An Effective Fusion Method to Enhance the Robustness of CNN
This work addresses robustness in CNNs for image classification, but it appears incremental as it focuses on improving module incorporation rather than introducing a new paradigm.
The paper tackles the problem of CNN vulnerability to adversarial perturbations in image classification by proposing a new fusion method that incorporates a denoising module and attention mechanism into ResNet18, achieving effectiveness and outperforming state-of-the-art methods on CIFAR10 under FGSM and PGD attacks.
With the development of technology rapidly, applications of convolutional neural networks have improved the convenience of our life. However, in image classification field, it has been found that when some perturbations are added to images, the CNN would misclassify it. Thus various defense methods have been proposed. The previous approach only considered how to incorporate modules in the network to improve robustness, but did not focus on the way the modules were incorporated. In this paper, we design a new fusion method to enhance the robustness of CNN. We use a dot product-based approach to add the denoising module to ResNet18 and the attention mechanism to further improve the robustness of the model. The experimental results on CIFAR10 have shown that our method is effective and better than the state-of-the-art methods under the attack of FGSM and PGD.