AI Classifiers Robustness to Adversarial Attacks: Manipulating Output of Classifiers by Small Input Perturbations
Deep Learning has shown an advanced performance in the state of the art applications in several domains. However, its vulnerability to adversarial attacks can remarkably ruin any powerful classifier's performance through manipulating the output estimation with small perturbations, that provoked Machine Learning research community to go through this topic. Hence, the explanation of the reason of this vulnerability remains undecided. In this project, we present the explanation of the existence of adversarial examples, and we analyze the features' behavior to adversarial perturbations, its responsibility toward this phenomena, and its impact on the robustness of an AI Classifier through the presentation of some experimental results using handwritten digits recognition using the MNIST dataset. The project is motivated by two perspectives of the explanation of the existence of adversarial examples. Finally, we present an evaluation of the proposed robustness enhancement method based on the explanation, that proves the efficiency of the proposed solution that could be applied on bigger scale applications to show more concrete performance.
Code (1)
Similar Papers 제목 키워드 기반
Robust Detection of Adversarial Attacks by Modeling the Intrinsic Properties of Deep Neural Networks
It has been shown that deep neural network (DNN) based classifiers are vulnerable to human-imperceptive adversarial perturbations which can cause DNN classifiers to output wrong predictions with high confidence. We propo…
Adversarial Imaging Pipelines
Adversarial attacks play an essential role in understanding deep neural network predictions and improving their robustness. Existing attack methods aim to deceive convolutional neural network (CNN)-based classifiers by m…
Exploring the role of Input and Output Layers of a Deep Neural Network in Adversarial Defense
Deep neural networks are learning models having achieved state of the art performance in many fields like prediction, computer vision, language processing and so on. However, it has been shown that certain inputs exist w…
Adversarial DefenseGuardian of the Ensembles: Introducing Pairwise Adversarially Robust Loss for Resisting Adversarial Attacks in DNN Ensembles
Adversarial attacks rely on transferability, where an adversarial example (AE) crafted on a surrogate classifier tends to mislead a target classifier. Recent ensemble methods demonstrate that AEs are less likely to misle…
DiversityImage ClassificationTowards Evaluating the Robustness of Chinese BERT Classifiers
Recent advances in large-scale language representation models such as BERT have improved the state-of-the-art performances in many NLP tasks. Meanwhile, character-level Chinese NLP models, including BERT for Chinese, hav…