Active Learning for Quality Engineering

This repo contains the code for replicating the results from the paper "Active Learning for Industrial Applications" published in Quality Engineering.

The images used in this paper are from the MVTec datasets and they can be requested here:

Active Learning for Industrial Applications

This repository contains the code for replicating results from the paper Active Learning for Industrial Applications, published in Quality Engineering. The project explores active learning strategies to improve data efficiency in real-world industrial applications, reducing labeling costs while maintaining high model performance.

📌 Overview

Supervised learning models often require large labeled datasets, which can be costly and time-consuming to obtain. Active learning addresses this by selecting the most informative samples for annotation, improving model efficiency with fewer labeled examples. Here we show how to apply active learning methods to industrial quality control and fault detection tasks, demonstrating how active learning improves labeling efficiency in manufacturing settings.

🛠 Setup Instructions

1. Clone the repository:

   git clone https://github.com/yourusername/active-learning-industrial.git
   cd active-learning-industrial

2. Install dependencies:

   pip install -r requirements.txt

3. Run the example notebook:

   jupyter notebook active_learning.ipynb

📂 Code Structure

• requirements.txt: file that contains a list of all the packages installed in the environment used, along with their version numbers.
• feature_extraction.ipynb: notebook showing how to obtain preprocessed features from the images using a pre-trained ResNet-18.
• sampling_strategies.py: functions for performing random sampling and margin sampling.
• active_learning.ipynb: notebook showing how the implementation of the active learning strategy.

📊 Visualizing Active Learning

Below is an illustration of clustering-based sampling in action:

The black points indicate samples with the most representative data points of each clusters (subpopulation), which are selected for annotation in active learning.

Comparison of Active Learning Strategies

From the paper, we analyze different active learning strategies in an industrial setting:

📜 Citation

If you use this code, please cite:

Cacciarelli, D., & Kulahci, M. (2025). Active Learning for Industrial Applications. Quality Engineering.