K² · Artificial intelligence
A forecasting model that never has to start over when new data arrives
The authors introduce TiRex-2, a recurrent xLSTM-based time series foundation model that generalizes the univariate TiRex to multivariate forecasting with both past and future covariates.
PP
MP
Patrick Podest, Marco Pichler, Elias Bürger et al.
10 authors · cs.LG
The 30-second scan
Most forecasting AIs have an expensive habit: every time a new data point shows up, they re-read the entire history from scratch to update their prediction.
- Existing Transformer-based time series foundation models capture cross-variate dependencies but incur quadratic complexity in context length and require full-history recomputation as new observations arrive.
- TiRex-2 uses a memory-centric recurrent design that operates at constant per-patch cost under streaming.
- The model combines a bidirectional time mixer with an asymmetric grouped-attention variate mixer, enabling integration of future-known covariates while preserving strict causality over target variables.
38.4M active parameters used in univariate modean additional 44.1M parameters activated for multivariate forecasting