اشترك بالحزمة الذهبية واحصل على وصول غير محدود شمرا أكاديميا
تسجيل مستخدم جديدNeural Attention Distillation: Erasing Backdoor Triggers from Deep Neural Networks
168
0
0.0
(
0
)
اسأل ChatGPT حول البحث
ﻻ يوجد ملخص باللغة العربية
Deep neural networks (DNNs) are known vulnerable to backdoor attacks, a training time attack that injects a trigger pattern into a small proportion of training data so as to control the models prediction at the test time. Backdoor attacks are notably dangerous since they do not affect the models performance on clean examples, yet can fool the model to make incorrect prediction whenever the trigger pattern appears during testing. In this paper, we propose a novel defense framework Neural Attention Distillation (NAD) to erase backdoor triggers from backdoored DNNs. NAD utilizes a teacher network to guide the finetuning of the backdoored student network on a small clean subset of data such that the intermediate-layer attention of the student network aligns with that of the teacher network. The teacher network can be obtained by an independent finetuning process on the same clean subset. We empirically show, against 6 state-of-the-art backdoor attacks, NAD can effectively erase the backdoor triggers using only 5% clean training data without causing obvious performance degradation on clean examples. Code is available in https://github.com/bboylyg/NAD.
قيم البحث
اقرأ أيضاً
Back-door attack poses a severe threat to deep learning systems. It injects hidden malicious behaviors to a model such that any input stamped with a special pattern can trigger such behaviors. Detecting back-door is hence of pressing need. Many exist
ing defense techniques use optimization to generate the smallest input pattern that forces the model to misclassify a set of benign inputs injected with the pattern to a target label. However, the complexity is quadratic to the number of class labels such that they can hardly handle models with many classes. Inspired by Multi-Arm Bandit in Reinforcement Learning, we propose a K-Arm optimization method for backdoor detection. By iteratively and stochastically selecting the most promising labels for optimization with the guidance of an objective function, we substantially reduce the complexity, allowing to handle models with many classes. Moreover, by iteratively refining the selection of labels to optimize, it substantially mitigates the uncertainty in choosing the right labels, improving detection accuracy. At the time of submission, the evaluation of our method on over 4000 models in the IARPA TrojAI competition from round 1 to the latest round 4 achieves top performance on the leaderboard. Our technique also supersedes three state-of-the-art techniques in terms of accuracy and the scanning time needed.
Machine unlearning has great significance in guaranteeing model security and protecting user privacy. Additionally, many legal provisions clearly stipulate that users have the right to demand model providers to delete their own data from training set
, that is, the right to be forgotten. The naive way of unlearning data is to retrain the model without it from scratch, which becomes extremely time and resource consuming at the modern scale of deep neural networks. Other unlearning approaches by refactoring model or training data struggle to gain a balance between overhead and model usability. In this paper, we propose an approach, dubbed as DeepObliviate, to implement machine unlearning efficiently, without modifying the normal training mode. Our approach improves the original training process by storing intermediate models on the hard disk. Given a data point to unlearn, we first quantify its temporal residual memory left in stored models. The influenced models will be retrained and we decide when to terminate the retraining based on the trend of residual memory on-the-fly. Last, we stitch an unlearned model by combining the retrained models and uninfluenced models. We extensively evaluate our approach on five datasets and deep learning models. Compared to the method of retraining from scratch, our approach can achieve 99.0%, 95.0%, 91.9%, 96.7%, 74.1% accuracy rates and 66.7$times$, 75.0$times$, 33.3$times$, 29.4$times$, 13.7$times$ speedups on the MNIST, SVHN, CIFAR-10, Purchase, and ImageNet datasets, respectively. Compared to the state-of-the-art unlearning approach, we improve 5.8% accuracy, 32.5$times$ prediction speedup, and reach a comparable retrain speedup under identical settings on average on these datasets. Additionally, DeepObliviate can also pass the backdoor-based unlearning verification.
Recent work has proposed the concept of backdoor attacks on deep neural networks (DNNs), where misbehaviors are hidden inside normal models, only to be triggered by very specific inputs. In practice, however, these attacks are difficult to perform an
d highly constrained by sharing of models through transfer learning. Adversaries have a small window during which they must compromise the student model before it is deployed. In this paper, we describe a significantly more powerful variant of the backdoor attack, latent backdoors, where hidden rules can be embedded in a single Teacher model, and automatically inherited by all Student models through the transfer learning process. We show that latent backdoors can be quite effective in a variety of application contexts, and validate its practicality through real-world attacks against traffic sign recognition, iris identification of lab volunteers, and facial recognition of public figures (politicians). Finally, we evaluate 4 potential defenses, and find that only one is effective in disrupting latent backdoors, but might incur a cost in classification accuracy as tradeoff.
Creating a state-of-the-art deep-learning system requires vast amounts of data, expertise, and hardware, yet research into embedding copyright protection for neural networks has been limited. One of the main methods for achieving such protection invo
lves relying on the susceptibility of neural networks to backdoor attacks, but the robustness of these tactics has been primarily evaluated against pruning, fine-tuning, and model inversion attacks. In this work, we propose a neural network laundering algorithm to remove black-box backdoor watermarks from neural networks even when the adversary has no prior knowledge of the structure of the watermark. We are able to effectively remove watermarks used for recent defense or copyright protection mechanisms while achieving test accuracies above 97% and 80% for both MNIST and CIFAR-10, respectively. For all backdoor watermarking methods addressed in this paper, we find that the robustness of the watermark is significantly weaker than the original claims. We also demonstrate the feasibility of our algorithm in more complex tasks as well as in more realistic scenarios where the adversary is able to carry out efficient laundering attacks using less than 1% of the original training set size, demonstrating that existing backdoor watermarks are not sufficient to reach their claims.
Recent researches show that deep learning model is susceptible to backdoor attacks. Many defenses against backdoor attacks have been proposed. However, existing defense works require high computational overhead or backdoor attack information such as
the trigger size, which is difficult to satisfy in realistic scenarios. In this paper, a novel backdoor detection method based on adversarial examples is proposed. The proposed method leverages intentional adversarial perturbations to detect whether an image contains a trigger, which can be applied in both the training stage and the inference stage (sanitize the training set in training stage and detect the backdoor instances in inference stage). Specifically, given an untrusted image, the adversarial perturbation is added to the image intentionally. If the prediction of the model on the perturbed image is consistent with that on the unperturbed image, the input image will be considered as a backdoor instance. Compared with most existing defense works, the proposed adversarial perturbation based method requires low computational resources and maintains the visual quality of the images. Experimental results show that, the backdoor detection rate of the proposed defense method is 99.63%, 99.76% and 99.91% on Fashion-MNIST, CIFAR-10 and GTSRB datasets, respectively. Besides, the proposed method maintains the visual quality of the image as the l2 norm of the added perturbation are as low as 2.8715, 3.0513 and 2.4362 on Fashion-MNIST, CIFAR-10 and GTSRB datasets, respectively. In addition, it is also demonstrated that the proposed method can achieve high defense performance against backdoor attacks under different attack settings (trigger transparency, trigger size and trigger pattern). Compared with the existing defense work (STRIP), the proposed method has better detection performance on all the three datasets, and is more efficient than STRIP.
الأسئلة المقترحة
سجل دخول لتتمكن من نشر تعليقات
التعليقات
جاري جلب التعليقات
سجل دخول لتتمكن من متابعة معايير البحث التي قمت باختيارها
