arXiv subjects

Oleg Smirnov

Publications and source records attributed to Oleg Smirnov.

At least 19 recordsLinked to original sources

arXiv2026

Towards Unified Approaches in Self-Supervised Event Stream Modeling: Progress and Prospects

The proliferation of digital interactions across diverse domains, such as healthcare, e-commerce, gaming, and finance, has resulted in the generation of vast volumes of event stream (ES) data. ES data comprises continuous sequences of timestamped events that encapsulate detailed contextual information relevant to each domain. While ES data holds significant potential for extracting actionable insights and enhancing decision-making, its effective utilization is hindered by challenges such as the scarcity of labeled data and the fragmented nature of existing research efforts. Self-Supervised Learning (SSL) has emerged as a promising paradigm to address these challenges by enabling the extraction of meaningful representations from unlabeled ES data. In this survey, we systematically review and synthesize SSL methodologies tailored for ES modeling across multiple domains, bridging the gaps between domain-specific approaches that have traditionally operated in isolation. We present a comprehensive taxonomy of SSL techniques, encompassing both predictive and contrastive paradigms, and analyze their applicability and effectiveness within different application contexts. Furthermore, we identify critical gaps in current research and propose a future research agenda aimed at developing scalable, domain-agnostic SSL frameworks for ES modeling. By unifying disparate research efforts and highlighting cross-domain synergies, this survey aims to accelerate innovation, improve reproducibility, and expand the applicability of SSL to diverse real-world ES challenges.

cs.LG↗

arXiv2026

Few-Shot Out of Domain Intent Detection with Covariance Corrected Mahalanobis Distance

Conversational agents like chatbots and voice assistants are trained to understand and respond to user intents. On encountering an utterance with an intent different from the ones they have been trained on, these agents are expected to classify the intent as `unknown' or `out of domain'. This problem is known as out of domain (OOD) intent detection. Podolskiy et al. (2021), showed that Mahalanobis distance can be used effectively for identifying OOD intents, outperforming competing approaches. However, their method fails to outperform the baselines in the practically important few-shot setting. In this paper we analyze the reason for low performance and propose a covariance corrected Mahalanobis distance for detecting out-of-domain intents.

cs.AI↗

arXiv2026

The Language You Ask In: Language-Conditioned Ideological Divergence in LLM Analysis of Contested Political Documents

Large language models are increasingly used to interpret politically contested questions, value-laden material on which there is no single correct answer, only competing interpretive traditions. We ask whether a model's choice among those traditions can turn on the language of the prompt rather than the content. Comparing two frontier models, ChatGPT 5.2 and Claude Opus 4.5, on one contested Ukrainian civil-society document under semantically matched Russian and Ukrainian prompts, we find that both shift along the same axis on identical source text: Russian prompts elicit delegitimizing readings of the document's authors and Ukrainian prompts legitimating ones. The magnitude is model-dependent but neither model is neutral: each adopts a language-dependent stance, and the difference is one of degree. Because contested political questions admit no correct reading against which to measure, we read this as language-conditioned variation in which interpretive tradition a model activates: the model neither holds a single stance nor surfaces the plurality of available ones, but silently adopts the dominant frame of the prompt's language. We draw out the consequences for pluralism-aware evaluation, which must probe the same content across the languages a model serves, and for pluralistic alignment in multilingual settings.

cs.CY↗

arXiv2026

Causality for Tabular Data Synthesis: A High-Order Structure Causal Benchmark Framework

Existing evaluations of tabular synthesis models rely primarily on low-order statistics and downstream task performance, leaving multivariate causal relationships that go beyond pairwise correlations largely unmeasured. We argue that a systematic evaluation on high-order structural information is a crucial first step in addressing this issue in tabular data synthesis. In this paper, we present high-order structural causal information as a natural form of prior knowledge and introduce a benchmark framework to evaluate tabular synthesis models. This framework allows us to generate benchmark datasets through a flexible range of data generation processes, allowing for the training of tabular synthesis models using these datasets for further evaluation. We propose multiple benchmark tasks, high-order metrics, and causal inference tasks as downstream tasks for evaluating the quality of synthetic data generated by the trained models. Our experiments demonstrate the effectiveness of the benchmark framework in evaluating the model's ability to capture high-order structural causal information. Furthermore, our benchmarking results provide an initial assessment of state-of-the-art tabular synthesis models. These results reveal significant gaps between ideal and actual performance and highlight how baseline methods differ. We position the framework as a controlled diagnostic benchmark for causal fidelity, complementing existing low-order and downstream evaluations. We open source the benchmark framework, including both code and data along with documentation, to support further research in this area.

cs.LG↗

arXiv2026

Tracing Large-scale Structure with the MeerKLASS On-the-Fly Survey: Angular Clustering of Radio Sources at 816 MHz

We present the first measurement of the angular two-point correlation function $w(θ)$ of radio sources from the MeerKAT Large Area Synoptic Survey (MeerKLASS) UHF on-the-fly (OTF) continuum Data Release~1. DR1 provides interferometric Stokes-$I$ imaging at a reference frequency of 816\,MHz over $\sim 800~\mathrm{deg}^2$ within the DESI footprint. We detect a positive clustering signal over $0.02^\circ \lesssim θ\lesssim 10^\circ$. The measurement is stable to reasonable variations of the depth mask and flux threshold on intermediate and large scales. Modelling the intermediate-scale signal ($0.112^\circ\leθ\le 1.36^\circ$) with a fixed-slope power law ($γ=1.8$) yields $A(1^\circ)=(1.434\pm0.475)\times10^{-3}$, corresponding to $\log_{10}A=-2.843^{+0.124}_{-0.175}$. We infer an effective large-scale bias by fitting $Λ$CDM projected-matter templates $w_{\rm DM}(θ)$ computed with \textsc{CAMB} and Limber projection, including an integral-constraint correction evaluated from random--random weights. Using two bracketing T-RECS redshift-distribution priors, we obtain $b_{\rm eff}=1.998\pm0.350$ (AGN prior) and $b_{\rm eff}=1.530\pm0.265$ (TOTAL prior), demonstrating that the dominant modelling uncertainty arises from $N(z)$. As a derived summary we Limber-invert the power-law amplitude to obtain $r_0=6.18\pm1.13$ and $5.59\pm1.02~h^{-1}\mathrm{Mpc}$ for the AGN and TOTAL priors, respectively. These results establish MeerKLASS UHF DR1 as a new wide-area, intermediate-frequency dataset for radio-continuum clustering. As MeerKLASS expands and overlapping optical/IR spectroscopy provides improved redshift calibration, future releases will enable population-split clustering and bias evolution measurements with substantially reduced modelling uncertainty.

astro-ph.CO↗

arXiv2026

Be Wary of Your Time Series Preprocessing

Normalization and scaling are fundamental preprocessing steps in time series modeling, yet their role in Transformer-based models remains underexplored from a theoretical perspective. In this work, we present the first formal analysis of how different normalization strategies, specifically instance-based and global scaling, impact the expressivity of Transformer-based architectures for time series representation learning. We propose a novel expressivity framework tailored to time series, which quantifies a model's ability to distinguish between similar and dissimilar inputs in the representation space. Using this framework, we derive theoretical bounds for two widely used normalization methods: Standard and Min-Max scaling. Our analysis reveals that the choice of normalization strategy can significantly influence the model's representational capacity, depending on the task and data characteristics. We complement our theory with empirical validation on classification and forecasting benchmarks using multiple Transformer-based models. Our results show that no single normalization method consistently outperforms others, and in some cases, omitting normalization entirely leads to superior performance. These findings highlight the critical role of preprocessing in time series learning and motivate the need for more principled normalization strategies tailored to specific tasks and datasets.

cs.LG↗

arXiv2026

Mechanisms Affecting Galaxies Nearby and Environmental Trends (MAGNET)

[ABRIDGED] Galaxy evolution is shaped by internal and external mechanisms that regulate the baryon cycle and star formation activity. We present a theoretical framework based on the GAlaxy Evolution and Assembly (GAEA) semi-analytic model. We extracted portions of simulated volumes that include isolated galaxies, pairs, group, and filament members at z ~ 0, specifically avoiding massive clusters. Galaxies were classified using both intrinsic (halo-based) and observational (2D projected) parameterizations, reconstructing their environmental histories from z = 2 and identifying mergers, tidal interactions, ram pressure stripping (RPS), and starvation. 2D information decreases isolated and group fractions while doubles pairs. More than half of galaxies remain unaffected by the investigated processes since z = 2. Among affected galaxies, mergers dominate at high stellar masses (40-60% at log(M*/Msun) > 10.5). Tidal interactions are less frequent, and their incidence increases with stellar mass. RPS dominates in groups and filaments at intermediate masses (~50%), while starvation ranges from 20 to 30%. The incidence of the different mechanisms depends strongly on both mass and environment, though their imprints on global properties are often subtle. Distinct evolutionary pathways emerge: log(M*/Msun) < 9.5, galaxies in groups and filaments have a faster mass growth than galaxies in the other environments, especially those undergoing starvation, mergers and, to less extent, RPS. Differences are reduced moving to higher masses, where no clear dependence on physical mechanism emerge, even though at these masses a clear star formation suppression is evident in mergers and starved galaxies. This theoretical investigation provides essential context for the recently started multi-wavelength program Mechanisms Affecting Galaxies Nearby and Environmental Trends (MAGNET), which we introduce here.

astro-ph.GA↗

arXiv2025

A direction-dependent framework for visibility plane mosaicing and primary beam correction

With the increasing sensitivity of modern radio interferometers, it has become important to image objects larger than the field of view while optimising sensitivity and image fidelity. We present a coherent visibility plane direction-dependent imaging, calibration and mosaicing framework. Our simulations and application to real MeerKAT data show that this joint deconvolution and primary beam correction approach, coupled with direction-dependent calibration, allows for deeper mosaics with greater fidelity and increased accuracy of recovered flux densities and spectral indices, especially beyond the half-power beam width. Our best-case mosaic produces precise flux values within a 6% uncertainty and spectral indices within 20\% throughout the imaged area, and is fully complete out to twice the radii and half the flux density than the image plane equivalent. The application to archival wideband MeerKAT 1283 MHz data produces the deepest high-resolution image of the Shapley Supercluster Core, with a sensitivity of 3.6 $μ$Jy/beam within the primary beam at a 7$^{\prime\prime}$ resolution, constituting a $\sim$ 50% increase in dynamic range over the image plane counterpart, and a fluxscale that is consistent within 10% across the entire field of view. The compute time for the direction-dependent visibility plane mosaic was comparable to the sum of the times needed to perform direction-dependent calibration on the individual pointings. Our results suggest that visibility plane mosaicing with its capability for deeper deconvolution could improve the efficiency of deep and wide surveys, particularly for on-the-fly mapping and studies of low surface brightness sources, and could form the basis of future calibration pipelines for SKA-scale instruments.

astro-ph.IM↗

arXiv2025

tensorflow-riemopt: A Library for Optimization on Riemannian Manifolds

This paper presents tensorflow-riemopt, a Python library for geometric machine learning in TensorFlow. The library provides efficient implementations of neural network layers with manifold-constrained parameters, geometric operations on Riemannian manifolds, and stochastic optimization algorithms for non-Euclidean spaces. Designed for integration with TensorFlow Extended, it supports both research prototyping and production deployment of machine learning pipelines. The code and documentation are distributed under the MIT license and available at https://github.com/master/tensorflow-riemopt

cs.MS↗

arXiv2025

The MeerKLASS UHF On-the-Fly Continuum Survey -- Data Release I

We present the first public data release (DR1) from the interferometric component of the MeerKAT Large Area Synoptic Survey (MeerKLASS) UHF survey, a legacy program demonstrating a novel on-the-fly (OTF) mapping technique. This release is based on 12 hours of early science observations covering approximately 800 deg$^2$ of the southern sky. We describe the data processing pipeline developed to calibrate and image these fast-scanning observations, producing high-fidelity continuum images at a central frequency of 816 MHz. The resulting mosaic reaches an RMS sensitivity of $\sim$35 $μ$Jy beam$^{-1}$ in its deepest regions, with a typical angular resolution of $\sim32'' \times 17''$. In these images, we identify $95483$ radio sources. We validate the catalogue through cross-matching with external surveys, confirming sub-arcsecond astrometric accuracy and a robust flux density scale. We compute the differential source counts, finding excellent agreement with existing measurements and validating our end-to-end processing. The success of this pilot study serves as a crucial proof of concept for the OTF observing strategy, and the public release of the images and source catalogue provides a valuable resource for a wide range of astrophysical studies. This work paves the way for the full MeerKLASS OTF survey and future large-area survey projects with the SKA.

astro-ph.GA↗

arXiv2025

Frequency Matters: When Time Series Foundation Models Fail Under Spectral Shift

Time series foundation models (TSFMs) have shown strong results on public benchmarks, prompting comparisons to a "BERT moment" for time series. Their effectiveness in industrial settings, however, remains uncertain. We examine why TSFMs often struggle to generalize and highlight spectral shift (a mismatch between the dominant frequency components in downstream tasks and those represented during pretraining) as a key factor. We present evidence from an industrial-scale player engagement prediction task in mobile gaming, where TSFMs underperform domain-adapted baselines. To isolate the mechanism, we design controlled synthetic experiments contrasting signals with seen versus unseen frequency bands, observing systematic degradation under spectral mismatch. These findings position frequency awareness as critical for robust TSFM deployment and motivate new pretraining and evaluation protocols that explicitly account for spectral diversity.

cs.LG↗

arXiv2025

Enhancing Graph Classification Robustness with Singular Pooling

Graph Neural Networks (GNNs) have achieved strong performance across a range of graph representation learning tasks, yet their adversarial robustness in graph classification remains underexplored compared to node classification. While most existing defenses focus on the message-passing component, this work investigates the overlooked role of pooling operations in shaping robustness. We present a theoretical analysis of standard flat pooling methods (sum, average and max), deriving upper bounds on their adversarial risk and identifying their vulnerabilities under different attack scenarios and graph structures. Motivated by these insights, we propose \textit{Robust Singular Pooling (RS-Pool)}, a novel pooling strategy that leverages the dominant singular vector of the node embedding matrix to construct a robust graph-level representation. We theoretically investigate the robustness of RS-Pool and interpret the resulting bound leading to improved understanding of our proposed pooling operator. While our analysis centers on Graph Convolutional Networks (GCNs), RS-Pool is model-agnostic and can be implemented efficiently via power iteration. Empirical results on real-world benchmarks show that RS-Pool provides better robustness than the considered pooling methods when subject to state-of-the-art adversarial attacks while maintaining competitive clean accuracy. Our code is publicly available at:\href{https://github.com/king/rs-pool}{https://github.com/king/rs-pool}.

cs.LG↗

arXiv2025

Pool Me Wisely: On the Effect of Pooling in Transformer-Based Models

Transformer models have become the dominant backbone for sequence modeling, leveraging self-attention to produce contextualized token representations. These are typically aggregated into fixed-size vectors via pooling operations for downstream tasks. While much of the literature has focused on attention mechanisms, the role of pooling remains underexplored despite its critical impact on model behavior. In this paper, we introduce a theoretical framework that rigorously characterizes the expressivity of Transformer-based models equipped with widely used pooling methods by deriving closed-form bounds on their representational capacity and the ability to distinguish similar inputs. Our analysis extends to different variations of attention formulations, demonstrating that these bounds hold across diverse architectural variants. We empirically evaluate pooling strategies across tasks requiring both global and local contextual understanding, spanning three major modalities: computer vision, natural language processing, and time-series analysis. Results reveal consistent trends in how pooling choices affect accuracy, sensitivity, and optimization behavior. Our findings unify theoretical and empirical perspectives, providing practical guidance for selecting or designing pooling mechanisms suited to specific tasks. This work positions pooling as a key architectural component in Transformer models and lays the foundation for more principled model design beyond attention alone.

cs.LG↗

arXiv2025

Prompt Tuning Decision Transformers with Structured and Scalable Bandits

Prompt tuning has emerged as a key technique for adapting large pre-trained Decision Transformers (DTs) in offline Reinforcement Learning (RL), particularly in multi-task and few-shot settings. The Prompting Decision Transformer (PDT) enables task generalization via trajectory prompts sampled uniformly from expert demonstrations -- without accounting for prompt informativeness. In this work, we propose a bandit-based prompt-tuning method that learns to construct optimal trajectory prompts from demonstration data at inference time. We devise a structured bandit architecture operating in the trajectory prompt space, achieving linear rather than combinatorial scaling with prompt size. Additionally, we show that the pre-trained PDT itself can serve as a powerful feature extractor for the bandit, enabling efficient reward modeling across various environments. We theoretically establish regret bounds and demonstrate empirically that our method consistently enhances performance across a wide range of tasks, high-dimensional environments, and out-of-distribution scenarios, outperforming existing baselines in prompt tuning.

cs.LG↗

arXiv2025

The MeerKAT Galaxy Cluster Legacy Survey -- II. Catalogue of the diffuse radio emission in MeerKAT-GCLS clusters

We present a follow-up study focused on cluster-scale diffuse radio emissions in 115 galaxy clusters in the Southern sky, using full-resolution (7.8") and tapered low-resolution (15") images from the MeerKAT Galaxy Cluster Legacy Survey (MGCLS). In this MGCLS follow-up paper, we update and quantitatively characterise the presence of diffuse radio emission and provide detailed catalogue information on its radio properties at 1.28 GHz. As the MGCLS sample is heterogeneous and was initially used as a test bed to reveal the scientific potential of MeerKAT, the reported numbers are subject to this special case. More than half ($\sim$54%; 62/115) of the observed MGCLS clusters present diffuse cluster radio emission, with the total number of diffuse radio sources or candidates detected being 103. Including candidates, we find that radio relics are the most frequently detected diffuse sources in MGCLS at 53% (55/103), followed by halos at 32% (33/103) and mini-halos at 10% (11/103), with only 3% (3/103) being Phoenixes. The sizes of the diffuse radio structures and candidates range from 55 kpc to over 2 Mpc, with $P_{1.28GHz}$ radio power ranging from $10^{22}$ W Hz$^{-1}$ to greater than $10^{25}$ W Hz$^{-1}$. In-band radio spectral index estimates reveal revived radio plasma emissions that exhibit steep radio spectral indices down to $α^{1656}_{908} \sim -3.5$. Mini-halos and their respective candidates are detected mainly in low-mass ($M_{500}$ $\leq$ 5$\times10^{14}$ M$_\odot$) and low-power ($P_{1.4GHz}$ $\leq$ $10^{23}$ W Hz$^{-1}$) systems. We suggest the presence of a statistically significant correlation between the 1.4 GHz radio power and the cluster mass for mini-halo (and candidate) systems.

astro-ph.GA↗

arXiv2025

HI gas in the rejuvenated radio galaxy PKS 2014-55

We present new high-spectral-resolution MeerKAT observations of $\mathrm{HI}$ absorption against the central region of the restarted, giant, X-shaped radio galaxy PKS2014-55, which exhibits morphological evidence of three distinct cycles of activity. We report a wide component $(\mathrm{FWHM}=38 \pm 7~\mathrm{kms^{-1}})$ redshifted to $96 \pm 50~\mathrm{kms^{-1}}$, a deep narrow detection $(\mathrm{FWHM}=19 \pm 6~\mathrm{kms^{-1}})$ redshifted to $160 \pm 40~\mathrm{kms^{-1}}$, and a shallow component $(\mathrm{FWHM}=22 \pm 6~\mathrm{kms^{-1}})$ redshifted to $240 \pm 40~\mathrm{kms^{-1}}$. One of the three components exceeds the typical rotational velocity of $100~\mathrm{kms^{-1}}$, suggesting complex kinematics of the inflowing gas. These $\mathrm{HI}$ observations support the correlation between the occurrence of $\mathrm{HI}$ absorption and the rejuvenation of radio activity.

astro-ph.GA↗

arXiv2025

Prompt-Tuning Bandits: Enabling Few-Shot Generalization for Efficient Multi-Task Offline RL

Prompting has emerged as the dominant paradigm for adapting large, pre-trained transformer-based models to downstream tasks. The Prompting Decision Transformer (PDT) enables large-scale, multi-task offline Reinforcement Learning (RL) pre-training by leveraging stochastic trajectory prompts to identify the target task. However, these prompts are sampled uniformly from expert demonstrations, overlooking a critical limitation: not all prompts are equally informative for differentiating between tasks. This limits generalization and adaptation, especially in low-data or open-world settings where sample efficiency is crucial. To address this issue, we propose a lightweight, inference-time, bandit-based prompt-tuning framework. The bandit explores and optimizes trajectory prompt selection to enhance task performance, while avoiding costly fine-tuning of the transformer backbone. Our experiments indicate not only clear performance gains due to bandit-based prompt-tuning, but also better sample complexity, scalability, and prompt space exploration compared to prompt-tuning baselines. These results highlights the importance of adaptive prompt selection mechanisms for efficient generalization in offline multi-task RL.

cs.LG↗

arXiv2025

Are We Really Measuring Progress? Transferring Insights from Evaluating Recommender Systems to Temporal Link Prediction

Recent work has questioned the reliability of graph learning benchmarks, citing concerns around task design, methodological rigor, and data suitability. In this extended abstract, we contribute to this discussion by focusing on evaluation strategies in Temporal Link Prediction (TLP). We observe that current evaluation protocols are often affected by one or more of the following issues: (1) inconsistent sampled metrics, (2) reliance on hard negative sampling often introduced as a means to improve robustness, and (3) metrics that implicitly assume equal base probabilities across source nodes by combining predictions. We support these claims through illustrative examples and connections to longstanding concerns in the recommender systems community. Our ongoing work aims to systematically characterize these problems and explore alternatives that can lead to more robust and interpretable evaluation. We conclude with a discussion of potential directions for improving the reliability of TLP benchmarks.

cs.LG↗