Reproduce “Improving the statistical efficiency of cross-conformal prediction” 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
Matteo Gasparin, Aaditya Ramdas
This work introduces new variants of cross-conformal prediction that achieve narrower prediction sets by incorporating novel p-value combination rules, leveraging exchangeability and external randomization, while rigorously preserving the existing finite-sample marginal coverage guarantee of at least 1-2 alpha.
Key insights
- Cross-conformal p-values exhibit exchangeability when the data folds are of equal size, a property that can be exploited for more efficient combination.
- Leveraging recent theoretical advancements in combining dependent and exchangeable p-values allows for the construction of statistically tighter prediction sets.
- Introducing external randomization can also lead to narrower prediction sets by reducing the conservatism of p-value combination rules, even under arbitrary dependence assumptions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.