aiqbench analysis
QiYao-I introduces a novel manifold-based architecture designed specifically to handle irregularly sampled and asynchronous multivariate time series, addressing a significant gap in existing foundation models that assume regular intervals. By mapping real timestamps into a learnable feature space and incorporating frequency-aware variable interactions, the model demonstrates superior generalization in zero-shot and few-shot scenarios on real-world benchmarks. This research advances the theoretical and practical capabilities of machine learning for complex, non-uniform data streams common in industrial IoT and financial analytics.
QiYao-I: A Manifold Based Foundation Model for Irregular Multivariate Time Series Forecasting
arXiv:2610.06936v1 Announce Type: new Abstract: Irregular multivariate time series forecasting is a challenging yet important problem in real-world applications, where observations are often irregularly sampled and asynchronously recorded across variables. Existing time series foundation models are mostly built on regularly sampled sequences, making them difficult to generalize to irregular time intervals and asynchronous cross-variable dependencies. To address these challenges, we propose QiYao-I, a manifold based foundation model for irregular multivariate time series forecasting. Specifically, we introduce a novel sampling-conditioned temporal manifold attention mechanism that maps real timestamps into a learnable temporal manifold feature space and injects temporal manifold biases into attention layers, enabling the model to capture both irregular time intervals and local sampling structures. Further, we propose a dynamic variable interaction mechanism with frequency awareness. It selectively performs cross-variable message passing under asynchronous observations. Extensive experiments on real-world irregular multivariate forecasting benchmarks demonstrate that QiYao-I achieves superior performance compared with both time series foundation models and end-to-end irregular forecasting models, showing strong generalization ability in zero-shot and few-shot settings.
Abstract:Irregular multivariate time series forecasting is a challenging yet important problem in real-world applications, where observations are often irregularly sampled and asynchronously recorded across variables. Existing time series foundation models are mostly built on regularly sampled sequences, making them difficult to generalize to irregular time intervals and asynchronous cross-variable dependencies. To address these challenges, we propose QiYao-I, a manifold based foundation model for irregular multivariate time series forecasting. Specifically, we introduce a novel sampling-conditioned temporal manifold attention mechanism that maps real timestamps into a learnable temporal manifold feature space and injects temporal manifold biases into attention layers, enabling the model to capture both irregular time intervals and local sampling structures. Further, we propose a dynamic variable interaction mechanism with frequency awareness. It selectively performs cross-variable message passing under asynchronous observations. Extensive experiments on real-world irregular multivariate forecasting benchmarks demonstrate that QiYao-I achieves superior performance compared with both time series foundation models and end-to-end irregular forecasting models, showing strong generalization ability in zero-shot and few-shot settings.
29 pages, 5 figures, 20 tables. Preprint
Machine Learning (cs.LG)
arXiv:2610.06936 [cs.LG]
(or arXiv:2610.06936v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2610.06936 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 266 words
Related coverage
A Hybrid Attention Model Learning Unified Time-aware Patch Representation for Irregular Multivariate Time Series Forecasting
arXiv · Sep 22, 2026
SurF: A Generative Model for Multivariate Irregular Time Series Forecasting
arXiv · Sep 10, 2026
7 Steps to Mastering Time Series Analysis with Python
KDnuggets · Jun 4, 2026
The Machines that Make the Machines
NVIDIA · Oct 7, 2026