Adversarial machine learning

At the MIT Spam Conference in January 2004, John Graham-Cumming showed that a machine-learning spam filter could be used to defeat another machine-learning spam filter by automatically learning which words to add to a spam email to get the email classified as not spam. In 2004, Nilesh Dalvi and others noted that linear classifiers used in spam filters could be defeated by simple "evasion attacks" as spammers inserted "good words" into their spam emails. (Around 2007, some spammers added random noise to fuzz words within "image spam" in order to defeat OCR-based filters.) In 2006, Marco Barreno and others published "Can Machine Learning Be Secure?", outlining a broad taxonomy of attacks. As late as 2013 many researchers continued to hope that non-linear classifiers (such as support vector machines and neural networks) might be robust to adversaries, until Battista Biggio and others demonstrated the first gradient-based attacks on such machine-learning models (2012–2013). In 2012, deep neural networks began to dominate computer vision problems; starting in 2014, Christian Szegedy and others demonstrated that deep neural networks could be fooled by adversaries, again using a gradient-based attack to craft adversarial perturbations. Further work would show that adversarial attacks are harder to produce in uncontrolled environments, due to the different environmental constraints that cancel out the effect of noise. For example, any small rotation or slight illumination on an adversarial image can destroy the adversariality. In addition, researchers such as Google Brain's Nick Frosst point out that it is much easier to make self-driving cars miss stop signs by physically removing the sign itself, rather than creating adversarial examples. Frosst also believes that the adversarial machine learning community incorrectly assumes models trained on a certain data distribution will also perform well on a completely different data distribution. He suggests that a new approach to machine learning should be explored, and is currently working on a unique neural network that has characteristics more similar to human perception than state-of-the-art approaches. Examples Examples include attacks in spam filtering, where spam messages are obfuscated through the misspelling of "bad" words or the insertion of "good" words; attacks in computer security, such as obfuscating malware code within network packets or modifying the characteristics of a network flow to mislead intrusion detection; attacks in biometric recognition where fake biometric traits may be exploited to impersonate a legitimate user; or to compromise users' template galleries that adapt to updated traits over time. Researchers showed that by changing only one-pixel it was possible to fool deep learning algorithms. Others 3-D printed a toy turtle with a texture engineered to make Google's object detection AI classify it as a rifle regardless of the angle from which the turtle was viewed. Creating the turtle required only low-cost commercially available 3-D printing technology. A machine-tweaked image of a dog was shown to look like a cat to both computers and humans. A 2019 study reported that humans can guess how machines will classify adversarial images. Researchers discovered methods for perturbing the appearance of a stop sign such that an autonomous vehicle classified it as a merge or speed limit sign. A data poisoning filter called Nightshade was released in 2023 by researchers at the University of Chicago. It was created for use by visual artists to put on their artwork to corrupt the data set of text-to-image models, which usually scrape their data from the internet without the consent of the image creator. McAfee attacked Tesla's former Mobileye system, fooling it into driving 50 mph over the speed limit, simply by adding a two-inch strip of black tape to a speed limit sign. Adversarial patterns on glasses or clothing designed to deceive facial-recognition systems or license-plate readers, have led to a niche industry of "stealth streetwear". An adversarial attack on a neural network can allow an attacker to inject algorithms into the target system. Researchers can also create adversarial audio inputs to disguise commands to intelligent assistants in benign-seeming audio; a parallel literature explores human perception of such stimuli. Clustering algorithms are used in security applications. Malware and computer virus analysis aims to identify malware families, and to generate specific detection signatures. In the context of malware detection, researchers have proposed methods for adversarial malware generation that automatically craft binaries to evade learning-based detectors while preserving malicious functionality. Optimization-based attacks such as GAMMA use genetic algorithms to inject benign content (for example, padding or new PE sections) into Windows executables, framing evasion as a constrained optimization problem that balances misclassification success with the size of the injected payload and showing transferability to commercial antivirus products. Complementary work uses generative adversarial networks (GANs) to learn feature-space perturbations that cause malware to be classified as benign; Mal-LSGAN, for instance, replaces the standard GAN loss with a least-squares objective and modified activation functions to improve training stability and produce adversarial malware examples that substantially reduce true positive rates across multiple detectors. == Challenges in applying machine learning to security ==

Researchers have observed that the constraints under which machine-learning techniques function in the security domain are different from those of common benchmark domains. Security data may change over time, include mislabeled samples, or reflect adversarial behavior, which complicates evaluation and reproducibility. Data collection issues Security datasets vary across formats, including binaries, network traces, and log files. Studies have reported that the process of converting these sources into features can introduce bias or inconsistencies. Prior approaches to this problem have included both data-level solutions and sequence-specific models. Methods like n-gram and Long Short-Term Memory (LSTM) networks can model sequential data, but their performance has been shown to decline significantly when malware samples are realistically proportioned in the training set, demonstrating the limitations in realistic security contexts. Spurious correlations result when a model learns to associate artifacts with a label, rather than the underlying security-relevant pattern. For example, a malware classifier might learn to identify a specific compiler artifact instead of malicious behavior itself. Biased parameter selection is a form of data snooping where model hyperparameters are tuned using the test set. Performance evaluation The choice of the evaluation metrics can impact the validity of the results. Having an inappropriate baseline involves failing to compare a new model against simpler, well-established baselines. Inappropriate performance measures means using metrics that do not align with the practical goals of the system. Reporting only "accuracy" is often described as insufficient for an intrusion detection system, where false-positive rates are considered critically important. Base rate fallacy is a failure to correctly interpret performance in the context of large class imbalances. Deployment and operation Deployment introduces challenges related to the performance and security in live environments. Lab-only evaluation is the practice of evaluating a system only in a controlled, static laboratory setting, which does not account for real-world challenges like concept drift and performance overhead. Inappropriate threat model refers to failing to consider the ML system itself as an attack surface. ==Attack modalities==

Taxonomy Attacks against (supervised) machine learning algorithms have been categorized along three primary axes: influence on the classifier, the security violation and their specificity. • Classifier influence: An attack can influence the classifier by disrupting the classification phase. This may be preceded by an exploration phase to identify vulnerabilities. The attacker's capabilities might be restricted by the presence of data manipulation constraints. • Security violation: An attack can supply malicious data that gets classified as legitimate. Malicious data supplied during training can cause legitimate data to be rejected after training. • Specificity: A targeted attack attempts to allow a specific intrusion/disruption. Alternatively, an indiscriminate attack creates general mayhem. This taxonomy has been extended into a more comprehensive threat model that allows explicit assumptions about the adversary's goal, knowledge of the attacked system, capability of manipulating the input data/system components, and on attack strategy. This taxonomy has further been extended to include dimensions for defense strategies against adversarial attacks. Strategies Below are some of the most commonly encountered attack scenarios. Data poisoning Poisoning consists of contaminating the training dataset with data designed to increase errors in the output. Given that learning algorithms are shaped by their training datasets, poisoning can effectively reprogram algorithms with potentially malicious intent. Concerns have been raised especially for user-generated training data, e.g. for content recommendation or natural language models. The ubiquity of fake accounts offers many opportunities for poisoning. Facebook reportedly removes around 7 billion fake accounts per year. Poisoning has been reported as the leading concern for industrial applications. On social medias, disinformation campaigns attempt to bias recommendation and moderation algorithms, to push certain content over others. A particular case of data poisoning is the backdoor attack, which aims to teach a specific behavior for inputs with a given trigger, e.g. a small defect on images, sounds, videos or texts. For instance, intrusion detection systems are often trained using collected data. An attacker may poison this data by injecting malicious samples during operation that subsequently disrupt retraining. Data poisoning techniques can also be applied to text-to-image models to alter their output, which is used by artists to defend their copyrighted works or their artistic style against imitation. Data poisoning can also happen unintentionally through model collapse, where models are trained on synthetic data. Byzantine attacks As machine learning is scaled, it often relies on multiple computing machines. In federated learning, for instance, edge devices collaborate with a central server, typically by sending gradients or model parameters. However, some of these devices may deviate from their expected behavior, e.g. to harm the central server's model or to bias algorithms towards certain behaviors (e.g., amplifying the recommendation of disinformation content). On the other hand, if the training is performed on a single machine, then the model is very vulnerable to a failure of the machine, or an attack on the machine; the machine is a single point of failure. In fact, the machine owner may themselves insert provably undetectable backdoors. The current leading solutions to make (distributed) learning algorithms provably resilient to a minority of malicious (a.k.a. Byzantine) participants are based on robust gradient aggregation rules. The robust aggregation rules do not always work especially when the data across participants has a non-iid distribution. Nevertheless, in the context of heterogeneous honest participants, such as users with different consumption habits for recommendation algorithms or writing styles for language models, there are provable impossibility theorems on what any robust learning algorithm can guarantee. Evasion Evasion attacks consist of exploiting the imperfection of a trained model. For instance, spammers and hackers often attempt to evade detection by obfuscating the content of spam emails and malware. Samples are modified to evade detection; that is, to be classified as legitimate. This does not involve influence over the training data. A clear example of evasion is image-based spam in which the spam content is embedded within an attached image to evade textual analysis by anti-spam filters. Another example of evasion is given by spoofing attacks against biometric verification systems. This can cause issues when either the training data or the model itself is sensitive and confidential. For example, model extraction could be used to extract a proprietary stock trading model which the adversary could then use for their own financial benefit. In the extreme case, model extraction can lead to model stealing, which corresponds to extracting a sufficient amount of data from the model to enable the complete reconstruction of the model. On the other hand, membership inference is a targeted model extraction attack, which infers the owner of a data point, often by leveraging the overfitting resulting from poor machine learning practices. Concerningly, this is sometimes achievable even without knowledge or access to a target model's parameters, raising security concerns for models trained on sensitive data, including but not limited to medical records and/or personally identifiable information. With the emergence of transfer learning and public accessibility of many state of the art machine learning models, tech companies are increasingly drawn to create models based on public ones, giving attackers freely accessible information to the structure and type of model being used. == Categories ==

Adversarial attacks and training in linear models There is a growing literature about adversarial attacks in linear models. Indeed, since the seminal work from Goodfellow at al. studying these models in linear models has been an important tool to understand how adversarial attacks affect machine learning models. The analysis of these models is simplified because the computation of adversarial attacks can be simplified in linear regression and classification problems. Moreover, adversarial training is convex in this case. Linear models allow for analytical analysis while still reproducing phenomena observed in state-of-the-art models. One prime example of that is how this model can be used to explain the trade-off between robustness and accuracy. Diverse work indeed provides analysis of adversarial attacks in linear models, including asymptotic analysis for classification and for linear regression. And, finite-sample analysis based on Rademacher complexity. A result from studying adversarial attacks in linear models is that it closely relates to regularization. Under certain conditions, it has been shown that • adversarial training of a linear regression model with input perturbations restricted by the infinity-norm closely resembles Lasso regression, and that • adversarial training of a linear regression model with input perturbations restricted by the 2-norm closely resembles Ridge regression. Adversarial deep reinforcement learning Adversarial deep reinforcement learning is an active area of research in reinforcement learning focusing on vulnerabilities of learned policies. In this research area, some studies initially showed that reinforcement learning policies are susceptible to imperceptible adversarial manipulations. While some methods have been proposed to overcome these susceptibilities, in the most recent studies it has been shown that these proposed solutions are far from providing an accurate representation of current vulnerabilities of deep reinforcement learning policies. Adversarial natural language processing Adversarial attacks on speech recognition have been introduced for speech-to-text applications, in particular for Mozilla's implementation of DeepSpeech and Google's Speech Commands model. Various pre-processing techniques have been proposed to defend against such attacks , but these methods may not be resilient to attackers aware of those defenses. == Specific attack types ==

There are a large variety of different adversarial attacks that can be used against machine learning systems. Many of these work on both deep learning systems as well as traditional machine learning models such as SVMs A high level sample of these attack types include: • Adversarial examples • Trojan and backdoor attacks • Model inversion • Membership inference Adversarial examples An adversarial example refers to specially crafted input that is designed to look "normal" to humans but causes misclassification to a machine learning model. Often, a form of specially designed "noise" is used to elicit the misclassifications. Below are some current techniques for generating adversarial examples in the literature (by no means an exhaustive list). • Gradient-based evasion attack • Projected Gradient Descent (PGD) • Carlini and Wagner (C&W) attack • Adversarial patch attack Black box attacks Black box attacks in adversarial machine learning assume that the adversary can only get outputs for provided inputs and has no knowledge of the model structure or parameters. In this case, the adversarial example is generated either using a model created from scratch, or without any model at all (excluding the ability to query the original model). In either case, the objective of these attacks is to create adversarial examples that are able to transfer to the black box model in question. Simple black-box adversarial attacks Simple black-box adversarial attacks is a query-efficient way to attack black-box image classifiers. Take a random orthonormal basis v_1, v_2, \dots, v_d in \R^d. The authors suggested the discrete cosine transform of the standard basis (the pixels). For a correctly classified image x, try x + \epsilon v_1, x - \epsilon v_1, and compare the amount of error in the classifier upon x + \epsilon v_1, x, x - \epsilon v_1. Pick the one that causes the largest amount of error. Repeat this for v_2, v_3, \dots until the desired level of error in the classifier is reached.It was discovered when the authors designed a simple baseline to compare with a previous black-box adversarial attack algorithm based on gaussian processes, and were surprised that the baseline worked even better. Square attack Square attacks were introduced in 2020 as a black-box evasion adversarial attack technique based on querying classification scores without the need for gradient information. As a score based black box attack, this adversarial approach is able to query probability distributions across model output classes, but has no other access to the model itself. According to the paper's authors, the proposed Square Attack required fewer queries than when compared to state-of-the-art score-based black box attacks at the time. \textbf{Targeted: } \min_{x^\prime}d(x^\prime, x) \text{ subject to } C(x^\prime) = c^* \textbf{Untargeted: } \min_{x^\prime}d(x^\prime, x) \text{ subject to } C(x^\prime) \neq C(x) To solve this problem, the attack proposes the following boundary function S for both the untargeted and targeted setting: The attack was called fast gradient sign method (FGSM), and it consists of adding a linear amount of in-perceivable noise to the image and causing a model to incorrectly classify it. This noise is calculated by multiplying the sign of the gradient with respect to the image we want to perturb by a small constant epsilon. As epsilon increases, the model is more likely to be fooled, but the perturbations become easier to identify as well. Shown below is the equation to generate an adversarial example where x is the original image, \epsilon is a very small number, \Delta_x is the gradient function, J is the loss function, \theta is the model weights, and y is the true label. adv_x = x + \epsilon \cdot sign(\Delta_xJ(\theta, x, y)) One important property of this equation is that the gradient is calculated with respect to the input image since the goal is to generate an image that maximizes the loss for the original image of true label y. In traditional gradient descent (for model training), the gradient is used to update the weights of the model since the goal is to minimize the loss for the model on a ground truth dataset. The Fast Gradient Sign Method was proposed as a fast way to generate adversarial examples to evade the model, based on the hypothesis that neural networks cannot resist even linear amounts of perturbation to the input. Carlini & Wagner In an effort to analyze existing adversarial attacks and defenses, researchers at the University of California, Berkeley, Nicholas Carlini and David Wagner in 2016 propose a faster and more robust method to generate adversarial examples. The attack proposed by Carlini and Wagner begins with trying to solve a difficult non-linear optimization equation: == Defenses ==

-style competitions • Game theoretic models • Sanitizing training data • Adversarial training • Gradient masking/obfuscation techniques: to prevent the adversary exploiting the gradient in white-box attacks. This family of defenses is deemed unreliable as these models are still vulnerable to black-box attacks or can be circumvented in other ways. • Ensembles of models have been proposed in literature, which have shown to be ineffective against evasion attacks but effective against data poisoning attacks. • Moving target defense ==See also==