While machine learning (ML) has made tremendous progress during the past decade, recent research has shown that ML models are vulnerable to various security and privacy attacks. So far, most of the attacks in this field focus on discriminative models, represented by classifiers. Meanwhile, little attention has been paid to the security and privacy risks of generative models, such as generative adversarial networks (GANs). In this paper, we propose the first set of training dataset property inference attacks against GANs. Concretely, the adversary aims to infer the macro-level training dataset property, i.e., the proportion of samples used to train a target GAN with respect to a certain attribute. A successful property inference attack can allow the adversary to gain extra knowledge of the target GAN's training dataset, thereby directly violating the intellectual property of the target model owner. Also, it can be used as a fairness auditor to check whether the target GAN is trained with a biased dataset. Besides, property inference can serve as a building block for other advanced attacks, such as membership inference. We propose a general attack pipeline that can be tailored to two attack scenarios, including the full black-box setting and partial black-box setting. For the latter, we introduce a novel optimization framework to increase the attack efficacy. Extensive experiments over four representative GAN models on five property inference tasks show that our attacks achieve strong performance. In addition, we show that our attacks can be used to enhance the performance of membership inference against GANs.
A membership inference attack (MIA) poses privacy risks for the training data of a machine learning model. With an MIA, an attacker guesses if the target data are a member of the training dataset. The state-of-the-art defense against MIAs, distillation for membership privacy (DMP), requires not only private data for protection but a large amount of unlabeled public data. However, in certain privacy-sensitive domains, such as medicine and finance, the availability of public data is not guaranteed. Moreover, a trivial method for generating public data by using generative adversarial networks significantly decreases the model accuracy, as reported by the authors of DMP. To overcome this problem, we propose a novel defense against MIAs that uses knowledge distillation without requiring public data. Our experiments show that the privacy protection and accuracy of our defense are comparable to those of DMP for the benchmark tabular datasets used in MIA research, Purchase100 and Texas100, and our defense has a much better privacy-utility trade-off than those of the existing defenses that also do not use public data for the image dataset CIFAR10.
In cross-device federated learning (FL) setting, clients such as mobiles cooperate with the server to train a global machine learning model, while maintaining their data locally. However, recent work shows that client's private information can still be disclosed to an adversary who just eavesdrops the messages exchanged between the client and the server. For example, the adversary can infer whether the client owns a specific data instance, which is called a passive membership inference attack. In this paper, we propose a new passive inference attack that requires much less computation power and memory than existing methods. Our empirical results show that our attack achieves a higher accuracy on CIFAR100 dataset (more than $4$ percentage points) with three orders of magnitude less memory space and five orders of magnitude less calculations.
Graph Neural Networks (GNNs) are widely adopted to analyse non-Euclidean data, such as chemical networks, brain networks, and social networks, modelling complex relationships and interdependency between objects. Recently, Membership Inference Attack (MIA) against GNNs raises severe privacy concerns, where training data can be leaked from trained GNN models. However, prior studies focus on inferring the membership of only the components in a graph, e.g., an individual node or edge. How to infer the membership of an entire graph record is yet to be explored. In this paper, we take the first step in MIA against GNNs for graph-level classification. Our objective is to infer whether a graph sample has been used for training a GNN model. We present and implement two types of attacks, i.e., training-based attacks and threshold-based attacks from different adversarial capabilities. We perform comprehensive experiments to evaluate our attacks in seven real-world datasets using five representative GNN models. Both our attacks are shown effective and can achieve high performance, i.e., reaching over 0.7 attack F1 scores in most cases. Furthermore, we analyse the implications behind the MIA against GNNs. Our findings confirm that GNNs can be even more vulnerable to MIA than the models with non-graph structures. And unlike the node-level classifier, MIAs on graph-level classification tasks are more co-related with the overfitting level of GNNs rather than the statistic property of their training graphs.
Machine Learning (ML) has emerged as a core technology to provide learning models to perform complex tasks. Boosted by Machine Learning as a Service (MLaaS), the number of applications relying on ML capabilities is ever increasing. However, ML models are the source of different privacy violations through passive or active attacks from different entities. In this paper, we present MixNN a proxy-based privacy-preserving system for federated learning to protect the privacy of participants against a curious or malicious aggregation server trying to infer sensitive attributes. MixNN receives the model updates from participants and mixes layers between participants before sending the mixed updates to the aggregation server. This mixing strategy drastically reduces privacy without any trade-off with utility. Indeed, mixing the updates of the model has no impact on the result of the aggregation of the updates computed by the server. We experimentally evaluate MixNN and design a new attribute inference attack, Sim, exploiting the privacy vulnerability of SGD algorithm to quantify privacy leakage in different settings (i.e., the aggregation server can conduct a passive or an active attack). We show that MixNN significantly limits the attribute inference compared to a baseline using noisy gradient (well known to damage the utility) while keeping the same level of utility as classic federated learning.
Federated learning (FL) has emerged as a promising privacy-aware paradigm that allows multiple clients to jointly train a model without sharing their private data. Recently, many studies have shown that FL is vulnerable to membership inference attacks (MIAs) that can distinguish the training members of the given model from the non-members. However, existing MIAs ignore the source of a training member, i.e., the information of which client owns the training member, while it is essential to explore source privacy in FL beyond membership privacy of examples from all clients. The leakage of source information can lead to severe privacy issues. For example, identification of the hospital contributing to the training of an FL model for COVID-19 pandemic can render the owner of a data record from this hospital more prone to discrimination if the hospital is in a high risk region. In this paper, we propose a new inference attack called source inference attack (SIA), which can derive an optimal estimation of the source of a training member. Specifically, we innovatively adopt the Bayesian perspective to demonstrate that an honest-but-curious server can launch an SIA to steal non-trivial source information of the training members without violating the FL protocol. The server leverages the prediction loss of local models on the training members to achieve the attack effectively and non-intrusively. We conduct extensive experiments on one synthetic and five real datasets to evaluate the key factors in an SIA, and the results show the efficacy of the proposed source inference attack.
Distribution inference, sometimes called property inference, infers statistical properties about a training set from access to a model trained on that data. Distribution inference attacks can pose serious risks when models are trained on private data, but are difficult to distinguish from the intrinsic purpose of statistical machine learning -- namely, to produce models that capture statistical properties about a distribution. Motivated by Yeom et al.'s membership inference framework, we propose a formal definition of distribution inference attacks that is general enough to describe a broad class of attacks distinguishing between possible training distributions. We show how our definition captures previous ratio-based property inference attacks as well as new kinds of attack including revealing the average node degree or clustering coefficient of a training graph. To understand distribution inference risks, we introduce a metric that quantifies observed leakage by relating it to the leakage that would occur if samples from the training distribution were provided directly to the adversary. We report on a series of experiments across a range of different distributions using both novel black-box attacks and improved versions of the state-of-the-art white-box attacks. Our results show that inexpensive attacks are often as effective as expensive meta-classifier attacks, and that there are surprising asymmetries in the effectiveness of attacks. Code is available at https://github.com/iamgroot42/FormEstDistRisks
While significant research advances have been made in the field of deep reinforcement learning, there have been no concrete adversarial attack strategies in literature tailored for studying the vulnerability of deep reinforcement learning algorithms to membership inference attacks. In such attacking systems, the adversary targets the set of collected input data on which the deep reinforcement learning algorithm has been trained. To address this gap, we propose an adversarial attack framework designed for testing the vulnerability of a state-of-the-art deep reinforcement learning algorithm to a membership inference attack. In particular, we design a series of experiments to investigate the impact of temporal correlation, which naturally exists in reinforcement learning training data, on the probability of information leakage. Moreover, we compare the performance of \emph{collective} and \emph{individual} membership attacks against the deep reinforcement learning algorithm. Experimental results show that the proposed adversarial attack framework is surprisingly effective at inferring data with an accuracy exceeding $84\%$ in individual and $97\%$ in collective modes in three different continuous control Mujoco tasks, which raises serious privacy concerns in this regard. Finally, we show that the learning state of the reinforcement learning algorithm influences the level of privacy breaches significantly.
The vulnerability of the Lottery Ticket Hypothesis has not been studied from the purview of Membership Inference Attacks. Through this work, we are the first to empirically show that the lottery ticket networks are equally vulnerable to membership inference attacks. A Membership Inference Attack (MIA) is the process of determining whether a data sample belongs to a training set of a trained model or not. Membership Inference Attacks could leak critical information about the training data that can be used for targeted attacks. Recent deep learning models often have very large memory footprints and a high computational cost associated with training and drawing inferences. Lottery Ticket Hypothesis is used to prune the networks to find smaller sub-networks that at least match the performance of the original model in terms of test accuracy in a similar number of iterations. We used CIFAR-10, CIFAR-100, and ImageNet datasets to perform image classification tasks and observe that the attack accuracies are similar. We also see that the attack accuracy varies directly according to the number of classes in the dataset and the sparsity of the network. We demonstrate that these attacks are transferable across models with high accuracy.
An over-the-air membership inference attack (MIA) is presented to leak private information from a wireless signal classifier. Machine learning (ML) provides powerful means to classify wireless signals, e.g., for PHY-layer authentication. As an adversarial machine learning attack, the MIA infers whether a signal of interest has been used in the training data of a target classifier. This private information incorporates waveform, channel, and device characteristics, and if leaked, can be exploited by an adversary to identify vulnerabilities of the underlying ML model (e.g., to infiltrate the PHY-layer authentication). One challenge for the over-the-air MIA is that the received signals and consequently the RF fingerprints at the adversary and the intended receiver differ due to the discrepancy in channel conditions. Therefore, the adversary first builds a surrogate classifier by observing the spectrum and then launches the black-box MIA on this classifier. The MIA results show that the adversary can reliably infer signals (and potentially the radio and channel information) used to build the target classifier. Therefore, a proactive defense is developed against the MIA by building a shadow MIA model and fooling the adversary. This defense can successfully reduce the MIA accuracy and prevent information leakage from the wireless signal classifier.
In the text processing context, most ML models are built on word embeddings. These embeddings are themselves trained on some datasets, potentially containing sensitive data. In some cases this training is done independently, in other cases, it occurs as part of training a larger, task-specific model. In either case, it is of interest to consider membership inference attacks based on the embedding layer as a way of understanding sensitive information leakage. But, somewhat surprisingly, membership inference attacks on word embeddings and their effect in other natural language processing (NLP) tasks that use these embeddings, have remained relatively unexplored. In this work, we show that word embeddings are vulnerable to black-box membership inference attacks under realistic assumptions. Furthermore, we show that this leakage persists through two other major NLP applications: classification and text-generation, even when the embedding layer is not exposed to the attacker. We show that our MI attack achieves high attack accuracy against a classifier model and an LSTM-based language model. Indeed, our attack is a cheaper membership inference attack on text-generative models, which does not require the knowledge of the target model or any expensive training of text-generative models as shadow models.
The expanding usage of complex machine learning methods like deep learning has led to an explosion in human activity recognition, particularly applied to health. In particular, as part of a larger body sensor network system, face and full-body analysis is becoming increasingly common for evaluating health status. However, complex models which handle private and sometimes protected data, raise concerns about the potential leak of identifiable data. In this work, we focus on the case of a deep network model trained on images of individual faces. Full-face video recordings taken from 493 individuals undergoing an eye-tracking based evaluation of neurological function were used. Outputs, gradients, intermediate layer outputs, loss, and labels were used as inputs for a deep network with an added support vector machine emission layer to recognize membership in the training data. The inference attack method and associated mathematical analysis indicate that there is a low likelihood of unintended memorization of facial features in the deep learning model. In this study, it is showed that the named model preserves the integrity of training data with reasonable confidence. The same process can be implemented in similar conditions for different models.
The collection and availability of big data, combined with advances in pre-trained models (e.g., BERT, XLNET, etc), have revolutionized the predictive performance of modern natural language processing tasks, ranging from text classification to text generation. This allows corporations to provide machine learning as a service (MLaaS) by encapsulating fine-tuned BERT-based models as APIs. However, BERT-based APIs have exhibited a series of security and privacy vulnerabilities. For example, prior work has exploited the security issues of the BERT-based APIs through the adversarial examples crafted by the extracted model. However, the privacy leakage problems of the BERT-based APIs through the extracted model have not been well studied. On the other hand, due to the high capacity of BERT-based APIs, the fine-tuned model is easy to be overlearned, but what kind of information can be leaked from the extracted model remains unknown. In this work, we bridge this gap by first presenting an effective model extraction attack, where the adversary can practically steal a BERT-based API (the target/victim model) by only querying a limited number of queries. We further develop an effective attribute inference attack which can infer the sensitive attribute of the training data used by the BERT-based APIs. Our extensive experiments on benchmark datasets under various realistic settings validate the potential vulnerabilities of BERT-based APIs. Moreover, we demonstrate that two promising defense methods become ineffective against our attacks, which calls for more effective defense methods.
Deep ensemble learning has been shown to improve accuracy by training multiple neural networks and averaging their outputs. Ensemble learning has also been suggested to defend against membership inference attacks that undermine privacy. In this paper, we empirically demonstrate a trade-off between these two goals, namely accuracy and privacy (in terms of membership inference attacks), in deep ensembles. Using a wide range of datasets and model architectures, we show that the effectiveness of membership inference attacks increases when ensembling improves accuracy. We analyze the impact of various factors in deep ensembles and demonstrate the root cause of the trade-off. Then, we evaluate common defenses against membership inference attacks based on regularization and differential privacy. We show that while these defenses can mitigate the effectiveness of membership inference attacks, they simultaneously degrade ensemble accuracy. We illustrate similar trade-off in more advanced and state-of-the-art ensembling techniques, such as snapshot ensembles and diversified ensemble networks. Finally, we propose a simple yet effective defense for deep ensembles to break the trade-off and, consequently, improve the accuracy and privacy, simultaneously.
Recently, it has been shown that Machine Learning models can leak sensitive information about their training data. This information leakage is exposed through membership and attribute inference attacks. Although many attack strategies have been proposed, little effort has been made to formalize these problems. We present a novel formalism, generalizing membership and attribute inference attack setups previously studied in the literature and connecting them to memorization and generalization. First, we derive a universal bound on the success rate of inference attacks and connect it to the generalization gap of the target model. Second, we study the question of how much sensitive information is stored by the algorithm about its training set and we derive bounds on the mutual information between the sensitive attributes and model parameters. Experimentally, we illustrate the potential of our approach by applying it to both synthetic data and classification tasks on natural images. Finally, we apply our formalism to different attribute inference strategies, with which an adversary is able to recover the identity of writers in the PenDigits dataset.
Deep Neural Network (DNN) models have been shown to have high empirical privacy leakages. Clinical language models (CLMs) trained on clinical data have been used to improve performance in biomedical natural language processing tasks. In this work, we investigate the risks of training-data leakage through white-box or black-box access to CLMs. We design and employ membership inference attacks to estimate the empirical privacy leaks for model architectures like BERT and GPT2. We show that membership inference attacks on CLMs lead to non-trivial privacy leakages of up to 7%. Our results show that smaller models have lower empirical privacy leakages than larger ones, and masked LMs have lower leakages than auto-regressive LMs. We further show that differentially private CLMs can have improved model utility on clinical domain while ensuring low empirical privacy leakage. Lastly, we also study the effects of group-level membership inference and disease rarity on CLM privacy leakages.
Membership inference attacks (MIAs) infer whether a specific data record is used for target model training. MIAs have provoked many discussions in the information security community since they give rise to severe data privacy issues, especially for private and sensitive datasets. Knowledge Graphs (KGs), which describe domain-specific subjects and relationships among them, are valuable and sensitive, such as medical KGs constructed from electronic health records. However, the privacy threat to knowledge graphs is critical but rarely explored. In this paper, we conduct the first empirical evaluation of privacy threats to knowledge graphs triggered by knowledge graph embedding methods (KGEs). We propose three types of membership inference attacks: transfer attacks (TAs), prediction loss-based attacks (PLAs), and prediction correctness-based attacks (PCAs), according to attack difficulty levels. In the experiments, we conduct three inference attacks against four standard KGE methods over three benchmark datasets. In addition, we also propose the attacks against medical KG and financial KG. The results demonstrate that the proposed attack methods can easily explore the privacy leakage of knowledge graphs.
Training deep neural networks requires gradient estimation from data batches to update parameters. Gradients per parameter are averaged over a set of data and this has been presumed to be safe for privacy-preserving training in joint, collaborative, and federated learning applications. Prior work only showed the possibility of recovering input data given gradients under very restrictive conditions - a single input point, or a network with no non-linearities, or a small 32x32 px input batch. Therefore, averaging gradients over larger batches was thought to be safe. In this work, we introduce GradInversion, using which input images from a larger batch (8 - 48 images) can also be recovered for large networks such as ResNets (50 layers), on complex datasets such as ImageNet (1000 classes, 224x224 px). We formulate an optimization task that converts random noise into natural images, matching gradients while regularizing image fidelity. We also propose an algorithm for target class label recovery given gradients. We further propose a group consistency regularization framework, where multiple agents starting from different random seeds work together to find an enhanced reconstruction of original data batch. We show that gradients encode a surprisingly large amount of information, such that all the individual images can be recovered with high fidelity via GradInversion, even for complex datasets, deep networks, and large batch sizes.
Machine learning (ML) models have been widely applied to various applications, including image classification, text generation, audio recognition, and graph data analysis. However, recent studies have shown that ML models are vulnerable to membership inference attacks (MIAs), which aim to infer whether a data record was used to train a target model or not. MIAs on ML models can directly lead to a privacy breach. For example, via identifying the fact that a clinical record that has been used to train a model associated with a certain disease, an attacker can infer that the owner of the clinical record has the disease with a high chance. In recent years, MIAs have been shown to be effective on various ML models, e.g., classification models and generative models. Meanwhile, many defense methods have been proposed to mitigate MIAs. Although MIAs on ML models form a newly emerging and rapidly growing research area, there has been no systematic survey on this topic yet. In this paper, we conduct the first comprehensive survey on membership inference attacks and defenses. We provide the taxonomies for both attacks and defenses, based on their characterizations, and discuss their pros and cons. Based on the limitations and gaps identified in this survey, we point out several promising future research directions to inspire the researchers who wish to follow this area. This survey not only serves as a reference for the research community but also provides a clear description for researchers outside this research domain. To further help the researchers, we have created an online resource repository, which we will keep updated with future relevant work. Interested readers can find the repository at https://github.com/HongshengHu/membership-inference-machine-learning-literature.
With an increase in low-cost machine learning APIs, advanced machine learning models may be trained on private datasets and monetized by providing them as a service. However, privacy researchers have demonstrated that these models may leak information about records in the training dataset via membership inference attacks. In this paper, we take a closer look at another inference attack reported in literature, called attribute inference, whereby an attacker tries to infer missing attributes of a partially known record used in the training dataset by accessing the machine learning model as an API. We show that even if a classification model succumbs to membership inference attacks, it is unlikely to be susceptible to attribute inference attacks. We demonstrate that this is because membership inference attacks fail to distinguish a member from a nearby non-member. We call the ability of an attacker to distinguish the two (similar) vectors as strong membership inference. We show that membership inference attacks cannot infer membership in this strong setting, and hence inferring attributes is infeasible. However, under a relaxed notion of attribute inference, called approximate attribute inference, we show that it is possible to infer attributes close to the true attributes. We verify our results on three publicly available datasets, five membership, and three attribute inference attacks reported in literature.