기본 정보
연구 분야
프로젝트
논문
구성원
article|
인용수 0
·2024
Poster: Integrating SVM into SNNs for Enhanced Robustness Against Adversarial Attack
Daeho Kim, Jae-hyeon Park, Jiwoo Lee, Eun‐Kyu Lee
초록

Adversarial attacks have been attractive topics in the field of security of deep learning. Recent research suggested that spiking neural networks (SNNs) could be more robust to attacks than conventional deep neural networks (DNNs). To enhance the robustness, this study proposes a hybrid model, integrating SNNs with Support Vector Machine, that has been adopted to improve accuracy performance primarily. Experimental studies demonstrate that the hybrid model performs better than a vanilla SNN model as well as a conventional DNN model, showing the potential of the hybrid model in the security domain. This study also provides an experimental analysis of the robustness of SNNs by varying internal parameters under different attack intensities. Experimental results show that careful selection of parameters improves robustness performance by more than 5 times, especially under strong attacks.

키워드
Robustness (evolution)Computer scienceAdversarial systemSupport vector machineArtificial intelligenceMachine learning
타입
article
IF / 인용수
- / 0
게재 연도
2024