Reproduce “Raw-Routed Mixture of Adapters: A Causal Intervention for Routing Collapse in Time Series Foundation Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Hung Phan, Thuy T. Nguyen, Minh Ngoc Dinh, Nhat-Quang Tran
The method adapts time-series foundation models with five expert heads and a gate that routes from raw windows. The experts use pretrained-backbone representations, while a lightweight convolutional gate sees unnormalized windows, preserving mean and scale cues removed by instance normalization. Across six forecasting datasets, it beat the best fixed adapter in all 54 comparisons, but did not help on low-signal Traffic.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.