Are Time-Indexed Foundation Models the Future of Time Series Imputation?

2025/11/08 by Naour, Etienne Le, Nabil, Tahar, Petralia, Adrien +1
Computer Science · Decision Sciences · Mathematics · #Data Quality and Management #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning in Healthcare #Statistical Methods and Bayesian Inference

paper · doi:10.48550/arxiv.2511.05980

openalex publication_date 2025/11/08 · openalex created_date 2025/11/12 · openalex updated_date 2026/07/28

Abstract

Foundation models for time series imputation remain largely unexplored. Recently, two such models, TabPFN-TS and MoTM, have emerged. These models share a common philosophy that places them within the family of time-indexed foundation models. This paper presents the first large-scale empirical study of these models for zero-shot imputation, which enables missing value recovery without retraining across a wide range of scenarios. We conduct extensive univariate experiments across 33 out-of-domain datasets (approximately 1.3M imputation windows) and evaluate their ability to integrate covariates at inference time to improve accuracy without fine-tuning. Our results demonstrate that time-indexed foundation models are a powerful and practical step toward achieving general-purpose, zero-shot imputation for real-world time series.

Citations

• TabPFN-2.5: Advancing the State of the Art in Tabular Foundation Models
• MoTM: Towards a Foundation Model for Time Series Imputation based on Continuous Modeling
• Zero-Shot Time Series Forecasting with Covariates via In-Context Learning
• TiRex: Zero-Shot Forecasting Across Long and Short Horizons with Enhanced In-Context Learning
• ImputeINR: Time Series Imputation via Implicit Neural Representations for Disease Diagnosis with Missing Data
• A synthetic dataset of French electric load curves with temperature conditioning
• TimeMixer++: A General Time Series Pattern Machine for Universal Predictive Analysis
• GIFT-Eval: A Benchmark For General Time Series Forecasting Model Evaluation
• NuwaTS: a Foundation Model Mending Every Incomplete Time Series
• TSLANet: Rethinking Transformers for Time Series Representation Learning
• Chronos: Learning the Language of Time Series
• MOMENT: A Family of Open Time-series Foundation Models
• Unified Training of Universal Time Series Forecasting Transformers
• A decoder-only foundation model for time-series forecasting
• ClimateLearn: Benchmarking Machine Learning for Weather and Climate Modeling
• BuildingsBench: A Large-Scale Dataset of 900K Buildings and Benchmark for Short-Term Load Forecasting
• Time Series Continuous Modeling for Imputation and Forecasting with Implicit Neural Representations
• TimesNet: Temporal 2D-Variation Modeling for General Time Series Analysis

Related

• MoTM: Towards a Foundation Model for Time Series Imputation based on Continuous Modeling
• NuwaTS: a Foundation Model Mending Every Incomplete Time Series
• TSI-Bench: Benchmarking Time Series Imputation
• NRTSI: Non-Recurrent Time Series Imputation
• Rethinking Evaluation in the Era of Time Series Foundation Models: (Un)known Information Leakage Challenges