Abstract: Highlights•Propose a bidirectional linear recurrent model to leverage forward and backward LRUs.•This method provably shows its universality in sequence approximation and stability.•Validate our method with sequential images and timeseries datasets and many baselines.•Introduce the first LRU based benchmark for timeseries forecasting.
External IDs:dblp:journals/inffus/LiuJHWLBS26
Loading