Paper measures sampling-frequency invariance of four time-series models
Original title: Quantifying the Sampling Frequency Invariance of Time-Series Foundation Models for Sensor Data Applications
MOMENT, Chronos-2, Moirai-MoE and TimesFM are scored with new FSI, FDI and CDI metrics.
Researchers stress-test four time-series foundation models used on sensor data whose sampling frequency changes across devices and modes: MOMENT, Chronos-2, Moirai-MoE and TimesFM. They measure embedding shift, forecasting degradation and downstream degradation with three metrics, FSI, FDI and CDI. The study asks whether behavior tracks loss of information bandwidth rather than the sampling-frequency value itself. Findings describe how invariance depends on the margin between a signal’s native rate and the bandwidth the task needs.
- The four models are MOMENT, Chronos-2, Moirai-MoE and TimesFM.
- Metrics are FSI for embeddings, FDI for forecasting and CDI for downstream tasks.
- The authors say this is the first study centered on sampling-frequency invariance.
- Invariance is framed as a property of each signal and model pair.
- AI models
- MOMENTChronos-2Moirai-MoETimesFM
From the original article
1 min read at arxiv.orgTime-Series Foundation Models (TSFMs) are increasingly used in sensor applications, where sampling frequency (SF) varies across sensors, devices, and operating modes (e. g. , power saving).
By Sourish Gunesh Dhekane · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.