arXiv subjects

Alexey Zaytsev

Publications and source records attributed to Alexey Zaytsev.

At least 37 records · Page 2Linked to original sources

arXiv2026

Position: agentic AI orchestration should be Bayes-consistent

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.

cs.AI↗

arXiv2026

Language steering in latent space to mitigate unintended code-switching

Multilingual Large Language Models (LLMs) often exhibit hallucinations such as unintended code-switching, reducing reliability in downstream tasks. We propose latent-space language steering, a lightweight inference-time method that identifies language directions via Principal Component Analysis (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 55\% across multiple language pairs on Qwen2.5 and Llama-3.2 models. Generation-based evaluation on Llama-3.2 further demonstrates 63--99\% reduction in Code-Switching Index across four language pairs ($p < 0.001$). We further analyze the layer-wise evolution of language representations, revealing that language identity concentrates in final layers with near-perfect linear separability.

cs.CL↗

arXiv2026

ALIEN: Aligned Entropy Head for Improving Uncertainty Estimation of LLMs

Uncertainty estimation remains a key challenge when adapting pre-trained language models to downstream classification tasks, with overconfidence often observed for difficult inputs. While predictive entropy provides a strong baseline for uncertainty estimation, it considers mainly aleatoric uncertainty and has limited capacity to capture effects, such as class overlap or ambiguous linguistic cues. We introduce Aligned Entropy - ALIEN, a lightweight method that refines entropy-based uncertainty by aligning it with prediction reliability. ALIEN trains a small uncertainty head initialized to produce the model's original entropy and subsequently fine-tuned with two regularization mechanisms. Experiments across seven classification datasets and two NER benchmarks, evaluated on five language models (RoBERTa, ELECTRA, LLaMA-2, Qwen2.5, and Qwen3), show that ALIEN consistently outperforms strong baselines across all considered scenarios in detecting incorrect predictions, while achieving the lowest calibration error. The proposed method introduces only a small inference overhead (in the order of milliseconds per batch on CPU) and increases the model's parameter count by just 0.002% for decoder models and 0.5% for encoder models, without requiring storage of intermediate states. It improves uncertainty estimation while preserving the original model architecture, making the approach practical for large-scale deployment with modern language models. Our results demonstrate that entropy can be effectively refined through lightweight supervised alignment, producing more reliable uncertainty estimates without modifying the backbone model. The code is available at 4.

cs.CL↗

arXiv2026

Enhancing the Reliability of Medical AI through Expert-guided Uncertainty Modeling

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.

cs.LG↗

arXiv2026

Uncertainty Estimation for the Open-Set Text Classification systems

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

cs.CL↗

arXiv2026

Probabilistic distances-based hallucination detection in LLMs with RAG

Detecting hallucinations in large language models (LLMs) is critical for their safety in many applications. Without proper detection, these systems often provide harmful, unreliable answers. In recent years, LLMs have been actively used in retrieval-augmented generation (RAG) settings. However, hallucinations remain even in this setting, and while numerous hallucination detection methods have been proposed, most approaches are not specifically designed for RAG systems. To overcome this limitation, we introduce a hallucination detection method based on estimating the distances between the distributions of prompt token embeddings and language model response token embeddings. The method examines the geometric structure of token hidden states to reliably extract a signal of factuality in text, while remaining friendly to long sequences. Extensive experiments demonstrate that our method achieves state-of-the-art or competitive performance. It also has transferability from solving the NLI task to the hallucination detection task, making it a fully unsupervised and efficient method with a competitive performance on the final task.

cs.CL↗

arXiv2026

Complexity-aware fine-tuning

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 three small open models ($\approx 3B$) 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.

cs.LG↗

arXiv2026

U-Former ODE: Fast Probabilistic Forecasting of Irregular Time Series

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.

cs.LG↗

arXiv2026

Strong Linear Baselines Strike Back: Closed-Form Linear Models as Gaussian Process Conditional Density Estimators for TSAD

Research in time series anomaly detection (TSAD) has largely focused on developing increasingly sophisticated, hard-to-train, and expensive-to-infer neural architectures. We revisit this paradigm and show that a simple linear autoregressive anomaly score with the closed-form solution provided by ordinary least squares (OLS) regression consistently matches or outperforms state-of-the-art deep detectors. From a theoretical perspective, we show that linear models capture a broad class of anomaly types, estimating a finite-history Gaussian process conditional density. From a practical side, across extensive univariate and multivariate benchmarks, the proposed approach achieves superior accuracy while requiring orders of magnitude fewer computational resources. Thus, future research should consistently include strong linear baselines and, more importantly, develop new benchmarks with richer temporal structures pinpointing the advantages of deep learning models.

cs.LG↗

arXiv2026

DeNOTS: Stable Deep Neural ODEs for Time Series

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.

cs.LG↗

arXiv2025

Foundation for unbiased cross-validation of spatio-temporal models for species distribution modeling

Evaluating the predictive performance of species distribution models (SDMs) under realistic deployment scenarios requires careful handling of spatial and temporal dependencies in the data. Cross-validation (CV) is the standard approach for model evaluation, but its design strongly influences the validity of performance estimates. When SDMs are intended for spatial or temporal transfer, random CV can lead to overoptimistic results due to spatial autocorrelation (SAC) among neighboring observations. We benchmark four machine learning algorithms (GBM, XGBoost, LightGBM, Random Forest) on two real-world presence-absence datasets, a temperate plant and an anadromous fish, using multiple CV designs: random, spatial, spatio-temporal, environmental, and forward-chaining. Two training data usage strategies (LAST FOLD and RETRAIN) are evaluated, with hyperparameter tuning performed within each CV scheme. Model performance is assessed on independent out-of-time test sets using AUC, MAE, and correlation metrics. Random CV overestimates AUC by up to 0.16 and produces MAE values up to 80 percent higher than spatially blocked alternatives. Blocking at the empirical SAC range substantially reduces this bias. Training strategy affects evaluation outcomes: LAST FOLD yields smaller validation-test discrepancies under strong SAC, while RETRAIN achieves higher test AUC when SAC is weaker. Boosted ensemble models consistently perform best under spatially structured CV designs. We recommend a robust SDM workflow based on SAC-aware blocking, blocked hyperparameter tuning, and external temporal validation to improve reliability under spatial and temporal shifts.

stat.AP↗

arXiv2025

PINE: Pipeline for Important Node Exploration in Attributed Networks

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.

cs.LG↗

arXiv2025

Parameter-Efficient Neural CDEs via Implicit Function Jacobians

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.

cs.LG↗

arXiv2025

Selective Adversarial Attacks on LLM Benchmarks

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.

cs.LG↗

arXiv2025

WWAggr: A Window Wasserstein-based Aggregation for Ensemble Change Point Detection

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.

stat.ML↗

arXiv2025

A theoretical framework for self-supervised contrastive learning for continuous dependent data

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.

cs.LG↗

arXiv2025

InDiD: Instant Disorder Detection via Representation Learning

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 data with change point moments for video data 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.

cs.LG↗

arXiv2025

Holistic Uncertainty Estimation For Open-Set Recognition

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.

cs.CV↗