CICADA: Cross-Domain Interpretable Coding for Anomaly Detection and Adaptation in Multivariate Time Series
Tian Lan, Yifei Gao, Yimeng Lu, Chen Zhang
TL;DR
CICADA tackles unsupervised anomaly detection in multivariate time series under latent, cross-domain shifts by combining a mixture-of-experts with selective meta-learning, adaptive expansion, and hierarchical attention to achieve robust, interpretable cross-domain detection. The method defines an anomaly score as $\\text{anosc}(\\mathcal{X}'_t) = \\sum_{j=1}^m \\sum_{k=1}^{m_j} w_j(\\mathcal{X}'_t) \\delta_j^{k ightarrow i'} A_j(\\mathcal{X}'_t; \\Theta_j^{k ightarrow i'})$, with domain-specific adaptations and a threshold $C$ for decision, while training optimizes $\\mathcal{L} = \\mathcal{L}_{\\text{reconstruction}} + \\lambda_1 \\mathcal{L}_{\\text{extraction}}$. The framework dynamically expands to accommodate new domains, assigns explicit meta-domain labels per expert, and uses hierarchical attention to quantify expert contributions, enhancing interpretability. Empirical results on synthetic, real industrial, and public datasets show CICADA outperforms state-of-the-art baselines in cross-domain detection metrics and provides interpretable domain attributions, enabling practical deployment in nonstationary environments. This work advances cross-domain anomaly detection by jointly addressing latent domains, negative transfer, and interpretability in multivariate time series.
Abstract
Unsupervised Time series anomaly detection plays a crucial role in applications across industries. However, existing methods face significant challenges due to data distributional shifts across different domains, which are exacerbated by the non-stationarity of time series over time. Existing models fail to generalize under multiple heterogeneous source domains and emerging unseen new target domains. To fill the research gap, we introduce CICADA (Cross-domain Interpretable Coding for Anomaly Detection and Adaptation), with four key innovations: (1) a mixture of experts (MOE) framework that captures domain-agnostic anomaly features with high flexibility and interpretability; (2) a novel selective meta-learning mechanism to prevent negative transfer between dissimilar domains, (3) an adaptive expansion algorithm for emerging heterogeneous domain expansion, and (4) a hierarchical attention structure that quantifies expert contributions during fusion to enhance interpretability further.Extensive experiments on synthetic and real-world industrial datasets demonstrate that CICADA outperforms state-of-the-art methods in both cross-domain detection performance and interpretability.