Alexandra Bazarova, Andrei Volodichev, Aleksandr Yugay, Andrey Shulga, Alina Ermilova, Konstantin Polev, Julia Belikova, Rauf Parchiev, Dmitry Simakov, Maxim Savchenko, Andrey Savchenko, Serguei Barannikov, Alexey Zaytsev. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

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 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 continuous dependent data, which allows the nearest samples to be semantically close to each other. In particular, we propose two possible ground truth similarity measures between objects — hard and soft closeness. Under it, we derive an analytical form for the estimated similarity matrix that accommodates both types of closeness between samples, thereby introducing dependency-aware loss functions. We validate our approach, 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.

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-theart 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.

AI-driven second-opinion systems play a crucial role in decision-making, especially in medicine, where accurate predictions guide clinicians. However, quantifying uncertainty in deep learning is challenging, as current methods often rely on hard class labels, which do not reflect true prediction confidence. This often results in overconfident predictions and slow convergence to true probabilities. To address this, we suggest a new method that separates uncertainty into two types: epistemic and aleatoric. We estimate these uncertainties using hard and soft confidence labels, with experts providing confidence levels that indicate the likelihood of misclassification. We release an updated blood typing dataset consisting of 3139 images with soft labels of uncertainty annotations from six experts and hard labels collected from medical records. Proposed approach improves SotA uncertainty estimation quality by two times for blood typing (classification) and by 62% for histology (segmentation)11The code is available at: https://github.com/createcolor/confidence-aware-uncertainty..

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.

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.

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.

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.

The performance of modern deep learning-based systems dramatically depends on the quality of input objects. For example, face recognition quality is lower for blurry or corrupted inputs. Moreover, it is difficult to predict the influence of input quality on the resulting accuracy in more complex scenarios. We propose a deep metric learning framework that allows for direct estimation of the uncertainty with almost no additional computational cost. The developed 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, thereby providing provably reasonable uncertainty estimation. We present results from comprehensive experiments on open-set classification tasks, including face recognition, which demonstrate the superior performance of ScaleFace compared to other uncertainty-aware face recognition approaches. We also extend our study to the task of text-to-image retrieval, showing that the proposed approach outperforms competitors by significant margins.

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.

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.

A memory efficient approach to ensembling neural networks is to share most weights among the ensembled models by means of a single reference network. We refer to this strategy as Embedded Ensembling (EE); its particular examples are BatchEnsembles and Monte-Carlo dropout ensembles. In this paper we perform a systematic theoretical and empirical analysis of embedded ensembles with different number of models. Theoretically, we use a Neural-Tangent-Kernel-based approach to derive the wide network limit of the gradient descent dynamics. In this limit, we identify two ensemble regimes - independent and collective - depending on the architecture and initialization strategy of ensemble models. We prove that in the independent regime the embedded ensemble behaves as an ensemble of independent models. We confirm our theoretical prediction with a wide range of experiments with finite networks, and further study empirically various effects such as transition between the two regimes, scaling of ensemble performance with the network width and number of models, and dependence of performance on a number of architecture and hyperparameter choices.

For sequential data, a change point is a moment of abrupt regime switch in data streams. Such changes appear in different scenarios, including simpler data from sensors and more challenging video surveillance data. We need to detect disorders as fast as possible. Classic approaches for change point detection (CPD) might underperform for semi-structured sequential data because they cannot process its structure without a proper representation. We propose a principled loss function that balances change detection delay and time to a false alarm. It approximates classic rigorous solutions but is differentiable and allows representation learning for deep models. We consider synthetic sequences, real-world data sensors and videos with change points. We carefully labelled available video data with change point moments and released it for the first time. Experiments suggest that complex data require meaningful representations tailored for the specificity of the CPD task --- and our approach provides them outperforming considered baselines. For example, for explosion detection in video, the F1 score for our method is 0.53 compared to baseline scores of 0.31 and 0.35.

We develop a neural network (NN) architecture aimed at the midterm prediction of earthquakes. Our data-based model aims to predict if an earthquake with a magnitude above a threshold takes place at a given small area of size 10 km$\times $10 km in a midterm range of 10–50 days from a given moment. Our deep NN model has a recurrent part long short term memory (LSTM) that accounts for time dependencies between earthquakes and a convolutional part that accounts for spatial dependencies. Obtained results show that NNs-based models beat baseline feature-based models that also account for spatio-temporal dependencies between different earthquakes. Moreover, each part of our network is essential for its quality. For historical data on Japan earthquakes, our model predicts the occurrence of an earthquake in a period of 10 to 50 days from a given moment with magnitude$M_{c} > 5$missing$2.09 \cdot 10^{3}$earthquakes out of$3.11 \cdot 10^{3}$and making$192 \cdot 10^{3}$false alarms. The baseline approach misses$2.07 \cdot 10^{3}$earthquakes but with a significantly higher number of false alarms$1004 \cdot 10^{3}$.

Macroeconomic indexes are of high importance for banks: many risk-control decisions utilize these indexes. A typical workflow of these indexes evaluation is costly and protracted, with a lag between the actual date and available index being a couple of months. Banks predict such indexes now using autoregressive models to make decisions in a rapidly changing environment. However, autoregressive models fail in complex scenarios related to appearances of crises. We propose to use clients’ financial transactions data from a large Russian bank to get such indexes. Financial transactions are long, and a number of clients is huge, so we develop an efficient approach that allows fast and accurate estimation of macroeconomic indexes based on a stream of transactions consisting of millions of transactions. The approach uses a neural networks paradigm and a smart sampling scheme.The results show that our neural network approach outperforms the baseline method on hand-crafted features based on transactions. Calculated embeddings show the correlation between the client’s transaction activity and bank macroeconomic indexes over time.

Machine learning models using transaction records as inputs are popular among financial institutions. The most efficient models use deep-learning architectures similar to those in the NLP community, posing a challenge due to their tremendous number of parameters and limited robustness. In particular, deep-learning models are vulnerable to adversarial attacks: a little change in the input harms the model's output. In this work, we examine adversarial attacks on transaction records data and defenses from these attacks. The transaction records data have a different structure than the canonical NLP or time-series data, as neighboring records are less connected than words in sentences, and each record consists of both discrete merchant code and continuous transaction amount. We consider a black-box attack scenario, where the attack doesn't know the true decision model and pay special attention to adding transaction tokens to the end of a sequence. These limitations provide a more realistic scenario, previously unexplored in the NLP world. The proposed adversarial attacks and the respective defenses demonstrate remarkable performance using relevant datasets from the financial industry. Our results show that a couple of generated transactions are sufficient to fool a deep-learning model. Further, we improve model robustness via adversarial training or separate adversarial examples detection. This work shows that embedding protection from adversarial attacks improves model robustness, allowing a wider adoption of deep models for transaction records in banking and finance.

During directional drilling, a bit may sometimes go to a nonproductive rock layer due to the gap about 20 m between the bit and high-fidelity rock-type sensors. The only way to detect the lithotype changes in time is the usage of measurements while drilling (MWD). However, there are no general mathematical modeling approaches that both well reconstruct the rock type based on MWD data and correspond to specifics of the oil and gas industry. In this letter, we present a data-driven procedure that utilizes MWD data for quick detection of changes in rock types. We propose the approach that combines traditional machine learning (ML) based on the solution of the rock-type classification problem with change detection procedures rarely used before in oil and gas industry. The data come from a newly developed oilfield in the north of western Siberia. The results suggest that we can detect a significant part of changes in rock types, reducing the change detection delay from 20 to 1.8 m and the number of false-positive alarms from 43 to 6 per well.

Engineering problems often involve data sources of variable fidelity with different costs of obtaining an observation. In particular, one can use both a cheap low fidelity function (e.g. a computational experiment with a CFD code) and an expensive high fidelity function (e.g. a wind tunnel experiment) to generate a data sample in order to construct a regression model of a high fidelity function. The key question in this setting is how the sizes of the high and low fidelity data samples should be selected in order to stay within a given computational budget and maximize accuracy of the regression model prior to committing resources on data acquisition. In this paper we obtain minimax interpolation errors for single and variable fidelity scenarios for a multivariate Gaussian process regression. Evaluation of the minimax errors allows us to identify cases when the variable fidelity data provides better interpolation accuracy than the exclusively high fidelity data for the same computational budget. These results allow us to calculate the optimal shares of variable fidelity data samples under the given computational budget constraint. Real and synthetic data experiments suggest that using the obtained optimal shares often outperforms natural heuristics in terms of the regression accuracy.

Most of the standard classification algorithms perform poorly when dealing with the case of imbalanced classes i.e. when there is a class to which the overwhelming majority of samples belong. There are many approaches that deal with this problem, among which SMOTE and SMOTE boosting, the common approach prefers overly simplistic models that lead to degradation of performance. Recent advances in statistical learning theory provide more adequate complexity penalties for weak classifiers, which stem from the Rademacher complexity terms in the ensemble generalization bounds. By adopting these advances and introducing a sample weight correction based on the classification margin at each iteration of boosting we get more precise models for imbalanced classification problems.