A1149
Title: Full-conformal novelty detection: A powerful and non-random approach
Authors: Junu Lee - University of Pennsylvania (United States)
Ilia Popov - University of Pennsylvania (United States)
Zhimei Ren - University of Pennsylvania (United States) [presenting]
Abstract: A powerful and non-random methodology is introduced for novelty detection, offering distribution-free false discovery rate (FDR) control guarantees. Building on the full-conformal inference framework and the concept of e-values, full-conformal e-values are introduced to quantify evidence for novelty relative to a given reference dataset. These e-values are then utilized by carefully crafted multiple-testing procedures to identify a set of novel units out-of-sample with provable finite-sample FDR control. Furthermore, the method is extended to address distribution shifts, accommodating scenarios where novelty detection must be performed on data drawn from a shifted distribution relative to the reference dataset. In all settings, the method is non-random and can perform powerfully with limited amounts of reference data. Empirical evaluations on synthetic and real-world datasets demonstrate that the approach significantly outperforms existing methods for novelty detection.
