arXiv:2606.10798v1 Announce Type: new Abstract: Pretrained time series foundation models (TSFMs) have enabled zero-shot forecasting on unseen target series. However, existing TSFMs often incur high computational cost and provide limited support for diverse variable types, often failing to account for covariates that exogenously influence target variability. To address these challenges, we propose CITRAS-FM, a tiny 7M-parameter TSFM that supports univariate, multivariate, and covariate-informed zero-shot forecasting with real-time CPU inference. Built on a patch-based, decoder-only Transformer,
The proliferation of time series data across industries and the increasing demand for real-time, resource-efficient predictive models are driving innovation in foundation models.
This development suggests a pathway to more accessible and deployable AI forecasting solutions that can run on less powerful hardware, expanding the reach of predictive analytics.
The ability to perform covariate-informed, zero-shot forecasting with a tiny 7M-parameter model on CPU enables broader adoption of advanced time series prediction without extensive compute resources.
- · Edge AI providers
- · SMEs with limited compute budgets
- · Industries relying on time series analysis (e.g., finance, logistics, IoT)
- · Small to medium data science teams
- · Companies reliant on large, inaccessible TSFMs
- · Traditional forecasting consultancies
Wider adoption and democratization of advanced time series forecasting across various sectors.
Increased efficiency and automation in operational decision-making, as more systems gain predictive capabilities.
The development of more specialized and tiny foundation models for other data types, accelerating the trend towards efficient edge AI.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG