Reproduce “Amortized Predictability-aware Training Framework for Time Series Forecasting and Classification” 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
Xu Zhang, Peng Wang, Yichen Li, Wei Wang
APTF reduces the influence of hard time-series examples without discarding them. It groups examples by training loss, assigns lower weights to higher-loss groups, and updates these groups as training progresses; an optional second model helps estimate predictability. On TimeMixer’s Fund3, WMAPE is 83.659 with APTF, versus 85.14 originally and 94.893 with Co-teaching. These results cover reported comparisons, not universal performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.