environmental (ENV) clustering, and a novel spatio-temporal method - under two proposed training schemes: LAST FOLD, widely used in spatial CV at the cost of data loss, and RETRAIN, which maximizes data usage but risks reintroducing SAC. LAST FOLD consistently yielded lower errors and stronger correlations. Spatial blocking at an optimal distance (SP 422) and ENV performed best, achieving Spearman and Pearson correlations of 0.485 and 0.548, respectively, although ENV may be unsuitable for long-term forecasts involving major environmental shifts. A spatio-temporal approach yielded modest benefits in our moderately variable dataset, but may excel with stronger temporal changes. These findings highlight the need to align CV approaches with the spatial and temporal structure of SDM data, ensuring rigorous validation and reliable predictive outcomes.

Ilya Kuleshov, Alexey Zaytsev

Neural Controlled Differential Equations (Neural CDEs, NCDEs) are a unique branch of methods, specifically tailored for analysing temporal sequences. However, they come with drawbacks, the main one being the number of parameters, required for the method's operation. In this paper, we propose an alternative, parameter-efficient look at Neural CDEs. It requires much fewer parameters, while also presenting a very logical analogy as the "Continuous RNN", which the Neural CDEs aspire to.

Ivan Dubrovsky, Anastasia Orlova, Illarion Iov et al.

Benchmarking outcomes increasingly govern trust, selection, and deployment of LLMs, yet these evaluations remain vulnerable to semantically equivalent adversarial perturbations. Prior work on adversarial robustness in NLP has emphasized text attacks that affect many models equally, leaving open the question of whether it is possible to selectively degrade or enhance performance while minimally affecting other models. We formalize this problem and study selective adversarial attacks on MMLU - a widely used benchmark designed to measure a language model's broad general knowledge and reasoning ability across different subjects. Using canonical attacks integrated into TextAttack framework, we introduce a protocol for selectivity assessment, develop a custom constraint to increase selectivity of attacks and propose a surrogate-LLM pipeline that generates selective perturbations. Empirically, we find that selective adversarial attacks exist and can materially alter relative rankings, challenging the fairness, reproducibility, and transparency of leaderboard-driven evaluation. Our results motivate perturbation-aware reporting and robustness diagnostics for LLM evaluation and demonstrate that even subtle edits can shift comparative judgments.

Andrey Goncharov, Nikolai Kondusov, Alexey Zaytsev

Multilingual Large Language Models (LLMs) often exhibit unintended code-switching, reducing reliability in downstream tasks. We propose latent-space language steering, a lightweight inference-time method that identifies language directions via PCA on parallel translations and steers token embeddings along these axes to control language identity. Our approach mitigates code-switching while preserving semantics with negligible computational overhead and requires only minimal parallel data for calibration. Empirically, we achieve 95-99\% language classification accuracy using a single principal component and reduce next-token distributional divergence by up to 42% across multiple language pairs on Qwen2.5 and Llama-3.2 models. We further analyze the layer-wise evolution of language representations, revealing that language identity concentrates in final layers with near-perfect linear separability.

Andrey Goncharov, Daniil Vyazhev, Petr Sychev et al.

General-purpose Large Language Models (LLMs) are frequently fine-tuned through supervised fine-tuning (SFT) to enhance performance in specific domains. Better results can be achieved by distilling the chain-of-thought of a larger model at the cost of numerous expensive calls and a much greater amount of data. We propose a novel blueprint for efficient fine-tuning that uses reasoning only for complex data identified by entropy. Specifically, across two small open models ($~3B$) we split the training data into complexity categories by a single token answer entropy (ROC AUC $0.73$), fine-tune large language models (LLMs) via SFT and distillation, and show that our pipeline significantly outperforms the standard SFT approach ($0.58$ vs $0.45$ average accuracy) and outperforms the distillation approach ($0.58$ vs $0.56$ average accuracy) while using $81%$ less data.

Alexander Marusov, Aleksandr Yugay, Alexey Zaytsev

Self-supervised learning (SSL) has emerged as a powerful approach to learning representations, particularly in the field of computer vision. However, its application to dependent data, such as temporal and spatio-temporal domains, remains underexplored. Besides, traditional contrastive SSL methods often assume \emph{semantic independence between samples}, which does not hold for dependent data exhibiting complex correlations. We propose a novel theoretical framework for contrastive SSL tailored to \emph{continuous dependent data}, which allows the nearest samples to be semantically close to each other. In particular, we propose two possible \textit{ground truth similarity measures} between objects -- \emph{hard} and \emph{soft} closeness. Under it, we derive an analytical form for the \textit{estimated similarity matrix} that accommodates both types of closeness between samples, thereby introducing dependency-aware loss functions. We validate our approach, \emph{Dependent TS2Vec}, on temporal and spatio-temporal downstream problems. Given the dependency patterns presented in the data, our approach surpasses modern ones for dependent data, highlighting the effectiveness of our theoretically grounded loss functions for SSL in capturing spatio-temporal dependencies. Specifically, we outperform TS2Vec on the standard UEA and UCR benchmarks, with accuracy improvements of $4.17$\% and $2.08$\%, respectively. Furthermore, on the drought classification task, which involves complex spatio-temporal patterns, our method achieves a $7$\% higher ROC-AUC score.

Alexander Stepikin, Evgenia Romanenkova, Alexey Zaytsev

Change Point Detection (CPD) aims to identify moments of abrupt distribution shifts in data streams. Real-world high-dimensional CPD remains challenging due to data pattern complexity and violation of common assumptions. Resorting to standalone deep neural networks, the current state-of-the-art detectors have yet to achieve perfect quality. Concurrently, ensembling provides more robust solutions, boosting the performance. In this paper, we investigate ensembles of deep change point detectors and realize that standard prediction aggregation techniques, e.g., averaging, are suboptimal and fail to account for problem peculiarities. Alternatively, we introduce WWAggr -- a novel task-specific method of ensemble aggregation based on the Wasserstein distance. Our procedure is versatile, working effectively with various ensembles of deep CPD models. Moreover, unlike existing solutions, we practically lift a long-standing problem of the decision threshold selection for CPD.

Artem Zabolotnyi, Roman Makarov, Mile Mitrovic et al.

Uncertainty estimation remains a critical challenge in adapting pre-trained language models to classification tasks, particularly under parameter-efficient fine-tuning approaches such as adapters. We introduce AdUE1, an efficient post-hoc uncertainty estimation (UE) method, to enhance softmax-based estimates. Our approach (1) uses a differentiable approximation of the maximum function and (2) applies additional regularization through L2-SP, anchoring the fine-tuned head weights and regularizing the model. Evaluations on five NLP classification datasets across four language models (RoBERTa, ELECTRA, LLaMA-2, Qwen) demonstrate that our method consistently outperforms established baselines such as Mahalanobis distance and softmax response. Our approach is lightweight (no base-model changes) and produces better-calibrated confidence.

Alexander Panyshev, Dmitry Vinichenko, Oleg Travkin et al.

Temporal Graph Networks (TGNs), while being accurate, face significant training inefficiencies due to irregular supervision signals in dynamic graphs, which induce sparse gradient updates. We first theoretically establish that aggregating historical node interactions into pseudo-labels reduces gradient variance, accelerating convergence. Building on this analysis, we propose History-Averaged Labels (HAL), a method that dynamically enriches training batches with pseudo-targets derived from historical label distributions. HAL ensures continuous parameter updates without architectural modifications by converting idle computation into productive learning steps. Experiments on the Temporal Graph Benchmark (TGB) validate our findings and an assumption about slow change of user preferences: HAL accelerates TGNv2 training by up to 15x while maintaining competitive performance. Thus, this work offers an efficient, lightweight, architecture-agnostic, and theoretically motivated solution to label sparsity in temporal graph learning.

Petr Sokerin, Dmitry Anikin, Sofia Krehova et al.

The emergence of deep learning led to the broad usage of neural networks in the time series domain for various applications, including finance and medicine. While powerful, these models are prone to adversarial attacks: a benign targeted perturbation of input data leads to significant changes in a classifier's output. However, formally small attacks in the time series domain become easily detected by the human eye or a simple detector model. We develop a concealed adversarial attack for different time-series models: it provides more realistic perturbations, being hard to detect by a human or model discriminator. To achieve this goal, the proposed adversarial attack maximizes an aggregation of a classifier and a trained discriminator loss. To make the attack stronger, we also propose a training procedure for a discriminator that provides broader coverage of possible attacks. Extensive benchmarking on six UCR time series datasets across four diverse architectures - including recurrent, convolutional, state-space, and transformer-based models - demonstrates the superiority of our attack for a concealability-efficiency trade-off. Our findings highlight the growing challenge of designing robust time series models, emphasizing the need for improved defenses against realistic and effective attacks.