Anomaly Heterogeneity Learning for Open-set Supervised Anomaly Detection

TL;DR

A novel approach, namely Anomaly Heterogeneity Learning (AHL), that simulates a diverse set of heterogeneous anomaly distributions and then utilizes them to learn a unified heterogeneous abnormality model in surrogate open-set environments is introduced.

Abstract

Open-set supervised anomaly detection (OSAD) - a recently emerging anomaly detection area - aims at utilizing a few samples of anomaly classes seen during training to detect unseen anomalies (i.e., samples from open-set anomaly classes), while effectively identifying the seen anomalies. Benefiting from the prior knowledge illustrated by the seen anomalies, current OSAD methods can often largely reduce false positive errors. However, these methods are trained in a closed-set setting and treat the anomaly examples as from a homogeneous distribution, rendering them less effective in generalizing to unseen anomalies that can be drawn from any distribution. This paper proposes to learn heterogeneous anomaly distributions using the limited anomaly examples to address this issue. To this end, we introduce a novel approach, namely Anomaly Heterogeneity Learning (AHL), that simulates a diverse set of heterogeneous anomaly distributions and then utilizes them to learn a unified heterogeneous abnormality model in surrogate open-set environments. Further, AHL is a generic framework that existing OSAD models can plug and play for enhancing their abnormality modeling. Extensive experiments on nine real-world anomaly detection datasets show that AHL can 1) substantially enhance different state-of-the-art OSAD models in detecting seen and unseen anomalies, and 2) effectively generalize to unseen anomalies in new domains. Code is available at https://github.com/mala-lab/AHL.

Figures (4)

• Figure 1: Current methods vs. our method $\texttt{AHL}$, where the anomaly samples of the same color indicates that they are treated as from one data distribution. Compared to existing methods that model a homogeneous anomaly distribution in a closed-set environment, $\texttt{AHL}$ simulates a diverse set of heterogeneous anomaly distributions (Sec. \ref{['subsec:hadg']}) and learns heterogeneous abnormality from them in a surrogate open environment (Sec. \ref{['subsec:cdl']}).
• Figure 2: Overview of our approach $\texttt{AHL}$. Its HADG component first generates $T$ heterogeneous anomaly distribution datasets from the training set, $\mathcal{T}=\{\mathcal{D}_i\}_{i=1}^T$, each of which includes a support set and open-set query set, i.e., $\mathcal{D}_i = \{\mathcal{D}^s_i, \mathcal{D}^q_i\}$. It then utilizes them to learn $T$ heterogeneous AD models $\{\phi_i\}_{i=1}^T$ in a simulated open-set environment and synthesizes these heterogeneous anomaly models into a unified AD model $g(\cdot)$ by a collaborative differential learning (CDL). Different $\phi_i$ learn anomaly distributions of various quality, so we also devise a model $\psi(\cdot)$ to assign an importance score to each $\phi_i$ to enhance the CDL component.
• Figure 3: Hyperparameter analysis of $\texttt{AHL}$ (DRA) based on the general setting using ten anomaly examples. Left: AUC results w.r.t. different number of clusters ($C$). Right: AUC results w.r.t. the length of sequences ($K$) as input to the sequential model $\psi$.
• Figure 4: Comparison of the AUC performance between DevNet and $\texttt{AHL}$(DRA) without the pseudo anomaly augmentation generation module. Here $wo.\ aug$ indicates the augmentation techniques are excluded during training.