Robust Yet Efficient Conformal Prediction Sets

TL;DR

This paper introduces a method to create robust conformal prediction sets that maintain coverage guarantees even under adversarial attacks, improving efficiency and applicability across data types.

Contribution

It derives provably robust conformal prediction sets with tighter bounds, addressing both evasion and poisoning attacks for continuous and discrete data.

Findings

Robust sets maintain coverage under adversarial attacks.

Tighter bounds lead to more efficient prediction sets.

Applicable to both feature and label perturbations.

Abstract

Conformal prediction (CP) can convert any model's output into prediction sets guaranteed to include the true label with any user-specified probability. However, same as the model itself, CP is vulnerable to adversarial test examples (evasion) and perturbed calibration data (poisoning). We derive provably robust sets by bounding the worst-case change in conformity scores. Our tighter bounds lead to more efficient sets. We cover both continuous and discrete (sparse) data and our guarantees work both for evasion and poisoning attacks (on both features and labels).