time-series

The paper introduces STCFormer, an adaptive spatio-temporal Transformer that dynamically clusters weather stations per temporal patch using a Cluster-Guided Attention Block combining local intra-cluster attention with global regional summaries, achieving the lowest 24-hour MSE across all eight temperature and wind forecasting tasks.

The paper proposes GenScript, an inference-only synthetic data pipeline that replaces per-dataset model training by passing a deterministic statistical profile of the source data to an LLM, which infers field semantics and integrity constraints that a coding agent then compiles into an auditable sampler. The approach unifies tabular, temporal, and relational data synthesis, building generators in minutes while closely matching leading methods in fidelity and preserving key integrity constraints.

DualCast is a dual-path LLM framework that extends a frozen Qwen3-8B backbone with a discrete financial vocabulary for bimodal financial time-series forecasting, combining a fast numerical path with a news-conditioned slow revision path optimized via group relative policy optimization.

Researchers from Allora Foundation and Liverpool John Moores University introduce the CZAR (Composite Zero-Agnostic Return) loss, a convex, asymmetric, piecewise-quadratic objective tailored for financial log-return prediction that prevents models from collapsing to zero-forecast behavior, and show improved directional accuracy on BTC intraday returns with LightGBM.

CADENCE 提出了一种置信度自适应的双专家网络，用于实现快速且准确的时间序列分类，是一篇 arXiv 上的学术预印本。

EXAONE Demand 1.0 is a time series foundation model designed for demand forecasting, utilizing a demand-specific corpus and adapter to outperform existing models on 22 datasets.

The paper proposes XACT, a framework for learning sparse attribution masks over time-frequency transforms to explain time-series classifiers, demonstrating improved precision and interpretability over baselines.

This paper proposes a method for improving frozen forecasters by combining static and online correctors to control downside risk, demonstrating gains up to 11.5% in benchmarks and reduced errors in electricity load forecasting.

The paper introduces fable.intermittent, an R package for benchmarking probabilistic forecasting methods for intermittent time series, and presents TWEES, a new exponential smoothing model using Tweedie distribution.

The paper proposes the Slicing-Graphing-Alignment (SGA) method to quantify uncertainty in multi-step forecasting for time series foundation models, demonstrating superior performance and revealing an empirical scaling law.

The paper proposes VINTAGE-TS, a revision-aware adaptation of a time-series foundation model that distinguishes observation time from information-availability time, and provides a framework for evaluating forecasts with data revisions.

This paper identifies blind spots in evaluating deep imbalanced regression, proposing balanced metrics and showing high tail-region instability across random seeds.

This paper introduces Event Signature Transfer (EST), a training-free, model-agnostic operator that constructs forecast scenarios by transferring event signatures from historical events onto time-series forecasts.

This paper introduces a synthetic multivariate time-series dataset for refrigerator predictive maintenance, generated using a physics-inspired simulator to support failure prediction and degradation analysis.

UniGIO is a generative framework that models global in-situ weather dynamics from incomplete observations, unifying forecasting, imputation, and generation tasks. It achieves state-of-the-art performance on the Weather-5K dataset with significant improvements in accuracy and extreme event capture.

This paper proposes a source-first taxonomy for zero-shot time-series forecasting, categorizing evidence sources and introducing an audit framework to improve benchmark transparency.

This paper proposes a multimodal anomaly detection framework for fault detection in mechanical systems using self-supervised cross-modal reconstruction and adaptive thresholding to improve robustness under distribution shifts.

A novel framework leveraging LLMs to generate synthetic time series data for manufacturing, demonstrating improved performance over traditional modeling techniques in downstream tasks like anomaly detection.

The article proposes a simpler one-dimensional version of time map plots for visualizing discrete events across multiple timescales, demonstrating it with last.fm scrobbles data.

DCRA introduces a diffusion-conditioned representation alignment framework to enhance robustness in time-series learning, particularly for clinical signals like EEG and ECG, by aligning representations across noise levels.