Post-hoc explanation methods are used with the intent of providing insights about neural networks and are sometimes said to help engender trust in their outputs. However, popular explanations methods have been found to be fragile to minor perturbations of input features or model parameters. Relying on constraint relaxation techniques from non-convex optimization, we develop a method that upper-bounds the largest change an adversary can make to a gradient-based explanation via bounded manipulation of either the input features or model parameters. By propagating a compact input or parameter set as symbolic intervals through the forwards and backwards computations of the neural network we can formally certify the robustness of gradient-based explanations. Our bounds are differentiable, hence we can incorporate provable explanation robustness into neural network training. Empirically, our method surpasses the robustness provided by previous heuristic approaches. We find that our training method is the only method able to learn neural networks with certificates of explanation robustness across all six datasets tested.

Explainability has been widely stated as a cornerstone of the responsible and trustworthy use of machine learning models. With the ubiquitous use of Deep Neural Network (DNN) models expanding to risk-sensitive and safety-critical domains, many methods have been proposed to explain the decisions of these models. Recent years have also seen concerted efforts that have shown how such explanations can be distorted (attacked) by minor input perturbations. While there have been many surveys that review explainability methods themselves, there has been no effort hitherto to assimilate the different methods and metrics proposed to study the robustness of explanations of DNN models. In this work, we present a comprehensive survey of methods that study, understand, attack, and defend explanations of DNN models. We also present a detailed review of different metrics used to evaluate explanation methods, as well as describe attributional attack and defense methods. We conclude with lessons and take-aways for the community towards ensuring robust explanations of DNN model predictions.

Gradient-based explanation is the cornerstone of explainable deep networks, but it has been shown to be vulnerable to adversarial attacks. However, existing works measure the explanation robustness based on $\ell_p$-norm, which can be counter-intuitive to humans, who only pay attention to the top few salient features. We propose explanation ranking thickness as a more suitable explanation robustness metric. We then present a new practical adversarial attacking goal for manipulating explanation rankings. To mitigate the ranking-based attacks while maintaining computational feasibility, we derive surrogate bounds of the thickness that involve expensive sampling and integration. We use a multi-objective approach to analyze the convergence of a gradient-based attack to confirm that the explanation robustness can be measured by the thickness metric. We conduct experiments on various network architectures and diverse datasets to prove the superiority of the proposed methods, while the widely accepted Hessian-based curvature smoothing approaches are not as robust as our method.

We propose a method to learn deep ReLU-based classifiers that are provably robust against normbounded adversarial perturbations on the training data. For previously unseen examples, the approach is guaranteed to detect all adversarial examples, though it may flag some non-adversarial examples as well. The basic idea is to consider a convex outer approximation of the set of activations reachable through a norm-bounded perturbation, and we develop a robust optimization procedure that minimizes the worst case loss over this outer region (via a linear program). Crucially, we show that the dual problem to this linear program can be represented itself as a deep network similar to the backpropagation network, leading to very efficient optimization approaches that produce guaranteed bounds on the robust loss. The end result is that by executing a few more forward and backward passes through a slightly modified version of the original network (though possibly with much larger batch sizes), we can learn a classifier that is provably robust to any norm-bounded adversarial attack. We illustrate the approach on a number of tasks to train classifiers with robust adversarial guarantees (e.g. for MNIST, we produce a convolutional classifier that provably has less than 5.8% test error for any adversarial attack with bounded ∞ norm less than = 0.1), and code for all experiments is available at http://github.com/ locuslab/convex_adversarial.

Despite their impressive performance on diverse tasks, neural networks fail catastrophically in the presence of adversarial inputs-imperceptibly but adversarially perturbed versions of natural inputs. We have witnessed an arms race between defenders who attempt to train robust networks and attackers who try to construct adversarial examples. One promise of ending the arms race is developing certified defenses, ones which are provably robust against all attackers in some family. These certified defenses are based on convex relaxations which construct an upper bound on the worst case loss over all attackers in the family. Previous relaxations are loose on networks that are not trained against the respective relaxation. In this paper, we propose a new semidefinite relaxation for certifying robustness that applies to arbitrary ReLU networks. We show that our proposed relaxation is tighter than previous relaxations and produces meaningful robustness guarantees on three different foreign networks whose training objectives are agnostic to our proposed relaxation.32nd Conference on Neural Information Processing Systems (NIPS 2018),

Adversarial examples have attracted significant attention in machine learning, but the reasons for their existence and pervasiveness remain unclear. We demonstrate that adversarial examples can be directly attributed to the presence of non-robust features: features (derived from patterns in the data distribution) that are highly predictive, yet brittle and (thus) incomprehensible to humans. After capturing these features within a theoretical framework, we establish their widespread existence in standard datasets. Finally, we present a simple setting where we can rigorously tie the phenomena we observe in practice to a misalignment between the (human-specified) notion of robustness and the inherent geometry of the data.

Adaptive attacks have (rightfully) become the de facto standard for evaluating defenses to adversarial examples. We find, however, that typical adaptive evaluations are incomplete. We demonstrate that thirteen defenses recently published at ICLR, ICML and NeurIPS-and which illustrate a diverse set of defense strategies-can be circumvented despite attempting to perform evaluations using adaptive attacks. While prior evaluation papers focused mainly on the end result-showing that a defense was ineffective-this paper focuses on laying out the methodology and the approach necessary to perform an adaptive attack. Some of our attack strategies are generalizable, but no single strategy would have been sufficient for all defenses. This underlines our key message that adaptive attacks cannot be automated and always require careful and appropriate tuning to a given defense. We hope that these analyses will serve as guidance on how to properly perform adaptive attacks against defenses to adversarial examples, and thus will allow the community to make further progress in building more robust models.

While neural networks have achieved high accuracy on standard image classification benchmarks, their accuracy drops to nearly zero in the presence of small adversarial perturbations to test inputs. Defenses based on regularization and adversarial training have been proposed, but often followed by new, stronger attacks that defeat these defenses. Can we somehow end this arms race? In this work, we study this problem for neural networks with one hidden layer. We first propose a method based on a semidefinite relaxation that outputs a certificate that for a given network and test input, no attack can force the error to exceed a certain value. Second, as this certificate is differentiable, we jointly optimize it with the network parameters, providing an adaptive regularizer that encourages robustness against all attacks. On MNIST, our approach produces a network and a certificate that no attack that perturbs each pixel by at most = 0.1 can cause more than 35% test error.

在本讨论文件中，我们调查了有关机器学习模型鲁棒性的最新研究。随着学习算法在数据驱动的控制系统中越来越流行，必须确保它们对数据不确定性的稳健性，以维持可靠的安全至关重要的操作。我们首先回顾了这种鲁棒性的共同形式主义，然后继续讨论训练健壮的机器学习模型的流行和最新技术，以及可证明这种鲁棒性的方法。从强大的机器学习的这种统一中，我们识别并讨论了该地区未来研究的迫切方向。

Interpretability provides a means for humans to verify aspects of machine learning (ML) models and empower human+ML teaming in situations where the task cannot be fully automated. Different contexts require explanations with different properties. For example, the kind of explanation required to determine if an early cardiac arrest warning system is ready to be integrated into a care setting is very different from the type of explanation required for a loan applicant to help determine the actions they might need to take to make their application successful. Unfortunately, there is a lack of standardization when it comes to properties of explanations: different papers may use the same term to mean different quantities, and different terms to mean the same quantity. This lack of a standardized terminology and categorization of the properties of ML explanations prevents us from both rigorously comparing interpretable machine learning methods and identifying what properties are needed in what contexts. In this work, we survey properties defined in interpretable machine learning papers, synthesize them based on what they actually measure, and describe the trade-offs between different formulations of these properties. In doing so, we enable more informed selection of task-appropriate formulations of explanation properties as well as standardization for future work in interpretable machine learning.

Neural network interpretation methods, particularly feature attribution methods, are known to be fragile with respect to adversarial input perturbations. To address this, several methods for enhancing the local smoothness of the gradient while training have been proposed for attaining \textit{robust} feature attributions. However, the lack of considering the normalization of the attributions, which is essential in their visualizations, has been an obstacle to understanding and improving the robustness of feature attribution methods. In this paper, we provide new insights by taking such normalization into account. First, we show that for every non-negative homogeneous neural network, a naive $\ell_2$-robust criterion for gradients is \textit{not} normalization invariant, which means that two functions with the same normalized gradient can have different values. Second, we formulate a normalization invariant cosine distance-based criterion and derive its upper bound, which gives insight for why simply minimizing the Hessian norm at the input, as has been done in previous work, is not sufficient for attaining robust feature attribution. Finally, we propose to combine both $\ell_2$ and cosine distance-based criteria as regularization terms to leverage the advantages of both in aligning the local gradient. As a result, we experimentally show that models trained with our method produce much more robust interpretations on CIFAR-10 and ImageNet-100 without significantly hurting the accuracy, compared to the recent baselines. To the best of our knowledge, this is the first work to verify the robustness of interpretation on a larger-scale dataset beyond CIFAR-10, thanks to the computational efficiency of our method.

Deep learning algorithms have been shown to perform extremely well on many classical machine learning problems. However, recent studies have shown that deep learning, like other machine learning techniques, is vulnerable to adversarial samples: inputs crafted to force a deep neural network (DNN) to provide adversary-selected outputs. Such attacks can seriously undermine the security of the system supported by the DNN, sometimes with devastating consequences. For example, autonomous vehicles can be crashed, illicit or illegal content can bypass content filters, or biometric authentication systems can be manipulated to allow improper access. In this work, we introduce a defensive mechanism called defensive distillation to reduce the effectiveness of adversarial samples on DNNs. We analytically investigate the generalizability and robustness properties granted by the use of defensive distillation when training DNNs. We also empirically study the effectiveness of our defense mechanisms on two DNNs placed in adversarial settings. The study shows that defensive distillation can reduce effectiveness of sample creation from 95% to less than 0.5% on a studied DNN. Such dramatic gains can be explained by the fact that distillation leads gradients used in adversarial sample creation to be reduced by a factor of 10 30 . We also find that distillation increases the average minimum number of features that need to be modified to create adversarial samples by about 800% on one of the DNNs we tested.

尽管深度神经网络（DNN）在感知和控制任务中表现出令人难以置信的性能，但几个值得信赖的问题仍然是开放的。其中一个最讨论的主题是存在对抗扰动的存在，它在能够量化给定输入的稳健性的可提供技术上开辟了一个有趣的研究线。在这方面，来自分类边界的输入的欧几里德距离表示良好被证明的鲁棒性评估，作为最小的经济适用的逆势扰动。不幸的是，由于NN的非凸性质，计算如此距离非常复杂。尽管已经提出了几种方法来解决这个问题，但据我们所知，没有提出可证明的结果来估计和绑定承诺的错误。本文通过提出两个轻量级策略来寻找最小的对抗扰动来解决这个问题。不同于现有技术，所提出的方法允许与理论上的近似距离的误差估计理论配制。最后，据报道，据报道了大量实验来评估算法的性能并支持理论发现。所获得的结果表明，该策略近似于靠近分类边界的样品的理论距离，导致可提供对任何对抗攻击的鲁棒性保障。

许多最先进的ML模型在各种任务中具有优于图像分类的人类。具有如此出色的性能，ML模型今天被广泛使用。然而，存在对抗性攻击和数据中毒攻击的真正符合ML模型的稳健性。例如，Engstrom等人。证明了最先进的图像分类器可以容易地被任意图像上的小旋转欺骗。由于ML系统越来越纳入安全性和安全敏感的应用，对抗攻击和数据中毒攻击构成了相当大的威胁。本章侧重于ML安全的两个广泛和重要的领域：对抗攻击和数据中毒攻击。

从外界培训的机器学习模型可能会被数据中毒攻击损坏，将恶意指向到模型的培训集中。对这些攻击的常见防御是数据消毒：在培训模型之前首先过滤出异常培训点。在本文中，我们开发了三次攻击，可以绕过广泛的常见数据消毒防御，包括基于最近邻居，训练损失和奇异值分解的异常探测器。通过增加3％的中毒数据，我们的攻击成功地将Enron垃圾邮件检测数据集的测试错误从3％增加到24％，并且IMDB情绪分类数据集从12％到29％。相比之下，没有明确占据这些数据消毒防御的现有攻击被他们击败。我们的攻击基于两个想法：（i）我们协调我们的攻击将中毒点彼此放置在彼此附近，（ii）我们将每个攻击制定为受限制的优化问题，限制旨在确保中毒点逃避检测。随着这种优化涉及解决昂贵的Bilevel问题，我们的三个攻击对应于基于影响功能的近似近似这个问题的方式; minimax二元性;和karush-kuhn-tucker（kkt）条件。我们的结果强调了对数据中毒攻击产生更强大的防御的必要性。

经认证的稳健性是安全关键应用中的深度神经网络的理想性质，流行的训练算法可以通过计算其Lipschitz常数的全球界限来认证神经网络的鲁棒性。然而，这种界限往往松动：它倾向于过度规范神经网络并降低其自然精度。绑定的Lipschitz绑定可以在自然和认证的准确性之间提供更好的权衡，但通常很难根据网络的非凸起计算。在这项工作中，我们通过考虑激活函数（例如Relu）和权重矩阵之间的相互作用，提出了一种有效和培训的\ emph {本地} Lipschitz上限。具体地，当计算权重矩阵的诱发标准时，我们消除了相应的行和列，其中保证激活函数在每个给定数据点的邻域中是常数，它提供比全局Lipschitz常数的可怕更严格的绑定神经网络。我们的方法可用作插入式模块，以拧紧在许多可认证的训练算法中绑定的Lipschitz。此外，我们建议夹住激活功能（例如，Relu和Maxmin），具有可读的上限阈值和稀疏性损失，以帮助网络实现甚至更严格的本地嘴唇尖端。在实验上，我们表明我们的方法始终如一地优于Mnist，CiFar-10和Tinyimagenet数据集的清洁和认证准确性，具有各种网络架构的清洁和认证的准确性。

多项式网络（PNS）最近在面部和图像识别方面表现出了有希望的表现。但是，PNS的鲁棒性尚不清楚，因此获得证书对于使其在现实世界应用中的采用至关重要。基于分支和绑定（BAB）技术的Relu神经网络（NNS）上的现有验证算法不能微不足道地应用于PN验证。在这项工作中，我们设计了一种新的边界方法，该方法配备了BAB，用于全球融合保证，称为VPN。一个关键的见解是，我们获得的边界比间隔结合的传播基线更紧密。这可以通过MNIST，CIFAR10和STL10数据集的经验验证进行声音和完整的PN验证。我们认为我们的方法对NN验证具有自身的兴趣。

Neural networks provide state-of-the-art results for most machine learning tasks. Unfortunately, neural networks are vulnerable to adversarial examples: given an input x and any target classification t, it is possible to find a new input x that is similar to x but classified as t. This makes it difficult to apply neural networks in security-critical areas. Defensive distillation is a recently proposed approach that can take an arbitrary neural network, and increase its robustness, reducing the success rate of current attacks' ability to find adversarial examples from 95% to 0.5%.In this paper, we demonstrate that defensive distillation does not significantly increase the robustness of neural networks by introducing three new attack algorithms that are successful on both distilled and undistilled neural networks with 100% probability. Our attacks are tailored to three distance metrics used previously in the literature, and when compared to previous adversarial example generation algorithms, our attacks are often much more effective (and never worse). Furthermore, we propose using high-confidence adversarial examples in a simple transferability test we show can also be used to break defensive distillation. We hope our attacks will be used as a benchmark in future defense attempts to create neural networks that resist adversarial examples.

背景信息：在过去几年中，机器学习（ML）一直是许多创新的核心。然而，包括在所谓的“安全关键”系统中，例如汽车或航空的系统已经被证明是非常具有挑战性的，因为ML的范式转变为ML带来完全改变传统认证方法。目的：本文旨在阐明与ML为基础的安全关键系统认证有关的挑战，以及文献中提出的解决方案，以解决它们，回答问题的问题如何证明基于机器学习的安全关键系统？'方法：我们开展2015年至2020年至2020年之间发布的研究论文的系统文献综述（SLR），涵盖了与ML系统认证有关的主题。总共确定了217篇论文涵盖了主题，被认为是ML认证的主要支柱：鲁棒性，不确定性，解释性，验证，安全强化学习和直接认证。我们分析了每个子场的主要趋势和问题，并提取了提取的论文的总结。结果：单反结果突出了社区对该主题的热情，以及在数据集和模型类型方面缺乏多样性。它还强调需要进一步发展学术界和行业之间的联系，以加深域名研究。最后，它还说明了必须在上面提到的主要支柱之间建立连接的必要性，这些主要柱主要主要研究。结论：我们强调了目前部署的努力，以实现ML基于ML的软件系统，并讨论了一些未来的研究方向。

To rigorously certify the robustness of neural networks to adversarial perturbations, most state-of-the-art techniques rely on a triangle-shaped linear programming (LP) relaxation of the ReLU activation. While the LP relaxation is exact for a single neuron, recent results suggest that it faces an inherent "convex relaxation barrier" as additional activations are added, and as the attack budget is increased. In this paper, we propose a nonconvex relaxation for the ReLU relaxation, based on a low-rank restriction of a semidefinite programming (SDP) relaxation. We show that the nonconvex relaxation has a similar complexity to the LP relaxation, but enjoys improved tightness that is comparable to the much more expensive SDP relaxation. Despite nonconvexity, we prove that the verification problem satisfies constraint qualification, and therefore a Riemannian staircase approach is guaranteed to compute a near-globally optimal solution in polynomial time. Our experiments provide evidence that our nonconvex relaxation almost completely overcome the "convex relaxation barrier" faced by the LP relaxation.