HCFM: Helmholtz Component Flow Matching for Time Series Anomaly Detection

Key points

• However, standard FM learns a monolithic vector field in which nominal transport and local density-changing behavior are entangled.1
• We propose Helmholtz Component Flow Matching (HCFM), an unsupervised TSAD framework that parameterizes the vector field prediction network with separate transport, compression, and residual heads.1
• This structured parameterization improves scalability and detection performance by separating volume-preserving nominal transport from density-changing compression, yielding cleaner anomaly signals from distinct deviations of nominal behavior.1
• Experiments on large-scale univariate and multivariate TSAD benchmarks show that HCFM consistently improves detection accuracy while maintaining efficient inference compared with 16 baselines.1

Sources (1)

• [1]HCFM: Helmholtz Component Flow Matching for Time Series Anomaly Detection
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 18, 06:58 AM

  Flow Matching (FM) is an efficient simulation-free approach for training flow-based generative models, making it a promising foundation for scalable time-series anomaly detection (TSAD).

  However, standard FM learns a monolithic vector field in which nominal transport and local density-changing behavior are entangled.

Extractive summary: sentences quoted from the sources.