Convolutional Neural Networks (CNNs) deployed in real-life applications such as autonomous vehicles have shown to be vulnerable to manipulation attacks, such as poisoning attacks and fine-tuning. Hence, it is essential to ensure the integrity and authenticity of CNNs because compromised models can produce incorrect outputs and behave maliciously. In this paper, we propose a self-contained tamper-proofing method, called DeepiSign, to ensure the integrity and authenticity of CNN models against such manipulation attacks. DeepiSign applies the idea of fragile invisible watermarking to securely embed a secret and its hash value into a CNN model. To verify the integrity and authenticity of the model, we retrieve the secret from the model, compute the hash value of the secret, and compare it with the embedded hash value. To minimize the effects of the embedded secret on the CNN model, we use a wavelet-based technique to transform weights into the frequency domain and embed the secret into less significant coefficients. Our theoretical analysis shows that DeepiSign can hide up to 1KB secret in each layer with minimal loss of the model's accuracy. To evaluate the security and performance of DeepiSign, we performed experiments on four pre-trained models (ResNet18, VGG16, AlexNet, and MobileNet) using three datasets (MNIST, CIFAR-10, and Imagenet) against three types of manipulation attacks (targeted input poisoning, output poisoning, and fine-tuning). The results demonstrate that DeepiSign is verifiable without degrading the classification accuracy, and robust against representative CNN manipulation attacks.
We have witnessed the continuing arms race between backdoor attacks and the corresponding defense strategies on Deep Neural Networks (DNNs). Most state-of-the-art defenses rely on the statistical sanitization of the "inputs" or "latent DNN representations" to capture trojan behaviour. In this paper, we first challenge the robustness of such recently reported defenses by introducing a novel variant of targeted backdoor attack, called "low-confidence backdoor attack". We also propose a novel defense technique, called "HaS-Nets". "Low-confidence backdoor attack" exploits the confidence labels assigned to poisoned training samples by giving low values to hide their presence from the defender, both during training and inference. We evaluate the attack against four state-of-the-art defense methods, viz., STRIP, Gradient-Shaping, Februus and ULP-defense, and achieve Attack Success Rate (ASR) of 99%, 63.73%, 91.2% and 80%, respectively. We next present "HaS-Nets" to resist backdoor insertion in the network during training, using a reasonably small healing dataset, approximately 2% to 15% of full training data, to heal the network at each iteration. We evaluate it for different datasets - Fashion-MNIST, CIFAR-10, Consumer Complaint and Urban Sound - and network architectures - MLPs, 2D-CNNs, 1D-CNNs. Our experiments show that "HaS-Nets" can decrease ASRs from over 90% to less than 15%, independent of the dataset, attack configuration and network architecture.
As an essential processing step in computer vision applications, image resizing or scaling, more specifically downsampling, has to be applied before feeding a normally large image into a convolutional neural network (CNN) model because CNN models typically take small fixed-size images as inputs. However, image scaling functions could be adversarially abused to perform a newly revealed attack called image-scaling attack, which can affect a wide range of computer vision applications building upon image-scaling functions. This work presents an image-scaling attack detection framework, termed as Decamouflage. Decamouflage consists of three independent detection methods: (1) rescaling, (2) filtering/pooling, and (3) steganalysis. While each of these three methods is efficient standalone, they can work in an ensemble manner not only to improve the detection accuracy but also to harden potential adaptive attacks. Decamouflage has a pre-determined detection threshold that is generic. More precisely, as we have validated, the threshold determined from one dataset is also applicable to other different datasets. Extensive experiments show that Decamouflage achieves detection accuracy of 99.9\% and 99.8\% in the white-box (with the knowledge of attack algorithms) and the black-box (without the knowledge of attack algorithms) settings, respectively. To corroborate the efficiency of Decamouflage, we have also measured its run-time overhead on a personal PC with an i5 CPU and found that Decamouflage can detect image-scaling attacks in milliseconds. Overall, Decamouflage can accurately detect image scaling attacks in both white-box and black-box settings with acceptable run-time overhead.
This work provides the community with a timely comprehensive review of backdoor attacks and countermeasures on deep learning. According to the attacker's capability and affected stage of the machine learning pipeline, the attack surfaces are recognized to be wide and then formalized into six categorizations: code poisoning, outsourcing, pretrained, data collection, collaborative learning and post-deployment. Accordingly, attacks under each categorization are combed. The countermeasures are categorized into four general classes: blind backdoor removal, offline backdoor inspection, online backdoor inspection, and post backdoor removal. Accordingly, we review countermeasures, and compare and analyze their advantages and disadvantages. We have also reviewed the flip side of backdoor attacks, which are explored for i) protecting intellectual property of deep learning models, ii) acting as a honeypot to catch adversarial example attacks, and iii) verifying data deletion requested by the data contributor.Overall, the research on defense is far behind the attack, and there is no single defense that can prevent all types of backdoor attacks. In some cases, an attacker can intelligently bypass existing defenses with an adaptive attack. Drawing the insights from the systematic review, we also present key areas for future research on the backdoor, such as empirical security evaluations from physical trigger attacks, and in particular, more efficient and practical countermeasures are solicited.
Artificial intelligence (AI) has been applied in phishing email detection. Typically, it requires rich email data from a collection of sources, and the data usually contains private information that needs to be preserved. So far, AI techniques are solely focusing on centralized data training that eventually accesses sensitive raw email data from the collected data repository. Thus, a privacy-friendly AI technique such as federated learning (FL) is a desideratum. FL enables learning over distributed email datasets to protect their privacy without the requirement of accessing them during the learning in a distributed computing framework. This work, to the best of our knowledge, is the first to investigate the applicability of training email anti-phishing model via FL. Building upon the Recurrent Convolutional Neural Network for phishing email detection, we comprehensively measure and evaluate the FL-entangled learning performance under various settings, including balanced and imbalanced data distribution among clients, scalability, communication overhead, and transfer learning. Our results positively corroborate comparable performance statistics of FL in phishing email detection to centralized learning. As a trade-off to privacy and distributed learning, FL has a communication overhead of 0.179 GB per global epoch per its clients. Our measurement-based results find that FL is suitable for practical scenarios, where data size variation, including the ratio of phishing to legitimate email samples, among the clients, are present. In all these scenarios, FL shows a similar performance of testing accuracy of around 98%. Besides, we demonstrate the integration of the newly joined clients with time in FL via transfer learning to improve the client-level performance. The transfer learning-enabled training results in the improvement of the testing accuracy by up to 2.6% and fast convergence.
Machine learning models have demonstrated vulnerability to adversarial attacks, more specifically misclassification of adversarial examples. In this paper, we investigate an attack-agnostic defense against adversarial attacks on high-resolution images by detecting suspicious inputs. The intuition behind our approach is that the essential characteristics of a normal image are generally consistent with non-essential style transformations, e.g., slightly changing the facial expression of human portraits. In contrast, adversarial examples are generally sensitive to such transformations. In our approach to detect adversarial instances, we propose an in\underline{V}ertible \underline{A}utoencoder based on the \underline{S}tyleGAN2 generator via \underline{A}dversarial training (VASA) to inverse images to disentangled latent codes that reveal hierarchical styles. We then build a set of edited copies with non-essential style transformations by performing latent shifting and reconstruction, based on the correspondences between latent codes and style transformations. The classification-based consistency of these edited copies is used to distinguish adversarial instances.
As communities represent similar opinions, similar functions, similar purposes, etc., community detection is an important and extremely useful tool in both scientific inquiry and data analytics. However, the classic methods of community detection, such as spectral clustering and statistical inference, are falling by the wayside as deep learning techniques demonstrate an increasing capacity to handle high-dimensional graph data with impressive performance. Thus, a survey of current progress in community detection through deep learning is timely. Structured into three broad research streams in this domain - deep neural networks, deep graph embedding, and graph neural networks, this article summarizes the contributions of the various frameworks, models, and algorithms in each stream along with the current challenges that remain unsolved and the future research opportunities yet to be explored.
This work is the first attempt to evaluate and compare felderated learning (FL) and split neural networks (SplitNN) in real-world IoT settings in terms of learning performance and device implementation overhead. We consider a variety of datasets, different model architectures, multiple clients, and various performance metrics. For learning performance, which is specified by the model accuracy and convergence speed metrics, we empirically evaluate both FL and SplitNN under different types of data distributions such as imbalanced and non-independent and identically distributed (non-IID) data. We show that the learning performance of SplitNN is better than FL under an imbalanced data distribution, but worse than FL under an extreme non-IID data distribution. For implementation overhead, we end-to-end mount both FL and SplitNN on Raspberry Pis, and comprehensively evaluate overheads including training time, communication overhead under the real LAN setting, power consumption and memory usage. Our key observations are that under IoT scenario where the communication traffic is the main concern, the FL appears to perform better over SplitNN because FL has the significantly lower communication overhead compared with SplitNN, which empirically corroborate previous statistical analysis. In addition, we reveal several unrecognized limitations about SplitNN, forming the basis for future research.