Timer-M1 is built on a primitive-based data synthesis and pretraining pipeline.
Pretraining on synthetic primitive patterns enables robust zero-shot multivariate time series forecasting
Timer-M1 uses temporal and relational primitives to generate diverse training episodes and sets new benchmarks on three forecasting evaluations
Chinese Tech
Haoran Zhang · Haixuan Liu · Xingjian Su · Yong Liu · Zhi Chen · Yuxuan Wang · +2 more
Tsinghua University · ByteDance
Research Digest··3 min read
The authors introduce Timer-M1, a multivariate time series foundation model pretrained on a pipeline that synthesizes elementary temporal patterns (trend, seasonality, regime changes) and cross-variate relations.
Why this paper
From ByteDance and Tsinghua University
In one line
Timer-M1 is a pretrained multivariate forecasting model that learns temporal and relational primitives for zero-shot generalization.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§