We propose a minimax formulation for removing backdoors from a given poisoned model based on a small set of clean data. This formulation encompasses much of prior work on backdoor removal. We propose the Implicit Bacdoor Adversarial Unlearning (I-BAU) algorithm to solve the minimax. Unlike previous work, which breaks down the minimax into separate inner and outer problems, our algorithm utilizes the implicit hypergradient to account for the interdependence between inner and outer optimization. We theoretically analyze its convergence and the generalizability of the robustness gained by solving minimax on clean data to unseen test data. In our evaluation, we compare I-BAU with six state-of-art backdoor defenses on seven backdoor attacks over two datasets and various attack settings, including the common setting where the attacker targets one class as well as important but underexplored settings where multiple classes are targeted. I-BAU's performance is comparable to and most often significantly better than the best baseline. Particularly, its performance is more robust to the variation on triggers, attack settings, poison ratio, and clean data size. Moreover, I-BAU requires less computation to take effect; particularly, it is more than $13\times$ faster than the most efficient baseline in the single-target attack setting. Furthermore, it can remain effective in the extreme case where the defender can only access 100 clean samples -- a setting where all the baselines fail to produce acceptable results.
相關內容
Adversarial regularization has been shown to improve the generalization performance of deep learning models in various natural language processing tasks. Existing works usually formulate the method as a zero-sum game, which is solved by alternating gradient descent/ascent algorithms. Such a formulation treats the adversarial and the defending players equally, which is undesirable because only the defending player contributes to the generalization performance. To address this issue, we propose Stackelberg Adversarial Regularization (SALT), which formulates adversarial regularization as a Stackelberg game. This formulation induces a competition between a leader and a follower, where the follower generates perturbations, and the leader trains the model subject to the perturbations. Different from conventional approaches, in SALT, the leader is in an advantageous position. When the leader moves, it recognizes the strategy of the follower and takes the anticipated follower's outcomes into consideration. Such a leader's advantage enables us to improve the model fitting to the unperturbed data. The leader's strategic information is captured by the Stackelberg gradient, which is obtained using an unrolling algorithm. Our experimental results on a set of machine translation and natural language understanding tasks show that SALT outperforms existing adversarial regularization baselines across all tasks. Our code is available at //github.com/SimiaoZuo/Stackelberg-Adv.
The neural network (NN) becomes one of the most heated type of models in various signal processing applications. However, NNs are extremely vulnerable to adversarial examples (AEs). To defend AEs, adversarial training (AT) is believed to be the most effective method while due to the intensive computation, AT is limited to be applied in most applications. In this paper, to resolve the problem, we design a generic and efficient AT improvement scheme, namely case-aware adversarial training (CAT). Specifically, the intuition stems from the fact that a very limited part of informative samples can contribute to most of model performance. Alternatively, if only the most informative AEs are used in AT, we can lower the computation complexity of AT significantly as maintaining the defense effect. To achieve this, CAT achieves two breakthroughs. First, a method to estimate the information degree of adversarial examples is proposed for AE filtering. Second, to further enrich the information that the NN can obtain from AEs, CAT involves a weight estimation and class-level balancing based sampling strategy to increase the diversity of AT at each iteration. Extensive experiments show that CAT is faster than vanilla AT by up to 3x while achieving competitive defense effect.
Deep neural networks have become an integral part of our software infrastructure and are being deployed in many widely-used and safety-critical applications. However, their integration into many systems also brings with it the vulnerability to test time attacks in the form of Universal Adversarial Perturbations (UAPs). UAPs are a class of perturbations that when applied to any input causes model misclassification. Although there is an ongoing effort to defend models against these adversarial attacks, it is often difficult to reconcile the trade-offs in model accuracy and robustness to adversarial attacks. Jacobian regularization has been shown to improve the robustness of models against UAPs, whilst model ensembles have been widely adopted to improve both predictive performance and model robustness. In this work, we propose a novel approach, Jacobian Ensembles-a combination of Jacobian regularization and model ensembles to significantly increase the robustness against UAPs whilst maintaining or improving model accuracy. Our results show that Jacobian Ensembles achieves previously unseen levels of accuracy and robustness, greatly improving over previous methods that tend to skew towards only either accuracy or robustness.
The success of deep learning has enabled advances in multimodal tasks that require non-trivial fusion of multiple input domains. Although multimodal models have shown potential in many problems, their increased complexity makes them more vulnerable to attacks. A Backdoor (or Trojan) attack is a class of security vulnerability wherein an attacker embeds a malicious secret behavior into a network (e.g. targeted misclassification) that is activated when an attacker-specified trigger is added to an input. In this work, we show that multimodal networks are vulnerable to a novel type of attack that we refer to as Dual-Key Multimodal Backdoors. This attack exploits the complex fusion mechanisms used by state-of-the-art networks to embed backdoors that are both effective and stealthy. Instead of using a single trigger, the proposed attack embeds a trigger in each of the input modalities and activates the malicious behavior only when both the triggers are present. We present an extensive study of multimodal backdoors on the Visual Question Answering (VQA) task with multiple architectures and visual feature backbones. A major challenge in embedding backdoors in VQA models is that most models use visual features extracted from a fixed pretrained object detector. This is challenging for the attacker as the detector can distort or ignore the visual trigger entirely, which leads to models where backdoors are over-reliant on the language trigger. We tackle this problem by proposing a visual trigger optimization strategy designed for pretrained object detectors. Through this method, we create Dual-Key Backdoors with over a 98% attack success rate while only poisoning 1% of the training data. Finally, we release TrojVQA, a large collection of clean and trojan VQA models to enable research in defending against multimodal backdoors.
Deep Neural Networks (DNNs) are vulnerable to invisible perturbations on the images generated by adversarial attacks, which raises researches on the adversarial robustness of DNNs. A series of methods represented by the adversarial training and its variants have proven as one of the most effective techniques in enhancing the DNN robustness. Generally, adversarial training focuses on enriching the training data by involving perturbed data. Despite of the efficiency in defending specific attacks, adversarial training is benefited from the data augmentation, which does not contribute to the robustness of DNN itself and usually suffers from accuracy drop on clean data as well as inefficiency in unknown attacks. Towards the robustness of DNN itself, we propose a novel defense that aims at augmenting the model in order to learn features adaptive to diverse inputs, including adversarial examples. Specifically, we introduce multiple paths to augment the network, and impose orthogonality constraints on these paths. In addition, a margin-maximization loss is designed to further boost DIversity via Orthogonality (DIO). Extensive empirical results on various data sets, architectures, and attacks demonstrate the adversarial robustness of the proposed DIO.
Adversarial training (i.e., training on adversarially perturbed input data) is a well-studied method for making neural networks robust to potential adversarial attacks during inference. However, the improved robustness does not come for free but rather is accompanied by a decrease in overall model accuracy and performance. Recent work has shown that, in practical robot learning applications, the effects of adversarial training do not pose a fair trade-off but inflict a net loss when measured in holistic robot performance. This work revisits the robustness-accuracy trade-off in robot learning by systematically analyzing if recent advances in robust training methods and theory in conjunction with adversarial robot learning can make adversarial training suitable for real-world robot applications. We evaluate a wide variety of robot learning tasks ranging from autonomous driving in a high-fidelity environment amenable to sim-to-real deployment, to mobile robot gesture recognition. Our results demonstrate that, while these techniques make incremental improvements on the trade-off on a relative scale, the negative side-effects caused by adversarial training still outweigh the improvements by an order of magnitude. We conclude that more substantial advances in robust learning methods are necessary before they can benefit robot learning tasks in practice.
Adversarial attack is a technique for deceiving Machine Learning (ML) models, which provides a way to evaluate the adversarial robustness. In practice, attack algorithms are artificially selected and tuned by human experts to break a ML system. However, manual selection of attackers tends to be sub-optimal, leading to a mistakenly assessment of model security. In this paper, a new procedure called Composite Adversarial Attack (CAA) is proposed for automatically searching the best combination of attack algorithms and their hyper-parameters from a candidate pool of \textbf{32 base attackers}. We design a search space where attack policy is represented as an attacking sequence, i.e., the output of the previous attacker is used as the initialization input for successors. Multi-objective NSGA-II genetic algorithm is adopted for finding the strongest attack policy with minimum complexity. The experimental result shows CAA beats 10 top attackers on 11 diverse defenses with less elapsed time (\textbf{6 $\times$ faster than AutoAttack}), and achieves the new state-of-the-art on $l_{\infty}$, $l_{2}$ and unrestricted adversarial attacks.
Recent advances in maximizing mutual information (MI) between the source and target have demonstrated its effectiveness in text generation. However, previous works paid little attention to modeling the backward network of MI (i.e., dependency from the target to the source), which is crucial to the tightness of the variational information maximization lower bound. In this paper, we propose Adversarial Mutual Information (AMI): a text generation framework which is formed as a novel saddle point (min-max) optimization aiming to identify joint interactions between the source and target. Within this framework, the forward and backward networks are able to iteratively promote or demote each other's generated instances by comparing the real and synthetic data distributions. We also develop a latent noise sampling strategy that leverages random variations at the high-level semantic space to enhance the long term dependency in the generation process. Extensive experiments based on different text generation tasks demonstrate that the proposed AMI framework can significantly outperform several strong baselines, and we also show that AMI has potential to lead to a tighter lower bound of maximum mutual information for the variational information maximization problem.
For better user experience and business effectiveness, Click-Through Rate (CTR) prediction has been one of the most important tasks in E-commerce. Although extensive CTR prediction models have been proposed, learning good representation of items from multimodal features is still less investigated, considering an item in E-commerce usually contains multiple heterogeneous modalities. Previous works either concatenate the multiple modality features, that is equivalent to giving a fixed importance weight to each modality; or learn dynamic weights of different modalities for different items through technique like attention mechanism. However, a problem is that there usually exists common redundant information across multiple modalities. The dynamic weights of different modalities computed by using the redundant information may not correctly reflect the different importance of each modality. To address this, we explore the complementarity and redundancy of modalities by considering modality-specific and modality-invariant features differently. We propose a novel Multimodal Adversarial Representation Network (MARN) for the CTR prediction task. A multimodal attention network first calculates the weights of multiple modalities for each item according to its modality-specific features. Then a multimodal adversarial network learns modality-invariant representations where a double-discriminators strategy is introduced. Finally, we achieve the multimodal item representations by combining both modality-specific and modality-invariant representations. We conduct extensive experiments on both public and industrial datasets, and the proposed method consistently achieves remarkable improvements to the state-of-the-art methods. Moreover, the approach has been deployed in an operational E-commerce system and online A/B testing further demonstrates the effectiveness.
Adversarial attacks to image classification systems present challenges to convolutional networks and opportunities for understanding them. This study suggests that adversarial perturbations on images lead to noise in the features constructed by these networks. Motivated by this observation, we develop new network architectures that increase adversarial robustness by performing feature denoising. Specifically, our networks contain blocks that denoise the features using non-local means or other filters; the entire networks are trained end-to-end. When combined with adversarial training, our feature denoising networks substantially improve the state-of-the-art in adversarial robustness in both white-box and black-box attack settings. On ImageNet, under 10-iteration PGD white-box attacks where prior art has 27.9% accuracy, our method achieves 55.7%; even under extreme 2000-iteration PGD white-box attacks, our method secures 42.6% accuracy. A network based on our method was ranked first in Competition on Adversarial Attacks and Defenses (CAAD) 2018 --- it achieved 50.6% classification accuracy on a secret, ImageNet-like test dataset against 48 unknown attackers, surpassing the runner-up approach by ~10%. Code and models will be made publicly available.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191