Tag
The paper presents a comparative study of forecasting models for neutron monitor time series, including seasonal baselines and quantum-inspired architectures, highlighting that simple and functional models perform well on periodic scientific data.
The paper presents TW3Cast, a time-series forecasting system that uses a frozen router of lightly fine-tuned foundation models to achieve top performance on the GIFT-Eval benchmark without agents or language models.
ROOSTER is a shared module that learns alignment between condition and target sequences for time-series forecasting and PPG-to-vital-sign reconstruction, achieving superior performance across multiple benchmarks.
FWBench introduces a benchmark for evaluating how language models select and use time-series forecasts to make cost-constrained decisions, comparing hosted and local configurations on electricity and cycle-hire datasets with efficient budget usage by GPT-6 Astra.
This paper introduces Kronecker coVariance Neural Networks (KVNNs), a temporal graph neural network that decouples spatial and temporal dependencies using Kronecker products to improve spatiotemporal data analysis, addressing limitations of traditional methods like ST-PCA.
Introduces leaky-integrator reconstruction to fix error accumulation in recursive differenced time-series forecasting, showing substantial error reduction across architectures and datasets.
The paper presents KG-Chronos-2, a method that enhances a frozen time-series foundation model with knowledge graph augmentation for improved surrogate forecasting of water-surface elevation in HEC-RAS, demonstrating significant RMSE reduction compared to baseline approaches.
This paper introduces a lightweight pre-encoder gate mechanism for Transformer-based time-series forecasting models to regulate covariate admission, with experiments showing competitive performance against baselines on multiple datasets.
The paper proposes a Physics Informed Recurrent Neural Network (PIRNN) that predicts unobservable physical variables to improve time series forecasting in physical processes, demonstrated through groundwater level predictions.
The paper benchmarks time-series foundation models for pedestrian crowd count forecasting across datasets, finding that foundation models excel in data-rich, seasonal regimes while simpler models can be competitive in limited data scenarios.
PPDL is a novel forecasting framework that integrates physical priors with deep learning to predict user retention ratios in multi-channel paid user acquisition, addressing challenges like channel heterogeneity and temporal patterns with validated improvements on industrial datasets.
The paper introduces FlowTSFM, an encoder-based time series foundation model that interprets depth as recurrent quantile transport, achieving competitive accuracy with fewer parameters compared to baselines like Chronos-2.
NanoForecast v0.5 is a 6.5M-parameter time series forecaster that achieves competitive performance against larger models like TimesFM by optimizing the training pipeline, resulting in a 43.8% reduction in MASE. The model and code are released under Apache 2.0.
Halo modifies deep forecasters to include scale estimation via heteroscedastic learning, significantly improving point forecast accuracy in benchmarks on electricity price markets.
The paper introduces CoSPOT, an LLM-based framework for online time series forecasting that uses compositional spectral prompts to efficiently adapt to non-stationary data and unseen patterns.
The paper compares foundation models with market-specific benchmarks for electricity price forecasting, finding that only TabPFN consistently outperforms statistically, but economic value varies by strategy and risk tolerance, suggesting foundation models cannot universally replace tailored models.
SAGE introduces a CLIP-based framework for time series forecasting that combines temporal, textual, and visual semantic information to improve accuracy, achieving state-of-the-art performance on long-term benchmarks.
This paper systematically investigates when auxiliary context helps in multi-modal time series forecasting, identifying two key conditions and demonstrating significant performance improvements when those conditions are met.
ICI-Time is a novel framework that reframes time series forecasting as a visual inpainting task, leveraging large vision models to enable adaptable forecasting without fine-tuning or architectural changes.
TimesFM 3.0 is a pretrained time-series foundation model by Google Research, released on Hugging Face with PyTorch weights for time-series forecasting tasks.