Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

One-Class Domain Adaptation via Meta-Learning

Stephanie Holly, Thomas Bierweiler, Stefan von Dosky, Ahmed Frikha, Clemens Heitzinger, Jana Eder

TL;DR

This paper tackles the challenge of domain adaptation under one-class supervision in industrial IoT, where distribution shifts and scarce anomaly labels hinder deployment. It proposes OC-DA MAML, a meta-learning framework that uses a task-sampling strategy restricting the support set to the normal class while keeping the query set class-balanced to mimic target-domain evaluation. A theoretical analysis shows that the approach tunes meta-parameters to enable rapid one-class adaptation across domains by enhancing gradient interactions between support and query updates. Empirically, OC-DA MAML demonstrates superior target-domain performance over standard MAML on both Rainbow-MNIST and a real-world Centrifugal-Pumps dataset, approaching in-domain performance and enabling robust cross-domain anomaly detection in practical settings.

Abstract

The deployment of IoT (Internet of Things) sensor-based machine learning models in industrial systems for anomaly classification tasks poses significant challenges due to distribution shifts, as the training data acquired in controlled laboratory settings may significantly differ from real-time data in production environments. Furthermore, many real-world applications cannot provide a substantial number of labeled examples for each anomalous class in every new environment. It is therefore crucial to develop adaptable machine learning models that can be effectively transferred from one environment to another, enabling rapid adaptation using normal operational data. We extended this problem setting to an arbitrary classification task and formulated the one-class domain adaptation (OC-DA) problem setting. We took a meta-learning approach to tackle the challenge of OC-DA, and proposed a task sampling strategy to adapt any bi-level meta-learning algorithm to OC-DA. We modified the well-established model-agnostic meta-learning (MAML) algorithm and introduced the OC-DA MAML algorithm. We provided a theoretical analysis showing that OC-DA MAML optimizes for meta-parameters that enable rapid one-class adaptation across domains. The OC-DA MAML algorithm is evaluated on the Rainbow-MNIST meta-learning benchmark and on a real-world dataset of vibration-based sensor readings. The results show that OC-DA MAML significantly improves the performance on the target domains and outperforms MAML using the standard task sampling strategy.

One-Class Domain Adaptation via Meta-Learning

TL;DR

This paper tackles the challenge of domain adaptation under one-class supervision in industrial IoT, where distribution shifts and scarce anomaly labels hinder deployment. It proposes OC-DA MAML, a meta-learning framework that uses a task-sampling strategy restricting the support set to the normal class while keeping the query set class-balanced to mimic target-domain evaluation. A theoretical analysis shows that the approach tunes meta-parameters to enable rapid one-class adaptation across domains by enhancing gradient interactions between support and query updates. Empirically, OC-DA MAML demonstrates superior target-domain performance over standard MAML on both Rainbow-MNIST and a real-world Centrifugal-Pumps dataset, approaching in-domain performance and enabling robust cross-domain anomaly detection in practical settings.

Abstract

The deployment of IoT (Internet of Things) sensor-based machine learning models in industrial systems for anomaly classification tasks poses significant challenges due to distribution shifts, as the training data acquired in controlled laboratory settings may significantly differ from real-time data in production environments. Furthermore, many real-world applications cannot provide a substantial number of labeled examples for each anomalous class in every new environment. It is therefore crucial to develop adaptable machine learning models that can be effectively transferred from one environment to another, enabling rapid adaptation using normal operational data. We extended this problem setting to an arbitrary classification task and formulated the one-class domain adaptation (OC-DA) problem setting. We took a meta-learning approach to tackle the challenge of OC-DA, and proposed a task sampling strategy to adapt any bi-level meta-learning algorithm to OC-DA. We modified the well-established model-agnostic meta-learning (MAML) algorithm and introduced the OC-DA MAML algorithm. We provided a theoretical analysis showing that OC-DA MAML optimizes for meta-parameters that enable rapid one-class adaptation across domains. The OC-DA MAML algorithm is evaluated on the Rainbow-MNIST meta-learning benchmark and on a real-world dataset of vibration-based sensor readings. The results show that OC-DA MAML significantly improves the performance on the target domains and outperforms MAML using the standard task sampling strategy.
Paper Structure (15 sections, 7 equations, 3 figures, 5 tables, 1 algorithm)

This paper contains 15 sections, 7 equations, 3 figures, 5 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Methodology
  3. Related Work
  4. Meta-Learning
  5. One-Class Domain Adaptation via Meta-Learning
  6. One-Class Domain Adaptation via Model-Agnostic Meta-Learning (OC-DA MAML)
  7. Analysis of OC-DA MAML
  8. Experiments
  9. Experimental Setup
  10. Results
  11. Conclusion and Future Work
  12. Appendix
  13. Data
  14. Task Setup
  15. Data Splits

Figures (3)

  • Figure 1: Example of a 3-way 1-shot learning task in the OC-DA setting.
  • Figure 2: Visualization of distribution shifts in the Centrifugal-Pumps dataset, showing the average amplitude [mm] of FFT vibration signals per frequency [Hz] for each class (1-normal, 5-cavitation, 6-hydraulic blockage, 7-dry running), recorded by the same pump operated within a steel framework vs. on a concrete surface.
  • Figure 3: Visualization of domain-specific information in the Centrifugal-Pumps dataset, showing the average amplitude [mm] of the FFT vibration signals per frequency [Hz] with a $95\%$ confidence interval. Each anomalous class (5-cavitation, 6-hydraulic blockage, 7-dry running) is compared to the normal class (1-normal class), recorded by same pump operated within a steel framework vs. on a concrete surface.