Mahalanobis feature-distance detection for anomalies and adversarial samples

What changed

Fits class means and a tied feature covariance, then combines layer-wise Mahalanobis scores with input preprocessing to detect anomalous or adversarial examples from pretrained representations.

Scientific scope

Fits class means and a tied feature covariance, then combines layer-wise Mahalanobis scores with input preprocessing to detect anomalous or adversarial examples from pretrained representations. Gaussian-discriminant interpretation is an assumption/model, not proof that arbitrary learned features are Gaussian. Layer ensemble weights and preprocessing use validation data; empirical detection is not certified resistance to adaptive attacks. Mahalanobis distance itself is inherited.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Gaussian-discriminant interpretation is an assumption/model, not proof that arbitrary learned features are Gaussian. Layer ensemble weights and preprocessing use validation data; empirical detection is not certified resistance to adaptive attacks. Mahalanobis distance itself is inherited.

Contributors

• Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.

• A Simple Unified Framework for Detecting Out-of-Distribution Samples and Adversarial Attacks

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction