Ali, H., Khalid, F., Tariq, H. A., Hanif, M. A., Ahmed, R., & Rehman, S. (2020). SSCNets: Robustifying DNNs using Secure Selective Convolutional Filters. IEEE Design and Test, 37(2), 58–65. https://doi.org/10.1109/mdat.2019.2961325
E384-02 - Forschungsbereich Systems on Chip E191-02 - Forschungsbereich Embedded Computing Systems
-
Zeitschrift:
IEEE Design and Test
-
ISSN:
2168-2356
-
Datum (veröffentlicht):
2020
-
Umfang:
8
-
Peer Reviewed:
Ja
-
Keywords:
Electrical and Electronic Engineering; Software; Hardware and Architecture; Deep learning; Robustness; Filtering; Feature extraction; Perturbation methods; Image edge detection; Training data
-
Abstract:
Training data is crucial in ensuring robust neural inference, and deep neural networks (DNNs) are heavily dependent on this assumption. However, DNNs can be exploited by adversaries that facilitate various attacks. Adversarial defenses include several techniques, some of which happen during the preprocessing stages (i.e., noise filtering, etc.). This article analyzes the impact of some preprocessing filters, and proposes a selective preprocessing method which increases robustness and reduces the computational complexity.
en
Forschungsschwerpunkte:
Information Systems Engineering: 50% Computer Engineering and Software-Intensive Systems: 50%
