Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Training-Free Time-Series Anomaly Detection: Leveraging Image Foundation Models

Nobuo Namura, Yuma Ichikawa

TL;DR

This work tackles the practical challenge of time-series anomaly detection without heavy neural-network training or hyperparameter tuning. It introduces ITF-TAD, a training-free pipeline that converts time-series data into scalograms via dual-mother-wavelet CWT, aggregates them into compact RGB-like images, and applies a pretrained image foundation model (PatchCore) for anomaly detection, providing frequency-aware localization. The approach demonstrates competitive or superior performance on five benchmark datasets (univariate and multivariate) compared to several deep-learning baselines, while also offering detailed visualizations across frequencies. By eliminating training requirements and enabling intuitive frequency-based explanations, ITF-TAD offers a scalable and practical solution for real-world TAD tasks with robust interpretability across multiple domains.

Abstract

Recent advancements in time-series anomaly detection have relied on deep learning models to handle the diverse behaviors of time-series data. However, these models often suffer from unstable training and require extensive hyperparameter tuning, leading to practical limitations. Although foundation models present a potential solution, their use in time series is limited. To overcome these issues, we propose an innovative image-based, training-free time-series anomaly detection (ITF-TAD) approach. ITF-TAD converts time-series data into images using wavelet transform and compresses them into a single representation, leveraging image foundation models for anomaly detection. This approach achieves high-performance anomaly detection without unstable neural network training or hyperparameter tuning. Furthermore, ITF-TAD identifies anomalies across different frequencies, providing users with a detailed visualization of anomalies and their corresponding frequencies. Comprehensive experiments on five benchmark datasets, including univariate and multivariate time series, demonstrate that ITF-TAD offers a practical and effective solution with performance exceeding or comparable to that of deep models.

Training-Free Time-Series Anomaly Detection: Leveraging Image Foundation Models

TL;DR

This work tackles the practical challenge of time-series anomaly detection without heavy neural-network training or hyperparameter tuning. It introduces ITF-TAD, a training-free pipeline that converts time-series data into scalograms via dual-mother-wavelet CWT, aggregates them into compact RGB-like images, and applies a pretrained image foundation model (PatchCore) for anomaly detection, providing frequency-aware localization. The approach demonstrates competitive or superior performance on five benchmark datasets (univariate and multivariate) compared to several deep-learning baselines, while also offering detailed visualizations across frequencies. By eliminating training requirements and enabling intuitive frequency-based explanations, ITF-TAD offers a scalable and practical solution for real-world TAD tasks with robust interpretability across multiple domains.

Abstract

Recent advancements in time-series anomaly detection have relied on deep learning models to handle the diverse behaviors of time-series data. However, these models often suffer from unstable training and require extensive hyperparameter tuning, leading to practical limitations. Although foundation models present a potential solution, their use in time series is limited. To overcome these issues, we propose an innovative image-based, training-free time-series anomaly detection (ITF-TAD) approach. ITF-TAD converts time-series data into images using wavelet transform and compresses them into a single representation, leveraging image foundation models for anomaly detection. This approach achieves high-performance anomaly detection without unstable neural network training or hyperparameter tuning. Furthermore, ITF-TAD identifies anomalies across different frequencies, providing users with a detailed visualization of anomalies and their corresponding frequencies. Comprehensive experiments on five benchmark datasets, including univariate and multivariate time series, demonstrate that ITF-TAD offers a practical and effective solution with performance exceeding or comparable to that of deep models.
Paper Structure (36 sections, 5 equations, 10 figures, 8 tables)

This paper contains 36 sections, 5 equations, 10 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Time-Series Anomaly Detection
  4. Image-Based Approaches
  5. Method
  6. Scalogram Generation with Dual Mother Wavelets
  7. Scalogram Aggregation
  8. Component-Wise PCA Mapping
  9. Random Mapping
  10. Channel Embedding
  11. Anomaly Detection with Image Foundation Model
  12. Image Dividing
  13. Anomaly Score Computation
  14. Experiments
  15. Experimental Setup
  16. ...and 21 more sections

Figures (10)

  • Figure 1: Data processing steps in ITF-TAD
  • Figure 2: Score partitioning on UCR-053
  • Figure 3: Anomaly scores from five models for the test section of UCR-060
  • Figure 4: Anomaly scores from five models for the test section of SMD-2-3
  • Figure 5: Random matrix generation using Latin hyper cube sampling
  • ...and 5 more figures