Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Stream-based Active Learning for Anomalous Sound Detection in Machine Condition Monitoring

Tuan Vu Ho, Kota Dohi, Yohei Kawaguchi

TL;DR

The paper tackles anomalous sound detection under limited anomalous data by introducing a stream-based active learning framework that updates the reference embedding set instead of retraining the neural network, enabling rapid adaptation in continuous audio streams.The method employs a hybrid sampling strategy combining least-certainty and random queries, guided by a budget manager to control labeling costs.Experiments on the DCASE 2023 Task 2 dataset show that AL improves ASD performance at low labeling budgets and outperforms baselines in partial AUC, illustrating practical benefits for evolving industrial monitoring scenarios.Overall, the approach offers a cost-effective, adaptively improving ASD system that can better handle unseen anomalies with limited labeled data.

Abstract

This paper introduces an active learning (AL) framework for anomalous sound detection (ASD) in machine condition monitoring system. Typically, ASD models are trained solely on normal samples due to the scarcity of anomalous data, leading to decreased accuracy for unseen samples during inference. AL is a promising solution to solve this problem by enabling the model to learn new concepts more effectively with fewer labeled examples, thus reducing manual annotation efforts. However, its effectiveness in ASD remains unexplored. To minimize update costs and time, our proposed method focuses on updating the scoring backend of ASD system without retraining the neural network model. Experimental results on the DCASE 2023 Challenge Task 2 dataset confirm that our AL framework significantly improves ASD performance even with low labeling budgets. Moreover, our proposed sampling strategy outperforms other baselines in terms of the partial area under the receiver operating characteristic score.

Stream-based Active Learning for Anomalous Sound Detection in Machine Condition Monitoring

TL;DR

The paper tackles anomalous sound detection under limited anomalous data by introducing a stream-based active learning framework that updates the reference embedding set instead of retraining the neural network, enabling rapid adaptation in continuous audio streams.The method employs a hybrid sampling strategy combining least-certainty and random queries, guided by a budget manager to control labeling costs.Experiments on the DCASE 2023 Task 2 dataset show that AL improves ASD performance at low labeling budgets and outperforms baselines in partial AUC, illustrating practical benefits for evolving industrial monitoring scenarios.Overall, the approach offers a cost-effective, adaptively improving ASD system that can better handle unseen anomalies with limited labeled data.

Abstract

This paper introduces an active learning (AL) framework for anomalous sound detection (ASD) in machine condition monitoring system. Typically, ASD models are trained solely on normal samples due to the scarcity of anomalous data, leading to decreased accuracy for unseen samples during inference. AL is a promising solution to solve this problem by enabling the model to learn new concepts more effectively with fewer labeled examples, thus reducing manual annotation efforts. However, its effectiveness in ASD remains unexplored. To minimize update costs and time, our proposed method focuses on updating the scoring backend of ASD system without retraining the neural network model. Experimental results on the DCASE 2023 Challenge Task 2 dataset confirm that our AL framework significantly improves ASD performance even with low labeling budgets. Moreover, our proposed sampling strategy outperforms other baselines in terms of the partial area under the receiver operating characteristic score.
Paper Structure (11 sections, 2 equations, 4 figures, 1 table, 2 algorithms)

This paper contains 11 sections, 2 equations, 4 figures, 1 table, 2 algorithms.

Table of Contents

  1. Introduction
  2. Anomalous Sound Detection Systems
  3. Proposed method
  4. Active learning framework for ASD
  5. Sampling strategy
  6. Experiments
  7. Dataset
  8. Implementation details
  9. Evaluation metrics
  10. Results
  11. Conclusion

Figures (4)

  • Figure 1: Overview of classification-based ASD system.
  • Figure 2: Active learning framework for anomalous sound detection.
  • Figure 3: Average AUC scores across 10 runs for each AL method, with error bars representing the $95\%$ confidence interval. The labeling budget denotes the relative ratio of queried samples to the total test samples.
  • Figure 4: Average pAUC scores across 10 runs for each AL method, with error bars representing the $95\%$ confidence interval. The labeling budget denotes the relative ratio of queried samples to the total test samples.