Reproduce “Which Histories Matter for Time Series Forecasting? Learning Predictive Relevance with Future Supervision” 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
Yong-Hoon Choi, Kwang-Hyun Park, Youngjin Cho
The paper introduces PRR, a time-series retrieval method that uses realized future compatibility to train ranking while keeping inference decisions past-only. It combines Pearson and learned-representation candidates, then reranks them using observed-past features. Across six datasets and four horizons, PRR beats Pearson retrieval in 23 of 24 conditions, reducing AnalogFutureMSE by 26.68% on average; better retrieval does not guarantee better forecasts.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.