Reproduce “Conformalized Adaptive Forecasting of Heterogeneous Trajectories” 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
Yanfei Zhou, Lars Lindemann, Matteo Sesia
CAFHT generates simultaneous forecasting bands for heterogeneous trajectories with finite-sample coverage guarantees. It adaptively adjusts prediction band widths based on individual trajectory unpredictability, leading to more informative and reliable uncertainty estimates in applications like motion planning.
Key insights
- Combining adaptive online conformal prediction with traditional split-conformal inference allows for simultaneously achieving dynamic adaptability to individual trajectory unpredictability and strong finite-sample coverage guarantees.
- Utilizing multiplicative conformity scores, which scale band expansion proportionally to the ACI output width, effectively addresses residual heteroscedasticity and enhances the informativeness of prediction bands in heterogeneous data.
- A dedicated data-splitting approach for tuning adaptive parameters like gamma ensures that the calibration process maintains statistical validity and prevents inflated coverage due to data reuse.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.