Research map / Time series / Foundation models for time series
TS-ICL: A Flexible Time-Indexed Foundation Model for Time Series via In-Context Learning
Etienne Le Naour, Tahar Nabil, Adrien Petralia
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Introduces TS-ICL, a 27M-parameter probabilistic in-context-learning encoder-regressor Transformer that unifies zero-shot forecasting and imputation, with covariates, trained on real and synthetic time series plus a DAG-based causal covariate prior.
Problem
placeholder
Approach
placeholder
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use TS-ICL time-indexed in-context regression instead of TabICLv2-TS | imputation · fm-impute-bench univariate | NMAE and CRPS: 17% and 15% | Improved |
| use TS-ICL instead of TabPFNv2.5-TS with covariates | imputation · fm-impute-bench known covariates | NMAE and CRPS: 36% and 35% | Improved |
| add covariates as inputs | TS-ICL imputation · fm-impute-bench known covariates | NMAE and CRPS: 39% and 38% | Improved |
| add covariates as inputs | TS-ICL forecasting · fev-bench 30 covariate datasets | CRPS | Improved |
| use TS-ICL instead of Chronos-2 with 30% look-back missing | forecasting · fev-bench univariate | MASE: 2.16 -> 1.77 | Improved |
| use TS-ICL instead of Chronos-2 with 50% look-back missing | forecasting · fev-bench univariate | MASE: 2.44 -> 1.89 | Improved |
| use TS-ICL instead of Chronos-2 with 70% look-back missing | forecasting · fev-bench univariate | MASE: 2.54 -> 2.16 | Improved |
| use TS-ICL instead of Chronos-2 with 90% look-back missing | forecasting · fev-bench univariate | MASE: 3.97 -> 3.63 | Improved |
| use TS-ICL instead of Chronos-2 with complete look-back | forecasting · fev-bench univariate | MASE: 1.62 -> 1.70 | Hurt |
| increase look-back missingness from 0% to 30% (Chronos-2) | Chronos-2 · fev-bench univariate | MASE: 1.62 -> 2.16 | Hurt |
| use TS-ICL instead of Chronos-2 for inference speed | forecasting · fev-bench univariate | inference time (s/window): 3.53 x 10^-3 -> 1.54 x 10^-2 | Hurt |
| use TS-ICL instead of TabICLv2-TS for inference speed | imputation · fm-impute-bench univariate | inference time (s/window): 3.07 x 10^-1 -> 6.51 x 10^-3 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- TiRex — TiRex: Zero-Shot Forecasting Across Long and Short Horizons with Enhanced In-Context Learning (NeurIPS 2025) · better on 1 of 1 shared result
- BRITS — BRITS: Bidirectional Recurrent Imputation for Time Series (NeurIPS 2018) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Multi-Scale Finetuning for Encoder-based Time Series Foundation Models (NeurIPS 2025 · similarity 0.90)
- In-Context Fine-Tuning for Time-Series Foundation Models (ICML 2025 · similarity 0.88)
- TS-RAG: Retrieval-Augmented Generation based Time Series Foundation Models are Stronger Zero-Shot Forecaster (NeurIPS 2025 · similarity 0.87)
- Timer-XL: Long-Context Transformers for Unified Time Series Forecasting (ICLR 2025 · similarity 0.87)
- S4M: S4 for multivariate time series forecasting with Missing values (ICLR 2025 · similarity 0.87)
- A Closer Look at Transformers for Time Series Forecasting: Understanding Why They Work and Where They Struggle (ICML 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
