Conformal Novelty Detection and Applications to Spatial Statistics and Collaborative Learning

by Martin Voigt Vejling (Université D'Aalborg)

Tuesday Nov 18, 2025, 11:15 AM → 12:15 PM Europe/Paris

Salle E. Picard (1R2-129)

Conformal novelty detection is a subfield of conformal inference concerned with one-class classification.

The study of conformal novelty detection is a very active area of research with many new discoveries in recent years [1]-[3].

 

In the first part of this talk, I will review some of the properties of so-called conformal p-values, including results from the recent literature as well as our contributions.

This includes:

(i) the distribution of a sequence of conformal p-values and their order statistics;

(ii) the false discovery rate control guarantees when using the Benjamini-Hochberg procedure;

(iii) the possibilities of familywise error rate control;

(iv) the (asymptotic and non-asymptotic) distribution of some combination statistics on the conformal p-values.

 

In the second part of the talk, I will discuss two application domains, specifically, spatial statistics and collaborative learning.

In spatial statistics, the theoretical developments from conformal novelty detection allows for efficient multiple Monte Carlo tests in the case of replicate point patterns.

In collaborative learning, statistical procedures providing data quality guarantees in a data sharing system can be developed using conformal novelty detection.

 

The talk is based on the manuscripts [4],[5].

References:

[1] Stephen Bates, Emmanuel Candès, Lihua Lei, Yaniv Romano and Matteo Sesia, "Testing for outliers with conformal p-values", vol. 51, The Annals of Statistics, 2023.

[2] Ariane Marandon, Lihua Lei, David Mary and Etienne Roquain, "Adaptive novelty detection with false discovery rate guarantee", vol. 52, The Annals of Statistics, 2024.

[3] Ulysse Gazin, Gilles Blanchard and Étienne Roquain, "Transductive conformal inference with adaptive scores", Proc. Conf. Artificial Intelligence and Statistics, 2024.

[4] Christophe A. N. Biscio, Adrien Mazoyer and Martin V. Vejling, "Conformal novelty detection for replicate point patterns with FDR or FWER control", Spatial Statistics, 2025.

[5] Martin V. Vejling, Shashi Raj Pandey, Christophe A. N. Biscio and Petar Popovski, "Conformal Data Contamination Tests for In-Distribution Data Acquisition", preprint, 2025.