TS-ICL: A Flexible Time-Indexed Foundation Model for Time Series via In-Context Learning

2026-06-04

AI Summary

This work addresses the limitation of existing time series foundation models, which are predominantly confined to forecasting and struggle to unify diverse real-world tasks such as handling irregular observations, imputing missing values, and managing degraded sampling. The authors propose TS-ICL, the first framework that integrates in-context learning with causal data priors through a probabilistic Transformer-based encoder-regressor architecture, casting all tasks into timestamp-aligned regression problems. Trained on synthetically generated causal dependency structures, TS-ICL achieves state-of-the-art performance in missing value imputation and maintains leading accuracy on both univariate and covariate-aware forecasting benchmarks, particularly excelling under partial observation conditions within the look-back window.

Abstract

Foundation models mark a profound paradigm shift in time series modeling, with task-specific models being superseded by general-purpose zero-shot models. Yet, current approaches primarily focus on forecasting, while real-world time series are often irregularly and partially observed, requiring models that can jointly forecast, impute missing values, and handle degraded sampling conditions. To address these challenges, we introduce TS-ICL, a novel probabilistic In-Context Learning encoder--regressor Transformer that unifies forecasting and imputation. TS-ICL formulates time series tasks as timestamp-aligned regression and naturally incorporates covariates by training on synthetic dependency structures generated from a novel causal data prior. Empirically, TS-ICL achieves a new state-of-the-art in imputation, while remaining competitive with leading forecasting foundation models across both univariate and covariate-aware benchmarks. It shows particularly strong performance in forecasting with partially observed look-back windows.

Problem

• time series
• forecasting
• imputation
• irregularly observed
• foundation models

Innovation

• In-Context Learning
• Time Series Foundation Model
• Unified Forecasting and Imputation
• Causal Data Prior
• Timestamp-Aligned Regression