Performance of Zero-Shot Time Series Foundation Models on Cloud Data

2025/02/18 by William Toner, T. Lee, Toner, William +7 · 4 voices · 5 citations
Computer Science · #FOS: Computer and information sciences #Machine Learning (cs.LG) #Time Series Analysis and Forecasting

paper · pdf · doi:10.48550/arxiv.2502.12944

openalex publication_date 2025/02/18 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Time series foundation models (FMs) have emerged as a popular paradigm for zero-shot multi-domain forecasting. FMs are trained on numerous diverse datasets and claim to be effective forecasters across multiple different time series domains, including cloud data. In this work we investigate this claim, exploring the effectiveness of FMs on cloud data. We demonstrate that many well-known FMs fail to generate meaningful or accurate zero-shot forecasts in this setting. We support this claim empirically, showing that FMs are outperformed consistently by simple linear baselines. We also illustrate a number of interesting pathologies, including instances where FMs suddenly output seemingly erratic, random-looking forecasts. Our results suggest a widespread failure of FMs to model cloud data.

Cited by

• TelecomTS: A Multi-Modal Observability Dataset for Time Series and Language Analysis
• Why Cannot Neural Networks Master Extrapolation? Insights from Physical Laws
• One-Embedding-Fits-All: Efficient Zero-Shot Time Series Forecasting by a Model Zoo
• BinConv: A Neural Architecture for Ordinal Encoding in Time-Series Forecasting
• This Time is Different: An Observability Perspective on Time Series Foundation Models

Discussions

• Performance of Zero-Shot Time Series Foundation Models on Cloud Data [hn, 4 points, 0 comments]
• Foundation models fail on the majority of real world (timeseries) datasets. Who knew?.. these guys -> arxiv.org/pdf/2502.12944 #BigData #TimeseriesAnalysis #DataAnalysis #DataEngineering [bsky, 2 points, 0 comments]
• Haciendo modelos de forecasting de ventas note que era muuucho mas sencillo copiar el PROMEDIO de los años anteriores y analizar las diferencias. Ahora salio un paper que confirma que estaba haciendo [bsky, 0 points, 0 comments]
• arxiv.org/abs/2502.1... [bsky, 0 points, 0 comments]

Related

• How Foundational are Foundation Models for Time Series Forecasting?
• Foundation Models and Fine-Tuning: Toward a New Generation of Models for Time Series Forecasting
• TSFM-Bench: A Comprehensive and Unified Benchmark of Foundation Models for Time Series Forecasting
• TimeFound: A Foundation Model for Time Series Forecasting
• Rethinking Evaluation in the Era of Time Series Foundation Models: (Un)known Information Leakage Challenges