An incremental adversarial training method enables timeliness and rapid new knowledge acquisition.

Publisher: Nature Publishing Group Country of Publication: England NLM ID: 101563288 Publication Model: Electronic Cited Medium: Internet ISSN: 2045-2322 (Electronic) Linking ISSN: 20452322 NLM ISO Abbreviation: Sci Rep Subsets: MEDLINE

Original Publication: London : Nature Publishing Group, copyright 2011-

Neural Networks, Computer* ; Brain-Computer Interfaces* ; Deep Learning*; Humans ; Algorithms

Competing Interests: Declarations. Competing interests: The authors declare no competing interests.
Adversarial training is an effective defense method for deep models against adversarial attacks. However, current adversarial training methods require retraining the entire neural network, which consumes a significant amount of computational resources, thereby affecting the timeliness of deep models and further hindering the rapid learning process of new knowledge. In response to the above problems, this article proposes an incremental adversarial training method (IncAT) and applies it to the field of brain computer interfaces (BCI). Within this method, we first propose a deep model called Neural Hybrid Assembly Network (NHANet) and then train it. Then, based on the original samples and the trained deep model, calculate the Fisher information matrix to evaluate the importance of deep neural network parameters on the original samples. Finally, when calculating the loss of adversarial samples and real labels, an Elastic Weight Consolidation (EWC) loss is added to limit the variation of important weights and bias parameters in the Neural Hybrid Assembly Network (NHANet). The above incremental adversarial training method was applied to the publicly available epilepsy brain computer interface dataset at the University of Bonn. The experimental results showed that when facing three different attack algorithms, including fast gradient sign method (FGSM), projected gradient descent (PGD) and basic iterative method (BIM), the method proposed in this paper achieved robust accuracies of 95.33%, 94.67%, and 93.60%, respectively, without affecting the accuracy of clean samples, which is 5.06%, 4.67%, and 2.67% higher than traditional training methods respectively, thus fully verifying the generalization and effectiveness of the method.
(© 2025. The Author(s).)

Phys Rev E Stat Nonlin Soft Matter Phys. 2001 Dec;64(6 Pt 1):061907. (PMID: 11736210)
IEEE Trans Pattern Anal Mach Intell. 2021 May;43(5):1483-1498. (PMID: 31794388)
IEEE Trans Neural Syst Rehabil Eng. 2020 Dec;28(12):2615-2626. (PMID: 33175681)
IEEE Trans Pattern Anal Mach Intell. 2025 Feb;47(2):961-977. (PMID: 39405140)

JJKH20250945KJ Science and Technology Development Project of the Department of Education of Jilin Province; JJKH20250945KJ Science and Technology Development Project of the Department of Education of Jilin Province; JJKH20250945KJ Science and Technology Development Project of the Department of Education of Jilin Province; JJKH20250945KJ Science and Technology Development Project of the Department of Education of Jilin Province; JJKH20250945KJ Science and Technology Development Project of the Department of Education of Jilin Province; 2022IT096 New Generation Information Technology Innovation Project of China University Industry, University and Research Innovation Fund

Date Created: 20251014 Date Completed: 20251014 Latest Revision: 20251017

20251017

PMC12521521

10.1038/s41598-025-19840-8

41087533