Research map / Time series / Foundation models for time series
Zeus: Towards Tuning-Free Foundation Model for Time Series Analysis
Yisong Fu, Zezhi Shao, Chengqing Yu, Yujie Li, Yongjun Xu, Xueqi Cheng, Fei Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use point-wise tokenization with U-shaped multi-scale hierarchy (Zeus) instead of patch-tokenized TSFMs | Zeus · Imputation, random and block masking | MSE: ETTh1 random: 0.382 (MOMENT) -> 0.079 (Zeus) | Improved |
| switch MOMENT from patch-level to point-level missing at inference | MOMENT · Imputation avg (ETTm1, ETTh2, Weather) | MSE: 0.143 -> 0.175 | Hurt |
| add predictive mask to MOTM (ablation) | Zeus · GIFT-Eval | MASE/CRPS | Improved |
| add multi-block mask to MOTM (ablation) | Zeus · Imputation | MSE/MAE | Improved |
| add single-block mask to MOTM (ablation) | Zeus · UCR anomaly detection and UEA classification | F1 and accuracy | Improved |
| use multi-scale U-shaped architecture instead of vanilla point-token Transformer | Zeus · Efficiency, length 4096 | self-attention FLOPs / speed / memory: 3.8x fewer attention FLOPs; 2.1x faster and 3.1x more memory efficient than Time-MoE base | Improved |
| Zeus zero-shot vs best zero-shot TSFM Timer | Zeus · Point forecasting (6 datasets) | MSE, MAE: MSE -40.3%, MAE -25.3% | Improved |
| Zeus zero-shot vs UniTS on anomaly detection | Zeus · UCR anomaly (42 datasets) | adjusted F1: 74.4 -> 90.0 | Improved |
| use 1-NN with Zeus features instead of MOMENT 1-NN | Zeus · UEA classification | accuracy: 60.5 -> 67.5 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 on most shared results.
- MOMENT — MOMENT: A Family of Open Time-series Foundation Models (ICML 2024) · better on 2 of 2 shared results
- Sundial — Sundial: A Family of Highly Capable Time Series Foundation Models (ICML 2025) · better on 1 of 1 shared result
- TiRex — TiRex: Zero-Shot Forecasting Across Long and Short Horizons with Enhanced In-Context Learning (NeurIPS 2025) · better on 1 of 1 shared result
- SVP-T — SVP-T: A Shape-Level Variable-Position Transformer for Multivariate Time Series Classification (AAAI 2023) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Kairos: Toward Adaptive and Parameter-Efficient Time Series Foundation Models (NeurIPS 2026 · similarity 0.88)
- Multi-Scale Finetuning for Encoder-based Time Series Foundation Models (NeurIPS 2025 · similarity 0.88)
- Less is More: Unlocking Specialization of Time Series Foundation Models via Structured Pruning (NeurIPS 2025 · similarity 0.87)
- Towards Neural Scaling Laws for Time Series Foundation Models (ICLR 2025 · similarity 0.86)
- MOMENT: A Family of Open Time-series Foundation Models (ICML 2024 · similarity 0.86)
- In-Context Fine-Tuning for Time-Series Foundation Models (ICML 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
