Reproduce “ENCP: Episode-Normalized Conformal Prediction for Vision-and-Language Navigation” 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
Vicky Feliren, A. Taufiq Asyhari, Muhamad Risqi U. Saputra
ENCP calibrates action sets across variable-length navigation episodes using conformal prediction. It divides each step’s nonconformity score by 2-p_max, then calibrates the worst adjusted score from each episode, providing marginal route-wide coverage when episodes are exchangeable. On R2R validation-unseen with DUET and THR at alpha=0.10, ENCP reaches 0.965 coverage with sets averaging 6.7 actions, but offers no guarantee across building shift.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.