Reproduce “TRACE: Time SeRies PArameter EffiCient FinE-tuning” 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
Yuze Li, Wei Zhu
TRACE adapts pretrained time-series forecasting models by compressing the prediction head and selectively updating LoRA modules using gated importance scores measured in masked contexts. On MOMENT-base, it reduces the 720-step head by 87.3%. On ETTh1, it reports MSE 0.470 and MAE 0.481 at horizon 720, outperforming AdaLoRA in that comparison; results favor TRACE on most, not all, listed long-horizon comparisons.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.