UtoMe: Observation-Uncertainty-Guided Token Merging for Weather Foundation Models
We introduce UtoMe, an observation-uncertainty-guided token merging framework that combines dynamics saliency with ensemble data assimilation spread as a relative input-reliability proxy.
ProofPaper ↗
Key points
- Weather transformers process dense reanalysis fields whose reliability varies across variables, locations, and weather regimes.
- Token merging can reduce computation, but feature similarity alone does not account for this heterogeneous input uncertainty.
- In ERA5 72-hour forecasting at two spatial resolutions, UtoMe improves Overall error over train-time ToMe at matched token budgets.
- Component ablations and spatial uncertainty controls support combining dynamics with aligned uncertainty when allocating a weather transformer's token budget.
Sources (1)
- [1]UtoMe: Observation-Uncertainty-Guided Token Merging for Weather Foundation ModelsarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 18, 04:12 AM
We introduce UtoMe, an observation-uncertainty-guided token merging framework that combines dynamics saliency with ensemble data assimilation spread as a relative input-reliability proxy.
Weather transformers process dense reanalysis fields whose reliability varies across variables, locations, and weather regimes.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 17, 2026model : add Kolibri-1 (Aleph-Alpha) support by riteme · Pull Request #30673 · ggml-org/llama.cpp
- Oct 17, 2026TideKV: Delayed KV Eviction via Layer-Wise Chunked Prefill for Memory-Efficient Long-Context LLM Inference
- Oct 16, 2026Hugging Face ❤️ TorchTPU: TPUs, Natively
- Oct 16, 2026Beyond Multi-Stage Pipelines: A Label-Correlation Ensemble for Multi-Label Retinal Fundus Disease Classification
- Oct 16, 2026MambaAct: Efficient Mamba Action Experts for Flow-Matching Vision-Language-Action Models
- Oct 15, 2026Granite Switch 4.1: Chat, Decisions, and Speech from One Endpoint