RMISC: A Large-scale Real-world Multivariate Corpus for Time Series Foundation Models

2026-07-07 · Source: Takara TLDR - Daily AI Papers · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, quick

Summary

The RMISC corpus has been established as a large-scale, high-quality, openly accessible, real-world multivariate time series archive. This corpus, comprising around 200 datasets and 142 billion time points across diverse domains, addresses a critical gap in Time Series Foundation Models (TSFMs) development. Current TSFMs are often pretrained on synthetic data, which may not capture complex real-world temporal dynamics. Experiments pretraining four advanced TSFMs on univariate, synthetic multivariate, and real-world multivariate data demonstrated that incorporating real-world multivariate data significantly improves zero-shot generalization performance for both univariate and multivariate TSFMs on standard in-distribution and out-of-distribution benchmarks. This research offers insights into the benefits of real-world data for developing more robust TSFMs.

Key takeaway

For Machine Learning Engineers developing Time Series Foundation Models, you should prioritize incorporating large-scale real-world multivariate datasets like RMISC for pretraining. Relying solely on synthetic data risks suboptimal zero-shot generalization performance. Real-world data, proven to improve benchmark results, yields more robust TSFMs for complex temporal dynamics and diverse applications.

Key insights

Training Time Series Foundation Models (TSFMs) with large-scale real-world multivariate data substantially enhances their zero-shot generalization capabilities.

Principles

• Real-world data improves TSFM generalization.
• Synthetic data lacks complex real-world dynamics.
• Multivariate data benefits both univariate and multivariate TSFMs.

Method

Establish RMISC, a large-scale real-world multivariate time series corpus. Pretrain four advanced TSFMs on univariate, synthetic multivariate, and real-world multivariate data, then evaluate zero-shot generalization on in-distribution and out-of-distribution benchmarks.

In practice

• Utilize RMISC for TSFM pretraining.
• Prioritize real-world over synthetic data.

Topics

• Time Series Foundation Models
• RMISC Corpus
• Multivariate Time Series
• Zero-shot Generalization
• Real-world Data Pretraining
• Machine Learning Datasets

Best for: Research Scientist, AI Scientist, Machine Learning Engineer

Related on AIssential

• TTM: Tiny Foundation Models for Multivariate Time-Series Forecasting — Tiny Time Mixers (TTMs) enable efficient, accurate time-series forecasting via compact pre-trained models and multi-level adaptation.
• Zero-shot Multivariate Time Series Forecasting Using Tabular Prior Fitted Networks — Multivariate time series forecasting can be reframed as scalar regression for zero-shot tabular foundation model application.
• Five Questions About Chronos-2, the Time Series Foundation Model — Time series foundation models like Chronos-2 enable zero-shot forecasting by learning recurring temporal shapes from diverse data.
• The digital quant: instant portfolio optimization with JointFM — JointFM is a zero-shot foundation model predicting full multivariate distributions for instant, domain-agnostic time-series forecasting.
• Interview with AAAI Fellow Yan Liu: machine learning for time series — Physics-informed foundation models enable accurate, adaptable time series analysis, especially in data-scarce scientific and real-world applications.
• Data Machina #261 — Generative AI and LLMs are being adapted to achieve state-of-the-art performance in diverse time-series tasks.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Takara TLDR - Daily AI Papers.