Adapting Prediction Sets to Distribution Shifts Without Labels
Keywords: Conformal prediction, Set-valued classification
Abstract: Recently there has been a surge of interest to deploy confidence set predictions rather than point predictions. Unfortunately, the effectiveness of such prediction sets is frequently impaired by distribution shifts in practice, and the challenge is often compounded by the lack of ground truth labels at test time. In this paper, we present a method for improving the quality of outputted prediction sets using only unlabeled data from the test domain. This is achieved by two new methods called $\texttt{ECP}$ and $\texttt{E{\small A}CP}$, that sit on top of existing set-valued classification methods and adjust their intervals according to the base model's own uncertainty evaluation on the unlabeled test data. Through extensive experiments on a number of large-scale datasets and neural network architectures, we show that our methods provide consistent improvement over existing conformal prediction based baselines and nearly match the performance of fully supervised methods.
Primary Area: probabilistic methods (Bayesian methods, variational inference, sampling, UQ, etc.)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 8615
Loading