Visible to the public Biblio

Filters: Keyword is deep neural networks  [Clear All Filters]
2021-06-28
Li, Meng, Zhong, Qi, Zhang, Leo Yu, Du, Yajuan, Zhang, Jun, Xiang, Yong.  2020.  Protecting the Intellectual Property of Deep Neural Networks with Watermarking: The Frequency Domain Approach. 2020 IEEE 19th International Conference on Trust, Security and Privacy in Computing and Communications (TrustCom). :402–409.
Similar to other digital assets, deep neural network (DNN) models could suffer from piracy threat initiated by insider and/or outsider adversaries due to their inherent commercial value. DNN watermarking is a promising technique to mitigate this threat to intellectual property. This work focuses on black-box DNN watermarking, with which an owner can only verify his ownership by issuing special trigger queries to a remote suspicious model. However, informed attackers, who are aware of the watermark and somehow obtain the triggers, could forge fake triggers to claim their ownerships since the poor robustness of triggers and the lack of correlation between the model and the owner identity. This consideration calls for new watermarking methods that can achieve better trade-off for addressing the discrepancy. In this paper, we exploit frequency domain image watermarking to generate triggers and build our DNN watermarking algorithm accordingly. Since watermarking in the frequency domain is high concealment and robust to signal processing operation, the proposed algorithm is superior to existing schemes in resisting fraudulent claim attack. Besides, extensive experimental results on 3 datasets and 8 neural networks demonstrate that the proposed DNN watermarking algorithm achieves similar performance on functionality metrics and better performance on security metrics when compared with existing algorithms.
2021-05-26
Boursinos, Dimitrios, Koutsoukos, Xenofon.  2020.  Trusted Confidence Bounds for Learning Enabled Cyber-Physical Systems. 2020 IEEE Security and Privacy Workshops (SPW). :228—233.

Cyber-physical systems (CPS) can benefit by the use of learning enabled components (LECs) such as deep neural networks (DNNs) for perception and decision making tasks. However, DNNs are typically non-transparent making reasoning about their predictions very difficult, and hence their application to safety-critical systems is very challenging. LECs could be integrated easier into CPS if their predictions could be complemented with a confidence measure that quantifies how much we trust their output. The paper presents an approach for computing confidence bounds based on Inductive Conformal Prediction (ICP). We train a Triplet Network architecture to learn representations of the input data that can be used to estimate the similarity between test examples and examples in the training data set. Then, these representations are used to estimate the confidence of set predictions from a classifier that is based on the neural network architecture used in the triplet. The approach is evaluated using a robotic navigation benchmark and the results show that we can computed trusted confidence bounds efficiently in real-time.

2021-05-13
Venceslai, Valerio, Marchisio, Alberto, Alouani, Ihsen, Martina, Maurizio, Shafique, Muhammad.  2020.  NeuroAttack: Undermining Spiking Neural Networks Security through Externally Triggered Bit-Flips. 2020 International Joint Conference on Neural Networks (IJCNN). :1–8.

Due to their proven efficiency, machine-learning systems are deployed in a wide range of complex real-life problems. More specifically, Spiking Neural Networks (SNNs) emerged as a promising solution to the accuracy, resource-utilization, and energy-efficiency challenges in machine-learning systems. While these systems are going mainstream, they have inherent security and reliability issues. In this paper, we propose NeuroAttack, a cross-layer attack that threatens the SNNs integrity by exploiting low-level reliability issues through a high-level attack. Particularly, we trigger a fault-injection based sneaky hardware backdoor through a carefully crafted adversarial input noise. Our results on Deep Neural Networks (DNNs) and SNNs show a serious integrity threat to state-of-the art machine-learning techniques.

2021-02-01
Mangaokar, N., Pu, J., Bhattacharya, P., Reddy, C. K., Viswanath, B..  2020.  Jekyll: Attacking Medical Image Diagnostics using Deep Generative Models. 2020 IEEE European Symposium on Security and Privacy (EuroS P). :139–157.
Advances in deep neural networks (DNNs) have shown tremendous promise in the medical domain. However, the deep learning tools that are helping the domain, can also be used against it. Given the prevalence of fraud in the healthcare domain, it is important to consider the adversarial use of DNNs in manipulating sensitive data that is crucial to patient healthcare. In this work, we present the design and implementation of a DNN-based image translation attack on biomedical imagery. More specifically, we propose Jekyll, a neural style transfer framework that takes as input a biomedical image of a patient and translates it to a new image that indicates an attacker-chosen disease condition. The potential for fraudulent claims based on such generated `fake' medical images is significant, and we demonstrate successful attacks on both X-rays and retinal fundus image modalities. We show that these attacks manage to mislead both medical professionals and algorithmic detection schemes. Lastly, we also investigate defensive measures based on machine learning to detect images generated by Jekyll.
2021-01-28
Kariyappa, S., Qureshi, M. K..  2020.  Defending Against Model Stealing Attacks With Adaptive Misinformation. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). :767—775.

Deep Neural Networks (DNNs) are susceptible to model stealing attacks, which allows a data-limited adversary with no knowledge of the training dataset to clone the functionality of a target model, just by using black-box query access. Such attacks are typically carried out by querying the target model using inputs that are synthetically generated or sampled from a surrogate dataset to construct a labeled dataset. The adversary can use this labeled dataset to train a clone model, which achieves a classification accuracy comparable to that of the target model. We propose "Adaptive Misinformation" to defend against such model stealing attacks. We identify that all existing model stealing attacks invariably query the target model with Out-Of-Distribution (OOD) inputs. By selectively sending incorrect predictions for OOD queries, our defense substantially degrades the accuracy of the attacker's clone model (by up to 40%), while minimally impacting the accuracy (\textbackslashtextless; 0.5%) for benign users. Compared to existing defenses, our defense has a significantly better security vs accuracy trade-off and incurs minimal computational overhead.

2021-01-25
Chen, J., Lin, X., Shi, Z., Liu, Y..  2020.  Link Prediction Adversarial Attack Via Iterative Gradient Attack. IEEE Transactions on Computational Social Systems. 7:1081–1094.
Increasing deep neural networks are applied in solving graph evolved tasks, such as node classification and link prediction. However, the vulnerability of deep models can be revealed using carefully crafted adversarial examples generated by various adversarial attack methods. To explore this security problem, we define the link prediction adversarial attack problem and put forward a novel iterative gradient attack (IGA) strategy using the gradient information in the trained graph autoencoder (GAE) model. Not surprisingly, GAE can be fooled by an adversarial graph with a few links perturbed on the clean one. The results on comprehensive experiments of different real-world graphs indicate that most deep models and even the state-of-the-art link prediction algorithms cannot escape the adversarial attack, such as GAE. We can benefit the attack as an efficient privacy protection tool from the link prediction of unknown violations. On the other hand, the adversarial attack is a robust evaluation metric for current link prediction algorithms of their defensibility.
2021-01-22
Mani, G., Pasumarti, V., Bhargava, B., Vora, F. T., MacDonald, J., King, J., Kobes, J..  2020.  DeCrypto Pro: Deep Learning Based Cryptomining Malware Detection Using Performance Counters. 2020 IEEE International Conference on Autonomic Computing and Self-Organizing Systems (ACSOS). :109—118.
Autonomy in cybersystems depends on their ability to be self-aware by understanding the intent of services and applications that are running on those systems. In case of mission-critical cybersystems that are deployed in dynamic and unpredictable environments, the newly integrated unknown applications or services can either be benign and essential for the mission or they can be cyberattacks. In some cases, these cyberattacks are evasive Advanced Persistent Threats (APTs) where the attackers remain undetected for reconnaissance in order to ascertain system features for an attack e.g. Trojan Laziok. In other cases, the attackers can use the system only for computing e.g. cryptomining malware. APTs such as cryptomining malware neither disrupt normal system functionalities nor trigger any warning signs because they simply perform bitwise and cryptographic operations as any other benign compression or encoding application. Thus, it is difficult for defense mechanisms such as antivirus applications to detect these attacks. In this paper, we propose an Operating Context profiling system based on deep neural networks-Long Short-Term Memory (LSTM) networks-using Windows Performance Counters data for detecting these evasive cryptomining applications. In addition, we propose Deep Cryptomining Profiler (DeCrypto Pro), a detection system with a novel model selection framework containing a utility function that can select a classification model for behavior profiling from both the light-weight machine learning models (Random Forest and k-Nearest Neighbors) and a deep learning model (LSTM), depending on available computing resources. Given data from performance counters, we show that individual models perform with high accuracy and can be trained with limited training data. We also show that the DeCrypto Profiler framework reduces the use of computational resources and accurately detects cryptomining applications by selecting an appropriate model, given the constraints such as data sample size and system configuration.
2021-01-15
Khodabakhsh, A., Busch, C..  2020.  A Generalizable Deepfake Detector based on Neural Conditional Distribution Modelling. 2020 International Conference of the Biometrics Special Interest Group (BIOSIG). :1—5.
Photo- and video-realistic generation techniques have become a reality following the advent of deep neural networks. Consequently, there are immense concerns regarding the difficulty in differentiating what content is real from what is synthetic. An example of video-realistic generation techniques is the infamous Deepfakes, which exploit the main modality by which humans identify each other. Deepfakes are a category of synthetic face generation methods and are commonly based on generative adversarial networks. In this article, we propose a novel two-step synthetic face image detection method in which general-purpose features are extracted in a first step, trivializing the task of detecting synthetic images. The anomaly detector predicts the conditional probabilities for observing every individual pixel in the image and is trained on pristine data only. The extracted anomaly features demonstrate true generalization capacity across widely different unknown synthesis methods while showing a minimal loss in performance with regard to the detection of known synthetic samples.
2021-01-11
Mihanpour, A., Rashti, M. J., Alavi, S. E..  2020.  Human Action Recognition in Video Using DB-LSTM and ResNet. 2020 6th International Conference on Web Research (ICWR). :133—138.

Human action recognition in video is one of the most widely applied topics in the field of image and video processing, with many applications in surveillance (security, sports, etc.), activity detection, video-content-based monitoring, man-machine interaction, and health/disability care. Action recognition is a complex process that faces several challenges such as occlusion, camera movement, viewpoint move, background clutter, and brightness variation. In this study, we propose a novel human action recognition method using convolutional neural networks (CNN) and deep bidirectional LSTM (DB-LSTM) networks, using only raw video frames. First, deep features are extracted from video frames using a pre-trained CNN architecture called ResNet152. The sequential information of the frames is then learned using the DB-LSTM network, where multiple layers are stacked together in both forward and backward passes of DB-LSTM, to increase depth. The evaluation results of the proposed method using PyTorch, compared to the state-of-the-art methods, show a considerable increase in the efficiency of action recognition on the UCF 101 dataset, reaching 95% recognition accuracy. The choice of the CNN architecture, proper tuning of input parameters, and techniques such as data augmentation contribute to the accuracy boost in this study.

Gautam, A., Singh, S..  2020.  A Comparative Analysis of Deep Learning based Super-Resolution Techniques for Thermal Videos. 2020 Third International Conference on Smart Systems and Inventive Technology (ICSSIT). :919—925.

Video streams acquired from thermal cameras are proven to be beneficial in diverse number of fields including military, healthcare, law enforcement, and security. Despite the hype, thermal imaging is increasingly affected by poor resolution, where it has expensive optical sensors and inability to attain optical precision. In recent years, deep learning based super-resolution algorithms are developed to enhance the video frame resolution at high accuracy. This paper presents a comparative analysis of super resolution (SR) techniques based on deep neural networks (DNN) that are applied on thermal video dataset. SRCNN, EDSR, Auto-encoder, and SRGAN are also discussed and investigated. Further the results on benchmark thermal datasets including FLIR, OSU thermal pedestrian database and OSU color thermal database are evaluated and analyzed. Based on the experimental results, it is concluded that, SRGAN has delivered a superior performance on thermal frames when compared to other techniques and improvements, which has the ability to provide state-of-the art performance in real time operations.

Rajapkar, A., Binnar, P., Kazi, F..  2020.  Design of Intrusion Prevention System for OT Networks Using Deep Neural Networks. 2020 11th International Conference on Computing, Communication and Networking Technologies (ICCCNT). :1–6.

The Automation industries that uses Supervisory Control and Data Acquisition (SCADA) systems are highly vulnerable for Network threats. Systems that are air-gapped and isolated from the internet are highly affected due to insider attacks like Spoofing, DOS and Malware threats that affects confidentiality, integrity and availability of Operational Technology (OT) system elements and degrade its performance even though security measures are taken. In this paper, a behavior-based intrusion prevention system (IPS) is designed for OT networks. The proposed system is implemented on SCADA test bed with two systems replicates automation scenarios in industry. This paper describes 4 main classes of cyber-attacks with their subclasses against SCADA systems and methodology with design of components of IPS system, database creation, Baselines and deployment of system in environment. IPS system identifies not only IT protocols but also Industry Control System (ICS) protocols Modbus and DNP3 with their inside communication fields using deep packet inspection (DPI). The analytical results show 99.89% accuracy on binary classification and 97.95% accuracy on multiclass classification of different attack vectors performed on network with low false positive rate. These results are also validated by actual deployment of IPS in SCADA systems with the prevention of DOS attack.

2020-12-11
Mikołajczyk, A., Grochowski, M..  2019.  Style transfer-based image synthesis as an efficient regularization technique in deep learning. 2019 24th International Conference on Methods and Models in Automation and Robotics (MMAR). :42—47.

These days deep learning is the fastest-growing area in the field of Machine Learning. Convolutional Neural Networks are currently the main tool used for the image analysis and classification purposes. Although great achievements and perspectives, deep neural networks and accompanying learning algorithms have some relevant challenges to tackle. In this paper, we have focused on the most frequently mentioned problem in the field of machine learning, that is relatively poor generalization abilities. Partial remedies for this are regularization techniques e.g. dropout, batch normalization, weight decay, transfer learning, early stopping and data augmentation. In this paper we have focused on data augmentation. We propose to use a method based on a neural style transfer, which allows to generate new unlabeled images of high perceptual quality that combine the content of a base image with the appearance of another one. In a proposed approach, the newly created images are described with pseudo-labels, and then used as a training dataset. Real, labeled images are divided into the validation and test set. We validated proposed method on a challenging skin lesion classification case study. Four representative neural architectures are examined. Obtained results show the strong potential of the proposed approach.

Cao, Y., Tang, Y..  2019.  Development of Real-Time Style Transfer for Video System. 2019 3rd International Conference on Circuits, System and Simulation (ICCSS). :183—187.

Re-drawing the image as a certain artistic style is considered to be a complicated task for computer machine. On the contrary, human can easily master the method to compose and describe the style between different images. In the past, many researchers studying on the deep neural networks had found an appropriate representation of the artistic style using perceptual loss and style reconstruction loss. In the previous works, Gatys et al. proposed an artificial system based on convolutional neural networks that creates artistic images of high perceptual quality. Whereas in terms of running speed, it was relatively time-consuming, thus it cannot apply to video style transfer. Recently, a feed-forward CNN approach has shown the potential of fast style transformation, which is an end-to-end system without hundreds of iteration while transferring. We combined the benefits of both approaches, optimized the feed-forward network and defined time loss function to make it possible to implement the style transfer on video in real time. In contrast to the past method, our method runs in real time with higher resolution while creating competitive visually pleasing and temporally consistent experimental results.

2020-12-07
Handa, A., Garg, P., Khare, V..  2018.  Masked Neural Style Transfer using Convolutional Neural Networks. 2018 International Conference on Recent Innovations in Electrical, Electronics Communication Engineering (ICRIEECE). :2099–2104.

In painting, humans can draw an interrelation between the style and the content of a given image in order to enhance visual experiences. Deep neural networks like convolutional neural networks are being used to draw a satisfying conclusion of this problem of neural style transfer due to their exceptional results in the key areas of visual perceptions such as object detection and face recognition.In this study, along with style transfer on whole image it is also outlined how transfer of style can be performed only on the specific parts of the content image which is accomplished by using masks. The style is transferred in a way that there is a least amount of loss to the content image i.e., semantics of the image is preserved.

2020-11-04
Khalid, F., Hanif, M. A., Rehman, S., Ahmed, R., Shafique, M..  2019.  TrISec: Training Data-Unaware Imperceptible Security Attacks on Deep Neural Networks. 2019 IEEE 25th International Symposium on On-Line Testing and Robust System Design (IOLTS). :188—193.

Most of the data manipulation attacks on deep neural networks (DNNs) during the training stage introduce a perceptible noise that can be catered by preprocessing during inference, or can be identified during the validation phase. There-fore, data poisoning attacks during inference (e.g., adversarial attacks) are becoming more popular. However, many of them do not consider the imperceptibility factor in their optimization algorithms, and can be detected by correlation and structural similarity analysis, or noticeable (e.g., by humans) in multi-level security system. Moreover, majority of the inference attack rely on some knowledge about the training dataset. In this paper, we propose a novel methodology which automatically generates imperceptible attack images by using the back-propagation algorithm on pre-trained DNNs, without requiring any information about the training dataset (i.e., completely training data-unaware). We present a case study on traffic sign detection using the VGGNet trained on the German Traffic Sign Recognition Benchmarks dataset in an autonomous driving use case. Our results demonstrate that the generated attack images successfully perform misclassification while remaining imperceptible in both “subjective” and “objective” quality tests.

Chacon, H., Silva, S., Rad, P..  2019.  Deep Learning Poison Data Attack Detection. 2019 IEEE 31st International Conference on Tools with Artificial Intelligence (ICTAI). :971—978.

Deep neural networks are widely used in many walks of life. Techniques such as transfer learning enable neural networks pre-trained on certain tasks to be retrained for a new duty, often with much less data. Users have access to both pre-trained model parameters and model definitions along with testing data but have either limited access to training data or just a subset of it. This is risky for system-critical applications, where adversarial information can be maliciously included during the training phase to attack the system. Determining the existence and level of attack in a model is challenging. In this paper, we present evidence on how adversarially attacking training data increases the boundary of model parameters using as an example of a CNN model and the MNIST data set as a test. This expansion is due to new characteristics of the poisonous data that are added to the training data. Approaching the problem from the feature space learned by the network provides a relation between them and the possible parameters taken by the model on the training phase. An algorithm is proposed to determine if a given network was attacked in the training by comparing the boundaries of parameters distribution on intermediate layers of the model estimated by using the Maximum Entropy Principle and the Variational inference approach.

Shen, J., Zhu, X., Ma, D..  2019.  TensorClog: An Imperceptible Poisoning Attack on Deep Neural Network Applications. IEEE Access. 7:41498—41506.

Internet application providers now have more incentive than ever to collect user data, which greatly increases the risk of user privacy violations due to the emerging of deep neural networks. In this paper, we propose TensorClog-a poisoning attack technique that is designed for privacy protection against deep neural networks. TensorClog has three properties with each of them serving a privacy protection purpose: 1) training on TensorClog poisoned data results in lower inference accuracy, reducing the incentive of abusive data collection; 2) training on TensorClog poisoned data converges to a larger loss, which prevents the neural network from learning the privacy; and 3) TensorClog regularizes the perturbation to remain a high structure similarity, so that the poisoning does not affect the actual content in the data. Applying our TensorClog poisoning technique to CIFAR-10 dataset results in an increase in both converged training loss and test error by 300% and 272%, respectively. It manages to maintain data's human perception with a high SSIM index of 0.9905. More experiments including different limited information attack scenarios and a real-world application transferred from pre-trained ImageNet models are presented to further evaluate TensorClog's effectiveness in more complex situations.

2020-10-29
Roseline, S. Abijah, Sasisri, A. D., Geetha, S., Balasubramanian, C..  2019.  Towards Efficient Malware Detection and Classification using Multilayered Random Forest Ensemble Technique. 2019 International Carnahan Conference on Security Technology (ICCST). :1—6.

The exponential growth rate of malware causes significant security concern in this digital era to computer users, private and government organizations. Traditional malware detection methods employ static and dynamic analysis, which are ineffective in identifying unknown malware. Malware authors develop new malware by using polymorphic and evasion techniques on existing malware and escape detection. Newly arriving malware are variants of existing malware and their patterns can be analyzed using the vision-based method. Malware patterns are visualized as images and their features are characterized. The alternative generation of class vectors and feature vectors using ensemble forests in multiple sequential layers is performed for classifying malware. This paper proposes a hybrid stacked multilayered ensembling approach which is robust and efficient than deep learning models. The proposed model outperforms the machine learning and deep learning models with an accuracy of 98.91%. The proposed system works well for small-scale and large-scale data since its adaptive nature of setting parameters (number of sequential levels) automatically. It is computationally efficient in terms of resources and time. The method uses very fewer hyper-parameters compared to deep neural networks.

2020-10-05
Liu, Donglei, Niu, Zhendong, Zhang, Chunxia, Zhang, Jiadi.  2019.  Multi-Scale Deformable CNN for Answer Selection. IEEE Access. 7:164986—164995.

The answer selection task is one of the most important issues within the automatic question answering system, and it aims to automatically find accurate answers to questions. Traditional methods for this task use manually generated features based on tf-idf and n-gram models to represent texts, and then select the right answers according to the similarity between the representations of questions and the candidate answers. Nowadays, many question answering systems adopt deep neural networks such as convolutional neural network (CNN) to generate the text features automatically, and obtained better performance than traditional methods. CNN can extract consecutive n-gram features with fixed length by sliding fixed-length convolutional kernels over the whole word sequence. However, due to the complex semantic compositionality of the natural language, there are many phrases with variable lengths and be composed of non-consecutive words in natural language, such as these phrases whose constituents are separated by other words within the same sentences. But the traditional CNN is unable to extract the variable length n-gram features and non-consecutive n-gram features. In this paper, we propose a multi-scale deformable convolutional neural network to capture the non-consecutive n-gram features by adding offset to the convolutional kernel, and also propose to stack multiple deformable convolutional layers to mine multi-scale n-gram features by the means of generating longer n-gram in higher layer. Furthermore, we apply the proposed model into the task of answer selection. Experimental results on public dataset demonstrate the effectiveness of our proposed model in answer selection.

Li, Xilai, Song, Xi, Wu, Tianfu.  2019.  AOGNets: Compositional Grammatical Architectures for Deep Learning. 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). :6213—6223.

Neural architectures are the foundation for improving performance of deep neural networks (DNNs). This paper presents deep compositional grammatical architectures which harness the best of two worlds: grammar models and DNNs. The proposed architectures integrate compositionality and reconfigurability of the former and the capability of learning rich features of the latter in a principled way. We utilize AND-OR Grammar (AOG) as network generator in this paper and call the resulting networks AOGNets. An AOGNet consists of a number of stages each of which is composed of a number of AOG building blocks. An AOG building block splits its input feature map into N groups along feature channels and then treat it as a sentence of N words. It then jointly realizes a phrase structure grammar and a dependency grammar in bottom-up parsing the “sentence” for better feature exploration and reuse. It provides a unified framework for the best practices developed in state-of-the-art DNNs. In experiments, AOGNet is tested in the ImageNet-1K classification benchmark and the MS-COCO object detection and segmentation benchmark. In ImageNet-1K, AOGNet obtains better performance than ResNet and most of its variants, ResNeXt and its attention based variants such as SENet, DenseNet and DualPathNet. AOGNet also obtains the best model interpretability score using network dissection. AOGNet further shows better potential in adversarial defense. In MS-COCO, AOGNet obtains better performance than the ResNet and ResNeXt backbones in Mask R-CNN.

2020-09-28
Liu, Kai, Zhou, Yun, Wang, Qingyong, Zhu, Xianqiang.  2019.  Vulnerability Severity Prediction With Deep Neural Network. 2019 5th International Conference on Big Data and Information Analytics (BigDIA). :114–119.
High frequency of network security incidents has also brought a lot of negative effects and even huge economic losses to countries, enterprises and individuals in recent years. Therefore, more and more attention has been paid to the problem of network security. In order to evaluate the newly included vulnerability text information accurately, and to reduce the workload of experts and the false negative rate of the traditional method. Multiple deep learning methods for vulnerability text classification evaluation are proposed in this paper. The standard Cross Site Scripting (XSS) vulnerability text data is processed first, and then classified using three kinds of deep neural networks (CNN, LSTM, TextRCNN) and one kind of traditional machine learning method (XGBoost). The dropout ratio of the optimal CNN network, the epoch of all deep neural networks and training set data were tuned via experiments to improve the fit on our target task. The results show that the deep learning methods evaluate vulnerability risk levels better, compared with traditional machine learning methods, but cost more time. We train our models in various training sets and test with the same testing set. The performance and utility of recurrent convolutional neural networks (TextRCNN) is highest in comparison to all other methods, which classification accuracy rate is 93.95%.
2020-09-21
Chow, Ka-Ho, Wei, Wenqi, Wu, Yanzhao, Liu, Ling.  2019.  Denoising and Verification Cross-Layer Ensemble Against Black-box Adversarial Attacks. 2019 IEEE International Conference on Big Data (Big Data). :1282–1291.
Deep neural networks (DNNs) have demonstrated impressive performance on many challenging machine learning tasks. However, DNNs are vulnerable to adversarial inputs generated by adding maliciously crafted perturbations to the benign inputs. As a growing number of attacks have been reported to generate adversarial inputs of varying sophistication, the defense-attack arms race has been accelerated. In this paper, we present MODEF, a cross-layer model diversity ensemble framework. MODEF intelligently combines unsupervised model denoising ensemble with supervised model verification ensemble by quantifying model diversity, aiming to boost the robustness of the target model against adversarial examples. Evaluated using eleven representative attacks on popular benchmark datasets, we show that MODEF achieves remarkable defense success rates, compared with existing defense methods, and provides a superior capability of repairing adversarial inputs and making correct predictions with high accuracy in the presence of black-box attacks.
2020-09-04
Song, Chengru, Xu, Changqiao, Yang, Shujie, Zhou, Zan, Gong, Changhui.  2019.  A Black-Box Approach to Generate Adversarial Examples Against Deep Neural Networks for High Dimensional Input. 2019 IEEE Fourth International Conference on Data Science in Cyberspace (DSC). :473—479.
Generating adversarial samples is gathering much attention as an intuitive approach to evaluate the robustness of learning models. Extensive recent works have demonstrated that numerous advanced image classifiers are defenseless to adversarial perturbations in the white-box setting. However, the white-box setting assumes attackers to have prior knowledge of model parameters, which are generally inaccessible in real world cases. In this paper, we concentrate on the hard-label black-box setting where attackers can only pose queries to probe the model parameters responsible for classifying different images. Therefore, the issue is converted into minimizing non-continuous function. A black-box approach is proposed to address both massive queries and the non-continuous step function problem by applying a combination of a linear fine-grained search, Fibonacci search, and a zeroth order optimization algorithm. However, the input dimension of a image is so high that the estimation of gradient is noisy. Hence, we adopt a zeroth-order optimization method in high dimensions. The approach converts calculation of gradient into a linear regression model and extracts dimensions that are more significant. Experimental results illustrate that our approach can relatively reduce the amount of queries and effectively accelerate convergence of the optimization method.
Wu, Yi, Liu, Jian, Chen, Yingying, Cheng, Jerry.  2019.  Semi-black-box Attacks Against Speech Recognition Systems Using Adversarial Samples. 2019 IEEE International Symposium on Dynamic Spectrum Access Networks (DySPAN). :1—5.
As automatic speech recognition (ASR) systems have been integrated into a diverse set of devices around us in recent years, security vulnerabilities of them have become an increasing concern for the public. Existing studies have demonstrated that deep neural networks (DNNs), acting as the computation core of ASR systems, is vulnerable to deliberately designed adversarial attacks. Based on the gradient descent algorithm, existing studies have successfully generated adversarial samples which can disturb ASR systems and produce adversary-expected transcript texts designed by adversaries. Most of these research simulated white-box attacks which require knowledge of all the components in the targeted ASR systems. In this work, we propose the first semi-black-box attack against the ASR system - Kaldi. Requiring only partial information from Kaldi and none from DNN, we can embed malicious commands into a single audio chip based on the gradient-independent genetic algorithm. The crafted audio clip could be recognized as the embedded malicious commands by Kaldi and unnoticeable to humans in the meanwhile. Experiments show that our attack can achieve high attack success rate with unnoticeable perturbations to three types of audio clips (pop music, pure music, and human command) without the need of the underlying DNN model parameters and architecture.
Taori, Rohan, Kamsetty, Amog, Chu, Brenton, Vemuri, Nikita.  2019.  Targeted Adversarial Examples for Black Box Audio Systems. 2019 IEEE Security and Privacy Workshops (SPW). :15—20.
The application of deep recurrent networks to audio transcription has led to impressive gains in automatic speech recognition (ASR) systems. Many have demonstrated that small adversarial perturbations can fool deep neural networks into incorrectly predicting a specified target with high confidence. Current work on fooling ASR systems have focused on white-box attacks, in which the model architecture and parameters are known. In this paper, we adopt a black-box approach to adversarial generation, combining the approaches of both genetic algorithms and gradient estimation to solve the task. We achieve a 89.25% targeted attack similarity, with 35% targeted attack success rate, after 3000 generations while maintaining 94.6% audio file similarity.