TimeRAF
Summary
TimeRAF is a retrieval-augmented forecasting model for zero-shot time-series foundation models. It retrieves relevant time series from a knowledge base and integrates them through Channel Prompting.
Official Artifacts
- Paper: https://arxiv.org/abs/2412.20810
- IEEE TKDE page: https://www.computer.org/csdl/journal/tk/2025/09/11031238/27uvv8498Uo
Role In The Wiki
TimeRAF is important for the retrieval/context branch of TSFMs and for benchmark hygiene around knowledge-base overlap.