arXiv AI

Differentiable Interval Bottlenecks for Interpretable Anomaly Detection in Numerical Data

DIFFINT is a reconstruction‑based anomaly detector that uses a differentiable autoencoder with a latent bottleneck composed of soft, axis‑aligned interval memberships. Each latent unit represents a human‑readable hyper‑rectangle in feature space, allowing the model to encode how strongly an instance falls inside each interval and to compute reconstruction error as the anomaly score. The method provides a certified lower bound on reconstruction error for points outside all active intervals, a suppression mechanism for sparse abnormalities, and a closed‑form, label‑free importance ranking for each (unit, feature) pair, achieving top performance on 48 ADBench benchmarks against 22 baselines.

Hugging Face Trending Papers
Sep 3

Differentiable Interval Bottlenecks for Interpretable Anomaly Detection in Numerical Data

DIFFINT is a reconstruction‑based anomaly detector that replaces the opaque latent bottleneck of a standard autoencoder with a set of soft, axis‑aligned interval memberships learned directly from raw numerical data. Each latent unit represents a human‑readable hyper‑rectangle, and an instance’s anomaly score is its reconstruction error weighted by how strongly it falls inside these intervals. The method provides a certified lower bound on reconstruction error for points outside all active intervals, a graded suppression mechanism for sparse anomalies, and a closed‑form, label‑free importance ranking for each (unit, feature) pair, achieving top performance on 48 ADBench benchmarks against 22 baselines. whyItMatters":"DIFFINT offers the first interpretable anomaly detector that maintains competitive performance while revealing which feature ranges drive each anomaly score, enabling practitioners to audit and understand model decisions without requiring anomaly labels."

arXiv AI
Sep 4

Witnesses Explain Anomalies

WAND is an unsupervised tabular anomaly detector that scores each point by how far its projection on unit‑sphere directions deviates from a sub‑Gaussian baseline. The directions that flag a point serve as its explanation, providing per‑feature attribution at no extra cost and recoverable via gradients. On 47 ADBench datasets, WAND matches or exceeds 16 baselines in ROC‑AUC while delivering more accurate, faithful explanations than post‑hoc SHAP, LIME, or ECOD, all with linear scoring time and a probe‑efficiency guarantee.

By Lamine Diop
arXiv AI
5d ago

Local Sparsity Enables Unsupervised LLM Safety Detection

The paper proposes a new unsupervised safety detection method for large language models that relies on anomaly detection rather than supervised training on unsafe data. By leveraging local sparsity in a linear representation space obtained via a sparse autoencoder, the authors develop a framework for locally masked SAE-based anomaly detection, providing theoretical support and empirical validation across multiple architectures and datasets. When calibrated with only 1% out-of-distribution data, the method achieves near‑optimal performance while using just 1–2% of SAE neurons for computation.

By Xin Chen, Gil Kur, Alexander Shevchenko, Andreas Krause
Hugging Face Trending Papers
6d ago

Local Sparsity Enables Unsupervised LLM Safety Detection

The paper proposes a novel unsupervised safety detection method for large language models that relies on local sparsity in a linear representation space recovered via a sparse autoencoder. By masking SAE neurons based on shared active support among nearby points, the authors develop a locally masked anomaly detection framework with theoretical backing. Experiments across multiple architectures and datasets—including capability‑testing and safety‑specific sets—show that using only 1–2% of SAE neurons and a small amount of out‑of‑distribution data yields near‑optimal safety detection performance.