Related Experiment Video
Updated: Jan 15, 2026
06:37
Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention
Published on: December 15, 2023
5.3K
Parameter-Free Statistical Generator-Based Class Incremental Learning for Multi-User Physical Layer Authentication in
Wanbing Zhao1, Yanru Guo1, Yuchen Huang1
1College of Computer Science, Sichuan University, Chengdu 610065, China.
Sensors (Basel, Switzerland)
|October 16, 2025
Summary
This study introduces PSG-CIL, a new framework for updating deep learning models in Industrial Internet of Things physical layer authentication. It efficiently handles new users without high computational costs, improving accuracy and reducing overhead.
Area of Science:
- Cybersecurity
- Machine Learning
- Industrial Internet of Things (IIoT)
Background:
- Deep learning (DL) based physical layer authentication (PLA) in the Industrial Internet of Things (IIoT) faces challenges with frequent user updates.
- Existing Class Incremental Learning (CIL) methods, particularly generative replay, suffer from high computational overhead due to heavy parameterized models, limiting their use in resource-constrained IIoT environments.
Purpose of the Study:
- To propose a parameter-free statistical generator-based CIL framework (PSG-CIL) for DL-based multi-user PLA in the IIoT.
- To reduce computational overhead and improve the efficiency of CIL for dynamic user environments in IIoT.
Main Methods:
- Developed a parameter-free statistical generator (PSG) that uses Gaussian sampling on user-specific means and variances to create pseudo-data without training additional models.
- Implemented a confidence-based pseudo-data selection mechanism to ensure the reliability of generated data.
- Introduced a dynamic adjustment mechanism for the loss weight to balance knowledge retention and adaptation to new users.
Main Results:
- PSG-CIL achieved superior accuracy compared to retraining from scratch and other CIL methods on real industrial datasets.
- In the AAP outer loop scenario, PSG-CIL reached 70.68% accuracy, significantly outperforming retraining from scratch (58.57%).
- The proposed framework maintained a lightweight scale, demonstrating reduced computational overhead.
Conclusions:
- PSG-CIL offers an effective and computationally efficient solution for DL-based multi-user PLA in the IIoT, addressing the limitations of existing CIL approaches.
- The parameter-free nature and confidence-based selection make PSG-CIL suitable for resource-constrained IIoT deployments.
- This framework enables robust and adaptable authentication systems in dynamic industrial environments.
