![[2509.25826] Kairos: Toward Adaptive and Parameter-Efficient Time Series Foundation Models](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2509.25826] Kairos: Toward Adaptive and Parameter-Efficient Time Series Foundation Models
Summary
The paper presents Kairos, a novel time series foundation model that enhances zero-shot generalization by decoupling temporal heterogeneity from model capacity, using a dynamic tokenization approach.
Why It Matters
Kairos addresses significant challenges in time series analysis by improving model adaptability and efficiency. This is crucial for applications requiring real-time data processing and decision-making, as it allows for better performance with fewer parameters, making it more accessible for various industries.
Key Takeaways
- Kairos introduces a dynamic patching tokenizer for better temporal abstraction.
- The model achieves superior zero-shot performance with fewer parameters.
- It utilizes a multi-granularity positional embedding for robust temporal modeling.
- Kairos is trained on a novel Predictability-Stratified Time-Series corpus.
- The approach enhances adaptability in time series foundation models.
Computer Science > Machine Learning arXiv:2509.25826 (cs) [Submitted on 30 Sep 2025 (v1), last revised 13 Feb 2026 (this version, v2)] Title:Kairos: Toward Adaptive and Parameter-Efficient Time Series Foundation Models Authors:Kun Feng, Shaocheng Lan, Yuchen Fang, Wenchao He, Lintao Ma, Xingyu Lu, Kan Ren View a PDF of the paper titled Kairos: Toward Adaptive and Parameter-Efficient Time Series Foundation Models, by Kun Feng and 6 other authors View PDF HTML (experimental) Abstract:Inherent temporal heterogeneity, such as varying sampling densities and periodic structures, has posed substantial challenges in zero-shot generalization for Time Series Foundation Models (TSFMs). Existing TSFMs predominantly rely on massive parameterization to absorb such heterogeneity, as their static tokenization and positional encoding schemes entangle diverse temporal patterns into a fixed representation space, encouraging memorization rather than adaptation. To address this limitation, we propose Kairos, a flexible and parameter-efficient TSFM that decouples temporal heterogeneity from model capacity through a novel tokenization perspective. Kairos introduces a dynamic patching tokenizer and a mixture-of-size encoding that adapt observational granularity to local information density, enabling fine-grained temporal abstraction without increasing model width or depth. In addition, we design a multi-granularity positional embedding based on dynamic rotary encodings, which conditions on instan...
