Enhancing Anomaly Detection with Interpretability and Generalization: ExIFFI and EIF+
The authors propose ExIFFI, a novel interpretability approach for the Extended Isolation Forest (EIF) algorithm, and introduce EIF+, an enhanced variant of EIF designed to improve generalization capabilities. The work aims to address the need for interpretable and effective anomaly detection models.
Improving Anomaly Discovery through Active Learning with Tree-based Ensembles
Tree-based anomaly detection ensembles are naturally suited for active learning, and the greedy querying strategy of seeking labels for instances with the highest anomaly scores is an efficient approach. Novel batch and streaming active learning algorithms are developed to improve the diversity of discovered anomalies and handle data drift, respectively.
Differential Privacy for Anomaly Detection: Analyzing the Trade-off Between Privacy and Explainability
Applying differential privacy (DP) to anomaly detection (AD) models significantly impacts their performance and explainability, with the trade-off varying across datasets and AD algorithms.
Understanding the Role of Normalization in Contrastive Representation Learning for Improved Out-of-Distribution Detection
Contrastive learning inherently promotes a large norm for the contrastive features of in-distribution samples, creating a separation between in-distribution and out-of-distribution data in the feature space. This property can be leveraged to improve out-of-distribution detection by incorporating out-of-distribution samples into the contrastive learning process.