Meta Learning-Driven Iterative Refinement for Robust Anomaly Detection in Industrial Inspection

TL;DR

The paper tackles robust anomaly detection in industrial inspection under noisy training data by proposing a meta-learning–driven iterative refinement framework that leverages Model-Agnostic Meta-Learning (MAML) to initialize and rapidly adapt model parameters. The approach combines a refined training loop with a hybrid anomaly scoring mechanism and density estimation via normalizing flows, complemented by dynamic IQR-based thresholding to adapt to changing score distributions. Experiments on MVTec AD and KSDD2 show substantial improvements in AUROC under noisy conditions and the ability to prune near out-of-distribution samples even with clean data, highlighting practical benefits for industrial deployment. Overall, the method delivers a robust, adaptable anomaly detector with mechanisms to prune training data and sustain high accuracy in real-world, variable industrial environments.

Abstract

This study investigates the performance of robust anomaly detection models in industrial inspection, focusing particularly on their ability to handle noisy data. We propose to leverage the adaptation ability of meta learning approaches to identify and reject noisy training data to improve the learning process. In our model, we employ Model Agnostic Meta Learning (MAML) and an iterative refinement process through an Inter-Quartile Range rejection scheme to enhance their adaptability and robustness. This approach significantly improves the models capability to distinguish between normal and defective conditions. Our results of experiments conducted on well known MVTec and KSDD2 datasets demonstrate that the proposed method not only excels in environments with substantial noise but can also contribute in case of a clear training set, isolating those samples that are relatively out of distribution, thus offering significant improvements over traditional models.

Figures (5)

• Figure 1: Comparison of anomaly detection models under various training conditions: Clean, Corrupted, and Iteratively Refined. This sequence demonstrates the effectiveness of our metalearning-driven iterative refinement process in improving the robustness and accuracy of anomaly detection in industrial applications.
• Figure 2: A schematic representation of the Model-Agnostic Meta-Learning (MAML) process, illustrating the inner and outer loop optimizations and highlighting adaptive adjustments to the model parameters $\theta$.
• Figure 3: Comparison of loss landscapes before and after MAML adaptations, clearly showing how MAML smooths the optimization landscape, facilitating more effective learning and generalization.
• Figure 4: AUROC performance across various classes of the MVTec-AD dataset.
• Figure 5: Overall results on MVTec-AD and KSDD2 datsets.