Repozytorium Uniwersytetu Zielonogórskiego udostępnia 9 345 obiektów cyfrowych
Obiekt
Tytuł: Developing Hessian-free second-order adversarial examples for adversarial training
Creator:
Qian, Yaguan ; Zhang, Liangjian ; Wang, Yugi ; Ji, Boyuan ; Yao, Tengteng ; Wang, Bin
Date:
Resource Type:
Contributor:
Korbicz, Józef (1951- ) - red. ; Uciński, Dariusz - red.
Group publication title:
Abstract:
Recent studies show that deep neural networks (DNNs) are extremely vulnerable to elaborately designed adversarial examples. Adversarial training, which uses adversarial examples as training data, has been proven to be one of the most effective methods of defense against adversarial attacks. However, most existing adversarial training methods use adversarial examples relying on first-order gradients, which perform poorly against second-order adversarial attacks and make it difficult to further improve the robustness of the model. ; In contrast to first-order gradients, second-order gradients provide a more accurate approximation of the loss landscape relative to natural examples. Therefore, our work focuses on constructing second-order adversarial examples and utilizing them for training DNNs. However, second-order optimization involves computing the Hessian inverse, which typically consumes considerable time. To address this issue, we propose an approximation method that transforms the problem into optimization within the Krylov subspace. ; Compared with the Euclidean space, the Krylov subspace method typically does not require storing the entire matrix. It only needs to store vectors and intermediate results, avoiding explicitly calculating the complete Hessian matrix. We approximate the adversarial direction by a linear combination of Hessian-vector products in the Krylov subspace to reduce the computation cost. Because of the non-symmetrical Hessian matrix, we use the generalized minimum residual to search for an approximate polynomial solution of the matrix. ; Our method efficiently reduces computational complexity and accelerates the training process. Extensive experiments conducted on the MNIST, CIFAR-10, and ImageNet-100 datasets demonstrate that our adversarial learning using second-order adversarial samples outperforms other first-order methods, leading to improved model robustness against various attacks.
Publisher:
Zielona Góra: Uniwersytet Zielonogórski
Resource Identifier:
DOI:
Pages:
Source:
AMCS, volume 34, number 3 (2024) ; kliknij tutaj, żeby przejść
Language:
License CC BY 4.0:
Rights:
Biblioteka Uniwersytetu Zielonogórskiego
Kolekcje, do których przypisany jest obiekt:
- Repository > Faculties > Faculty of Computer, Electrical and Control Engineering
- Repository > Types of work > Articles
- Repository > Scientific journals and UZ publishing series > International Journal of Applied Mathematics and Computer Science (AMCS) > International Journal of Applied Mathematics and Computer Science (AMCS) (2024)
Data ostatniej modyfikacji:
5 sie 2025
Data dodania obiektu:
5 sie 2025
Liczba wyświetleń treści obiektu:
36
Wszystkie dostępne wersje tego obiektu:
https://zbc.uz.zgora.pl/repozytorium/publication/101873
Wyświetl opis w formacie RDF:
Wyświetl opis w formacie OAI-PMH:
-
-
-
-
International Journal of Applied Mathematics and Computer Science (AMCS), volume 34, number 1 (2024)
-
International Journal of Applied Mathematics and Computer Science (AMCS), volume 34, number 2 (2024)
-
International Journal of Applied Mathematics and Computer Science (AMCS), volume 34, number 3 (2024)
-
International Journal of Applied Mathematics and Computer Science (AMCS), volume 34, number 4 (2024)
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
| Nazwa wydania | Data |
|---|---|
| Developing Hessian-free second-order adversarial examples for adversarial training | 5 sie 2025 |
Obiekty Podobne
Zhou, Qian Sun, Bo Korbicz, Józef (1951- ) - red. Uciński, Dariusz - red.
Bernardo, Lucas Salvador Damaševičius, Robertas Albuquerque, Victor Hugo C. de Maskeliunas, Rytis Kusy, Maciej - ed. Scherer, Rafał - ed. Krzyżak, Adam - ed.
Yilmaz, Mustafa Berkay Öztürk, Kagan Korbicz, Józef (1951- ) - red. Uciński, Dariusz - red.
Koziarski, Michał Cyganek, Bogusław Korbicz, Józef (1951- ) - red. Uciński, Dariusz - red.
Liu, Zhongshan Yu, Bin Zhang, Li Wang, Wensi Yao, Baozhen - ed. Wang, Shuaian - ed. Asian, Sobhan - ed.
