Reproduce “Unified Taxonomy for Multivariate Time Series Anomaly Detection using Deep Learning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Bruna Alves, Armando J. Pinho, Sónia Gouveia
Researchers from the University of Aveiro developed a unified taxonomy with eleven dimensions to systematize Deep Learning-based Multivariate Time Series Anomaly Detection (MTSAD) methods. The framework classifies 409 papers published between 2019 and 2024, revealing key methodological trends and interdependencies, and was validated with 190 recent publications from 2025.
Key insights
- Choices across the taxonomy's dimensions are interdependent, with a common profile involving window-format input, reconstruction error, regression task, Sum of Errors loss, and nested spatio-temporal dependency.
- The field shows increasing adoption of multi-objective, contrastive, and adversarial loss functions, alongside a growing emphasis on explicit spatio-temporal feature extraction modules.
- The taxonomy remains stable and adaptable to emerging paradigms in MTSAD, including generative models, contrastive learning strategies, and early applications of Large Language Models.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.