Neural language models have demonstrated impressive performance in various tasks but remain vulnerable to word-level adversarial attacks. Word-level adversarial attacks can be formulated as a combinatorial optimization problem, and thus, an attack method can be decomposed into search space and search method. Despite the significance of these two components, previous works inadequately distinguish them, which may lead to unfair comparisons and insufficient evaluations. In this paper, to address the inappropriate practices in previous works, we perform thorough ablation studies on the search space, illustrating the substantial influence of search space on attack efficiency, effectiveness, and imperceptibility. Based on the ablation study, we propose two standardized search spaces: the Search Space for ImPerceptibility (SSIP) and Search Space for EffecTiveness (SSET). The reevaluation of eight previous attack methods demonstrates the success of SSIP and SSET in achieving better trade-offs between efficiency, effectiveness, and imperceptibility in different scenarios, offering fair and comprehensive evaluations of previous attack methods and providing potential guidance for future works.

Data leakage is one of the most severe security threats that can compromise sensitive data through breaches or unauthorized access. Existing techniques usually adopt encryption or access control protection methods, but inevitably affects the data usability and introduce significant overhead. In this paper, we propose a novel technique for data leakage prevention in collaborative systems by dynamically broadening the deceptive attack surface. Our proposed technique offers an adaptive deception strategy that leverages historical user behaviors and current operations to generate deceptive data, and we developed an amplifying-based calculation method to enhance the accuracy of user trust degree evaluation. Furthermore, we introduce three deceptive indicators to evaluate our technique. Experimental results show that our technique can effectively prevent data leakage while preserving data usability and imposing minimal overhead to the system.

Automatic detecting rumors on social media has become a challenging task. Previous studies focus on learning indicative clues from conversation threads for identifying rumorous information. However, these methods only model rumorous conversation threads from various views but fail to fuse multi-view features very well. In this paper, we propose a novel multi-view fusion framework for rumor representation learning and classification. It encodes the multiple views based on Graph Convolutional Networks (GCN) and leverages Convolutional Neural Networks (CNN) to capture the consistent and complementary information among all views and fuse them. Experimental results on two public datasets demonstrate that our method outperforms state-of-the-art approaches.

Federated learning decentralizes the learning process, yet it does not provide adequate privacy protection. Current countermeasures predominantly rely on Local Differential Privacy (LDP) techniques. While larger noise injection offers stronger privacy, it also leads to a degradation in model performance and necessitates additional training iterations. Existing methods struggle to strike a balance between user privacy, model usability, and training efficiency. To address this, we propose an adaptive noise scaling method grounded in mutual information. This method is designed to dynamically optimize the noise scale, thereby safeguarding user privacy, enhancing training efficiency, and ensuring model usability. Specifically, we first estimate the mutual information between user training data and gradient updates in each iteration using Mutual Information Neural Estimation (MINE). Subsequently, we propose to dynamically adjust the optimal noise scale in each round of federated learning’s local training, based on the estimated mutual information. Experimental results demonstrate that using mutual information to dynamically adjust the noise scale reduces the number of training iterations by over 50%, while maintaining the same level of user privacy and data model availability.

Mobile Edge Computing (MEC) has emerged as a distributed computing paradigm offering low-latency services to users. However, the distributed and intricate deployment inevitably makes it challenge to ensure the reliability of MEC services. Anomaly detection using the streaming data of MEC services is an essential way to address the challenge. In this paper, to improve the detection accuracy and efficiency, we propose an MTDF-Detection framework for MEC services, jointly extracting Multi-Timescale and Dimensional Feature (MTDF), including the long-term trend, periodic, short-term fluctuation and auxiliary parameters (e.g., system maintenance and offloading task). In this framework, to reduce the computing costs, we adopt a Bidirectional Simple Recurrent Unit (Bi-SRU) to obtain contextual feature; and we design an adaptive m-Sigma algorithm to determine the dynamic threshold with real-time streaming data. Extensive experiments are conducted on a real-world dataset, and the results demonstrate that the MTDF-Detection framework outperforms the state-of-the-art schemes in terms of accuracy and efficiency.

Federated learning (FL) exhibits vulnerabilities to poisoning attacks, where Byzantine FL clients send malicious model updates to hamper the accuracy of the global model. However, these efforts are being circumvented by some more advanced stealthy poisoning attacks. In this paper, we propose an Enclave-aided Byzantine-robust Federated Aggregation (EBFA) framework. In particular, at each FL epoch, we first evaluate the layer-wise cosine similarity between the guide model (learned from an extra validation dataset) and local models, and then, utilize the boxplot method to construct a region of outliers to find Byzantine clients. To avoid the interference to the robust federated aggregation caused by classical privacy-preserving method, such as differential privacy and homomorphic encryption, we further design an efficient privacy-preserving scheme for robust aggregation via Trusted Execution Environment (TEE); and, to improve the efficiency, we only deploy the privacy-sensitive aggregation operations within resource limited TEE (or enclave). Finally, we perform extensive experiments on different datasets, and demonstrate that our proposed EBFA outperforms the state-of-the-art Byzantine-robust schemes (e.g., FLTrust) under non-IID settings. Moreover, our proposed enclave-aided privacy-preserving scheme could significantly improve the efficiency (over 40% for Alexnet) in comparison with the TEE-only scheme.

Microservice applications are gaining popularity as cloud-native embraced by the IT industry. However, they suffer from a latency problem because they intrinsically involve mass inter-microservice data exchange that introduces additional latency. The remarkable gap of data transmission speed between co-location and cross-server communication necessitates the need of utilizing microservice deployment strategies to accelerate data transmission and reduce end-to-end latency. Previous works characterize communication dependencies among microservices and put strongly inter-dependent ones on the same server to reduce communication overhead. Unfortunately, they overlook characteristics of microservice applications, and the resulting modeling is neither fine-grained enough nor comprehensive, which misleads microservice deployment and ultimately results in prolonged communication delay. To address the above problems, we propose a novel microser-vice deployment strategy based on fine-grained and comprehensive modeling of microservice dependencies, on the basis of an in-depth analysis of service characteristics. In this paper, dependencies are measured on the request level and combined with data sharing relationships to jointly decide microservice division, aiming to facilitate inter-microservice communication. After that, a selective scale-up strategy is designed to further reduce cross-server communication and shorten response latency. Extensive experiments on a real-world microservice application demonstrate that our method is effective in mitigating commu-nication overhead and can reduce end-to-end latency by 15.64% ~ 29.18 % compared with the state-of-the-art methods.

Abstract As a distributed learning paradigm, federated learning is supposed to protect data privacy without exchanging users’ local data. Even so, the gradient inversion attack, in which the adversary can reconstruct the original data from shared training gradients, has been widely deemed as a severe threat. Nevertheless, most existing researches are confined to impractical assumptions and narrow range of applications. To mitigate these shortcomings, we propose a comprehensive framework for gradient inversion attack, with well-designed algorithms for image and label reconstruction. For image reconstruction, we fully utilize the generative image prior, which derives from wide-used generative models, to improve the reconstructed results, by additional means of iterative optimization on mixed spaces and gradient-free optimizer. For label reconstruction, we design an adaptive recovery algorithm regarding real data distribution, which can adjust previous attacks to more complex scenarios. Moreover, we incorporate a gradient approximation method to efficiently fit our attack for FedAvg scenario. We empirically verify our attack framework using benchmark datasets and ablation studies, considering loose assumptions and complicated circumstances. We hope this work can greatly reveal the necessity of privacy protection in federated learning, while urge more effective and robust defense mechanisms.

Abstract Unprotected gradient exchange in federated learning (FL) systems may lead to gradient leakage-related attacks. CKKS is a promising approximate homomorphic encryption scheme to protect gradients, owing to its unique capability of performing operations directly on ciphertexts. However, configuring CKKS security parameters involves a trade-off between correctness, efficiency, and security. An evaluation gap exists regarding how these parameters impact computational performance. Additionally, the maximum vector length that CKKS can once encrypt, recommended by Homomorphic Encryption Standardization, is 16384, hampers its widespread adoption in FL when encrypting layers with numerous neurons. To protect gradients’ privacy in FL systems while maintaining practical performance, we comprehensively analyze the influence of security parameters such as polynomial modulus degree and coefficient modulus on homomorphic operations. Derived from our evaluation findings, we provide a method for selecting the optimal multiplication depth while meeting operational requirements. Then, we introduce an adaptive segmented encryption method tailored for CKKS, circumventing its encryption length constraint and enhancing its processing ability to encrypt neural network models. Finally, we present FedSHE , a privacy-preserving and efficient Fed erated learning scheme with adaptive S egmented CKKS H omomorphic E ncryption. FedSHE is implemented on top of the federated averaging (FedAvg) algorithm and is available at https://github.com/yooopan/FedSHE . Our evaluation results affirm the correctness and effectiveness of our proposed method, demonstrating that FedSHE outperforms existing homomorphic encryption-based federated learning research efforts in terms of model accuracy, computational efficiency, communication cost, and security level.

The recent advances of deep learning in various mobile and Internet-of-Things applications, coupled with the emergence of edge computing, have led to a strong trend of performing deep learning inference on the edge servers located physically close to the end devices. This trend presents the challenge of how to meet the quality-of-service requirements of inference tasks at the resource-constrained network edge, especially under variable or even bursty inference workloads. Solutions to this challenge have not yet been reported in the related literature. In the present paper, we tackle this challenge by means of workload-adaptive inference request scheduling: in different workload states, via adaptive inference request scheduling policies, different models with diverse model sizes can play different roles to maintain high-quality inference services. To implement this idea, we propose a request scheduling framework for general-purpose edge inference serving systems. Theoretically, we prove that, in our framework, the problem of optimizing the inference request scheduling policies can be formulated as a Markov decision process (MDP). To tackle such an MDP, we use reinforcement learning and propose a policy optimization approach. Through extensive experiments, we empirically demonstrate the effectiveness of our framework in the challenging practical case where the MDP is partially observable.

Neural language models have achieved superior performance.However, these models also suffer from the pathology of overconfidence in the out-of-distribution examples, potentially making the model difficult to interpret and making the interpretation methods fail to provide faithful attributions.In this paper, we explain the model pathology from the view of sentence representation and argue that the counter-intuitive bias degree and direction of the out-of-distribution examples' representation cause the pathology.We propose a Contrastive learning regularization method using Adversarial examples for Alleviating the Pathology (ConAAP), which calibrates the sentence representation of out-of-distribution examples.ConAAP generates positive and negative examples following the attribution results and utilizes adversarial examples to introduce direction information in regularization.Experiments show that ConAAP effectively alleviates the model pathology while slightly impacting the generalization ability on in-distribution examples and thus helps interpretation methods obtain more faithful results.

In the data-driven economy, data sharing is a key approach to unleashing the true value of data. Blockchain, as a decentralized ledger, can provide a trusted data sharing platform in an untrusted environment. However, existing blockchain-based data sharing schemes suffer from inefficiency and inadequate protection of security and privacy. To address the above issues, we propose a blockchain-based privacy-preserving data sharing scheme with security-enhanced access control. In the scheme, a secure data sharing architecture using dual-blockchain and the interplanetary file system (IPFS) is presented to provide decentralized and scalable storage. Based on the architecture, a blockchain-assisted multi-authority attribute-based encryption (BA-MA-ABE) algorithm with efficient attribute revocation and computation is designed in our work. Our BA-MA-ABE lever-ages blockchain to securely manage partial decryption keys and provides fine-grained access control over encrypted data. We also devise smart contracts that can support traceable access control over the flow of data while protecting user identity privacy with verifiable attribute credentials. In comparison with some existing work, our scheme shows more comprehensive security features with lower user computation overhead.

Federated Learning (FL) is a novel machine learning paradigm that enables multiple participants to collaboratively train a machine learning model by aggregating local gradients from each client without sharing sensitive train data with each other. The clients’ gradients plaintext transmission makes FL systems vulnerable to inference attacks, which aim to infer clients’ data from their model updates. Masking local gradients with additive homomorphic encryption (especially the Paillier scheme) before gradients aggregate is a straightforward way to ensure security. Unfortunately, this method needs traversing and encrypting gradients element-wisely, resulting in low computation efficiency and high communication cost. In this paper, we present BatchAgg, an efficient aggregation protocol for FL utilizing ciphertext packing technique provided by approximate homomorphic encryption scheme. Instead of encrypting gradients individually, BatchAgg encrypts a gradient vector into one ciphertext and homomorphically computes batch operations. Specifically, BatchAgg is built on the federated averaging protocol. In detail, we implement a federated image classification model for datasets horizontal split as a baseline and replace the Paillier-based aggregation protocol with BatchAgg to accelerate model training with lower communication cost. Evaluation results show that BatchAgg achieves 60× training speedup while reducing the communication cost by 57% compared with Paillier under the same security level. Moreover, BatchAgg can be easily integrated into existing models while assuring security throughout federated training without causing performance loss.

Microservice architecture is embraced by a growing number of enterprises due to the benefits of modularity and flexibility. However, being composed of numerous interdependent microservices, it is prone to cascading failures and afflicted by the arising problem of troubleshooting, which entails arduous efforts to identify the root cause node and ensure service availability. Previous works use call graph to characterize causality relation-ships of microservices but not completely or comprehensively, leading to an insufficient search of potential root cause nodes and consequently poor accuracy in culprit localization. In this paper, we propose ImpactTracer to address the above problems. ImpactTracer builds impact graph to provide a com-plete view of fault propagation in microservices and uses a novel backward tracing algorithm that exhaustively traverses the impact graph to identify the root cause node accurately. Extensive experiments on a real-world dataset demonstrate that ImpactTracer is effective in identifying the root cause node and outperforms the state-of-the-art methods by at least 72%, significantly facilitating troubleshooting in microservices.

Malware classification methods based on memory image representation have received increasing attention. However, the characteristics of the memory management mechanism and efficiency of the classification model are not well considered in previous works, which hinders the classifier from extracting high-quality features and consequently results in poor performance. Motivated by this, we propose a novel malware classification method. First, we add an Efficient Convolutional Block Attention Module (E-CBAM) to select important features with fewer parameters and less computational cost. Then, we integrate our attention module into a pre-trained EfficientNet-B0 to extract features efficiently. Moreover, data augmentation and label smoothing are adopted to mitigate model overfitting. Finally, extensive experiments on a realistic dataset testify to the performance and superiority of our method in both known and unknown malware classification.

Traditional defense methods are hard to change the inherent vulnerabilities of static data storage, single data access, and deterministic data content, leading to frequent data leakage incidents. Moving target defense (MTD) techniques can increase data diversity and unpredictability by dynamically shifting the data attack surface. However, in the existing methods, the data lacks sufficient dynamics due to insufficient shifting space and shifting frequency of attack surface, and legitimate users are inevitably greatly affected. This study proposes a data MTD method that the data changes dynamically based on real-time multi-source user access information. Through the multidimensional user stratification mechanism, we establish a novel dynamic data model that uses the combination of random deception strategies to convert metadata properties and content of data based on the user risk levels, while data remains unchanged for legitimate users. Multiple sets of experiments demonstrate the effectiveness and low consumption of our data dynamic defense approach.

We present NadGPT, a transformer-based semi-supervised framework for network anomaly detection. It is known that transformer models are good at modeling long sequence data such as network traffic; however, without sufficient ground-truth labels, transformer models tend to suffer from over-fitting thus leading to inferior performance. Inspired by the recent success of GPT models in natural language processing (NLP), we propose a new auxiliary self-supervised task plugged to the backbone transformer, which enables GPT-like auto-regressive training on network traffic sequence without using ground-truth labels. Experiments demonstrate the proposed method greatly reduces the requirements of labels in network anomaly detection. For example, on ISCX 2012 dataset, given only 0.05% training labels our semi-supervised approach obtains nontrivial 81.7% (2-class) and 64.9% (5-class) Fl-scores on the validation set, which is far better than the supervised counterparts using the same training data. We hope our research could inspire more label-efficient methods in network traffic analysis.

Federated learning enables distributed and collaborative learning among multiple participants while protecting their privacy. However, due to its distributed nature, federated learning is vulnerable to Byzantine attacks. These attacks can poison the data or directly modify the model parameters, making the global model performance degrade or even leaving a backdoor. Existing strategies for mitigating Byzantine attacks require a priori information about the number of attackers or require additional validation datasets. However, prior knowledge of the number of Byzantine clients or the collection of representative validation datasets is not always feasible in practice. Moreover, recent research has shown that well-designed attacks can make malicious updates indistinguishable from benign ones by making them highly similar, thus bypassing existing defense methods that rely on these metrics.To tackle these problems, we propose Dynamic Clustering based Federated Learning (DCFL), a novel Byzantine robust FL approach without any additional validation datasets. The main idea behind DCFL is to rigorously constrain the magnitude and direction of local updates through norms and signs. To achieve this, we propose a novel metric that can effectively distinguish malicious updates from benign updates in terms of direction, which can help the server eliminate malicious updates before final aggregation. Our experiments on three datasets demonstrate the effectiveness of DCFL in mitigating various popular Byzantine attacks. Remarkably, the accuracy of the global model learned in the adversarial setting is even close to that of FedAVG under no attack.

Federated Learning (FL) has recently drawn considerable attention, enabling multiple end devices to collaboratively learn global models without collecting device data. In reality, end devices can usually be distributed in non-correlated environments and generate non-IID data, which may lead to the Artificial Intelligence (AI) model weight divergence among devices and model accuracy degradation after aggregation. In this paper, to address the non-IID data problem for FL, we treat this problem among end devices as a distribution adaptation problem among multiple source domains and analyze the feasibility of feature augmentation, and then propose a novel method called Label-wisE Distribution Adaptive Federated Learning (LEDA-FL). First, to reduce the divergence in the label-wise feature space, we integrate the modified Conditional Variational AutoEncoder (CVAE) to align the label-wise feature distributions among clients. Second, we augment the label-wise features for FL clients to improve the FL performance (test accuracy and communication efficiency). Finally, we conduct an extensive experiment on five popular datasets, and the experimental results show that our proposed method improves the test accuracy of the global model (e.g.,6.2% test accuracy improvement on CIFAR100 compared to FedProx) and the communication efficiency of FL (e.g., about 60% reduction in communication cost on CIFAR100 compared to FedProx).

Telemetry data is an important resource to detect anomaly of satellites in orbit. In recent years, the telemetry data-driven satellite anomaly detection has drawn great attention from academia and industries. However, in prior arts, the time-or frequency-domain data feature is usually separately utilized; and, to reduce the compute complexity, a small portion of data dimensions are usually selected from the telemetry data manually, leading to reduction in detection accuracy. Motivated by this, we propose a novel satellite telemetry data anomaly detection approach using high-dimension telemetry data, and the Multiple Factors and Co-Attention based LSTM (MFCA-LSTM) model. Specifically, to achieve an accurate prediction of telemetry sequence, we first propose the MFCA-LSTM model, jointly considering the time- and frequency-domain data feature, and the auxiliary information, e.g., telecommand and mission planning. Then, by using the MFCA-LSTM model, we construct a two-level anomaly detection framework to efficiently detect full-dimension data; and, to improve the accuracy rate and reduce the false alarm rate, we further propose an adaptive Tukey test algorithm to determine the dynamic thresholds. Finally, we perform extensive experiments based on two real-world datasets, and the results testify the effectiveness of our proposed framework.

Robust keypoint detection on omnidirectional images against large perspective variations, is a key problem in many computer vision tasks. In this paper, we propose a perspectively equivariant keypoint learning framework named OmniKL for addressing this problem. Specifically, the framework is composed of a perspective module and a spherical module, each one including a keypoint detector specific to the type of the input image and a shared descriptor providing uniform description for omnidirectional and perspective images. In these detectors, we propose a differentiable candidate position sorting operation for localizing keypoints, which directly sorts the scores of the candidate positions in a differentiable manner and returns the globally top-K keypoints on the image. This approach does not break the differentiability of the two modules, thus they are end-to-end trainable. Moreover, we design a novel training strategy combining the self-supervised and co-supervised methods to train the framework without any labeled data. Extensive experiments on synthetic and real-world 360° image datasets demonstrate the effectiveness of OmniKL in detecting perspectively equivariant keypoints on omnidirectional images. Our source code are available online at https://github.com/vandeppce/sphkpt.

Mobile edge computing (MEC), extending computing services from cloud to edge, is recognized as one of key pillars to facilitate real-time services and tackle backhaul bottleneck. However, it is not economically efficient to attach intensive security appliances to every MEC node to defend application-level DDoS attacks and ensure the availability of services. Thus, we explore the elasticity of security defense among MEC nodes by proposing a COoperative DEfense (CODE) framework for MEC, referred to asCODE4MEC. CODE4MEC aims to adapt to traffic changes by coordinating container-carried defensive resources among cooperative MEC nodes in an automatic way. Towards this aim, we propose four control plane functions to enable a life-cycle management for CODE4MEC, namely, CODE triggering, scheduling, coordination and releasing. However, an effective CODE4MEC requires non-trivial algorithmic schemes, in particular for CODE scheduling and coordination functions. We thus design an online combinatorial auction mechanism for real-time CODE scheduling, and prove a tighter performance bound relative to prior arts. As for CODE coordination, a flow-based traffic and context information coordination scheme is proposed to enable classical defense schemes to work properly and efficiently. Finally, using a combination of real testbed and simulation evaluations, we validate the effectiveness of CODE4MEC.

Neural networks are vulnerable to adversarial examples. The adversary can successfully attack a model even without knowing model architecture and parameters, i.e., under a black-box scenario. Previous works on word-level attacks widely use word importance ranking (WIR) methods and complex search methods, including greedy search and heuristic algorithms, to find optimal substitutions. However, these methods fail to balance the attack success rate and the cost of attacks, such as the number of queries to the model and the time consumption. In this paper, We propose PAthological woRd Saliency sEarch (PARSE) that performs the search under dynamic search space following the subarea importance. Experiments show that PARSE can achieve comparable attack success rates to complex search methods while saving numerous queries and time, e.g., saving at most 74% of queries and 90% of time compared with greedy search when attacking the examples from Yelp dataset. The adversarial examples crafted by PARSE are also of high quality, highly transferable, and can effectively improve model robustness in adversarial training.

Lightweight virtualization technology represented by the container is developing rapidly and plays an increasingly important role in cloud computing. Container technology is efficient and flexible, but the kernel-sharing property also means its isolation is incomplete, making multi-tenancy container clouds vulnerable to co-resident attacks. The existing placement strategy either only considers the resource allocation problem or only considers the co-resident security problem of the virtualization technology on the same host. In this paper, we present SecCPS, a Secure Container Placement Strategy using deep reinforcement learning in a typical cloud scene. SecCPS learns to find the optimal placement to mitigate co-resident threats between containers running on virtual machines. In the design of SecCPS, we (1) present the challenges of securely placing containers running on virtual machines and establish metrics to describe co-residence in container clouds; (2) develop a secure container placement strategy using deep reinforcement learning as a preventive measure for co-residence; (3) implement the strategy and evaluate its effectiveness and scalability. The results show that SecCPS can effectively reduce co-residence compared with existing strategies and is scalable and flexible to large-scale container deployment scenarios.

Hardware-assisted virtualization is widely adopted in cloud computing and desktop platforms, significantly enhancing the security and efficiency of full virtualization systems by dividing multiple guest operating systems (OSes) and providing isolation between guest OSes and host OS. Thus, malware is usually analyzed in a guest OS to prevent the host OS from being compromised. However, an advanced malware will deliberately masquerade as a benign program when it detects that it is in a virtualized environment. The ideal goal of virtualization is to be utterly transparent to the guest processes, i.e., a guest process cannot be aware that it runs in a virtualized environment. Nowadays, although hardware-assisted virtualization detection is challenging, we present three novel approaches that can be leveraged to detect hardware-assisted virtualization. Our detection approaches are mainly based on the following features: i) The available CPU instruction sets are different in native and virtualized environments; ii) Modification to specific control register flag has different effects in native and virtualized environments; iii) The physical addresses in native and virtualized environments have different effects on the eviction of the level 2 CPU cache. We have conducted intensive experiments in both native and cloud environments. The experimental results show that our approaches can effectively detect the existence of hardware-assisted virtualization.

In the insider threat detection domain, data imbalance is a well-known problem. Most existing solutions, including rebalancing datasets and anomaly detection, have problems such as model overfitting, high cost, and high False Positive Rate (FPR). Therefore, how to effectively detect insider threats on an imbalanced dataset is a challenge. This paper proposes a new Siamese-architecture Insider Threat Detection (SITD) method, which detects insider threat by judging whether the input sample pairs belong to the same category instead of directly classifying a sample while avoiding the abovementioned problems. In addition, we improve the contrastive loss function to make the model pay more attention to the samples pairs of different categories, which significantly enhances the detection performance. Experimental results show that SITD outperforms other insider detection methods on the imbalanced CERT dataset. Moreover, SITD can achieve a good result no matter how imbalanced the dataset is.

The application of deep neural networks (DNNs) on 360-degree images has achieved remarkable progress in the recent years. However, DNNs have been demonstrated to be vulnerable to well-crafted adversarial examples, which may trigger severe safety problems in the real-world applications based on 360-degree images. In this paper, we propose an adversarial attack targeting spherical images, called 360-attactk, that transfers adversarial perturbations from perspective-view (PV) images to a final adversarial spherical image. Given a target spherical image, we first represent it with a set of planar PV images, and then perform 2D attacks on them to obtain adversarial PV images. Considering the issue of the projective distortion between spherical and PV images, we propose a distortion-aware attack to reduce the negative impact of distortion on attack. Moreover, to reconstruct the final adversarial spherical image with high aggressiveness, we calculate the spherical saliency map with a novel spherical spectrum method and next propose a saliency-aware fusion strategy that merges multiple inverse perspective projections for the same position on the spherical image. Extensive experimental results show that 360-attack is effective for disturbing spherical images in the black-box setting. Our attack also proves the presence of adversarial transferability from Z2 to SO(3) groups.

The safety of Deep Neural Networks (DNNs) processing omnidirectional images (ODIs) is an under-researched topic. In this paper, we propose a novel sparse attack, named Single-Perspective (SP) Attack, towards fooling these models by perturbing only one perspective image (PI) rendered from the target ODI. The attack is launched from the perspective domain, and finally the perturbation is transferred to the original ODI. To this end, we propose an effective PI position searching algorithm based on Bayesian Optimization, and then corrupt the PI centered on the desirable position with unconstrained/constrained perturbations. Extensive experiments on synthetic and real-world omnidirectional datasets demonstrate that SP Attack can overcome the projection deformation of ODIs, and mislead the neural networks by limiting the perturbations in a single patch on the target ODI.

Panoramic images are usually viewed through Head Mounted Displays (HMDs), which renders only a narrow field of view from the raw panoramic image. This distinctive viewing feature has largely been ignored when inpainting panoramic images. To address this issue, we propose a viewport-oriented generative adversarial panoramic image inpainting network in this paper. For capturing the distorted features accurately in the generating process of equirectangular projection (ERP) panoramic image, a latitude-adaptive feature fusion module is devised to aggregate the latitude-level features in ERP image and less-distorted patch-level viewport-domain features. Furthermore, a novel cross-domain discriminator is proposed to force the inpainting network to generate more plausible results in viewports. Extensive experiments show that our model achieves better performance compared to the baseline methods, especially in the viewport images.

Deep learning models are vulnerable to adversarial examples that add small perturbations to original inputs, which can help to evaluate the robustness and expose deficiencies of deep learning models. Current research mainly focus on English text adversarial attacks. Due to the differences between Chinese and English, the attack methods can not directly be used in Chinese attacks. In the mean time, in the Chinese text adversarial examples generation field, existing methods can't corporate the multi-modal characteristics of the Chinese language automatically. Therefore, in this paper, we present GPSAttack, a unified glyphs, phonetics, and semantics multi-modal attack method that can generate deceitful texts efficiently against Chinese text classification models. We conducted exhaustive attack experiments on convolutional, recurrent, and BERT networks to evaluate our method. The results show that our method achieves up to 90% attack success rate while modifying less than 4 characters per sentence and remaining readable to humans.

Textual adversarial examples crafted with well-designed perturbation can mislead state-of-the-art natural language models. Most previous works on word-level black-box attacks propose different text substituting strategies based on the word saliency determined by Leave One Out (LOO) methods, while the attack effectiveness is actually limited due to the model pathology. The word saliency determined by LOO methods can be severely affected by the model pathology, and unconscious bias is introduced. In this paper, we propose a word saliency method called Second-Order Enhanced Word Saliency (SOEWS), which considers the overfitting information obtained from the second-order model pathological behavior that helps to reduce the bias of LOO methods. Extensive experiments show that our method outperforms current black-box word saliency methods and can boost the effectiveness of previous attack frameworks. Performing adversarial training utilizing our method best improves model robustness, and human evaluation shows that the adversarial examples crafted by our method are almost imperceptible to humans. We also conduct case study to intuitively demonstrate the superiority of our method.

Deep Neural Networks (DNNs) have been widely used in everyday life owing to their impressive performance in complex machine learning tasks. The performance however comes at the cost of high computational complexity, which hinders the application of many DNN models in resource-constrained Internet-of-Things (IoT) and mobile devices. Device-edge collaborative DNN inference (referred to as co-inference) is an effective way to address the issue. However, it requires non-trivial algorithmic design, since the compound performance indicators including the inference efficiency and accuracy, and the data privacy have to be jointly considered. In this paper, we extend the degree of flexibility of the classical co-inference schemes, and propose a joint model, partitioning point and privacy differential intensity adaptation framework for co-inference, which comprises of the offline and online phases. In the offline phase, we train the co-inference model set that consists of a series of sub-models with different complexities, and profile the necessary performance of the sub-models. On that basis, we design an efficient algorithm for the online phase to promptly choose the sub-model, partitioning point and privacy differential intensity to meet the latency constraint and achieve the optimal accuracy-privacy tradeoff. Finally, extensive evaluations are carried out to demonstrate the effectiveness of our proposed framework.

Abstract Rotational-XOR cryptanalysis is a very recent technique for ARX ciphers. In this paper, the probability propagation formula of RX-cryptanalysis in modular addition is extended, and the calculation of RX-difference probability for any rotation parameter ($0

With the capability of trading accuracy for latency on-the-fly, the technique of adaptive early-exit inference has emerged as a promising line of research to accelerate the deep learning inference. However, studies in this line of research commonly use a group of thresholds to control the accuracy-latency trade-off, where a thorough and general methodology on how to determine these thresholds has not been conducted yet, especially with regard to the common requirements of average inference latency. To address this issue and enable latency-aware adaptive early-exit inference, in the present paper, we approximately formulate the threshold determination problem of finding the accuracy-maximum threshold setting that meets a given average latency requirement, and then propose a threshold determination method to tackle our formulated non-convex problem. Theoretically, we prove that, for certain parameter settings, our method finds an approximate stationary point of the formulated problem. Empirically, on top of various models across multiple datasets (CIFAR-10, CIFAR-100, ImageNet and two time-series datasets), we show that our method can well handle the average latency requirements, and consistently finds good threshold settings in negligible time.

Cloud computing provides a shared pool of configurable computing resources and significantly improves the utilization of computing resources, but it also introduces security threats. We focus on the virtual machine (VM) co-residency, which allows an attacker to launch cross- VM attacks against target VMs. To tackle this problem, we propose a secure and efficient VM allocation strategy to reduce the cross- VM attack threats while ensuring the efficiency of the cloud data center. First, we establish several metrics related to security and efficiency for the cloud data center. Then, we establish a constrained optimization model. Next, we allocate and migrate VMs based on typical suspicious or vulnerable VM features, and solve the optimization problem through our improved NSGA-II allocation. Finally, we implement our allocation strategy and conduct intensive experiments. The experimental results show that our allocation strategy performs better than the existing stratezies and orovides cloud vendors with tradeoff solutions.

With the rapid advancements in Augmented Reality, the number of AR users is gradually increasing and the multiuser AR ecosystem is on the rise. Currently, AR applications usually present results without limitations, which causes great latent danger to users, so it is necessary to apply strategies to ensure the safe output of AR. Due to the environmental diversities among the distributed users, the traditional approaches designed for single-user AR are not efficient for multi-user AR applications. Considering the characteristics of multi-user AR scenarios, we propose a multi-user AR output strategy model based on Federated Reinforcement Learning. With the device-fog-cloud hierarchical architecture, the proposed models are obtained first by Reinforcement Learning on users' devices, and are then hierarchically aggregated on the fog nodes and cloud server. We performed extensive AR simulations in Unity and obtained the results that show our method can avoid several security problems existent in multi-user AR applications.

Similar to conventional video, the increasingly popular 360 virtual reality (VR) video requires copyright protection mechanisms. The classic approach for copyright protection is the introduction of a digital watermark into the video sequence. Due to the nature of spherical panorama, traditional watermarking schemes that are dedicated to planar media cannot work efficiently for 360 VR video. In this article, we propose a spherical wavelet watermarking scheme to accommodate 360 VR video. With our scheme, the watermark is first embedded into the spherical wavelet transform domain of the 360 VR video. The spherical geometry of the 360 VR video is used as the host space for the watermark so that the proposed watermarking scheme is compatible with the multiple projection formats of 360 VR video. Second, the just noticeable difference model, suitable for head-mounted displays (HMDs), is used to control the imperceptibility of the watermark on the viewport. Third, besides detecting the watermark from the spherical projection, the proposed watermarking scheme also supports detecting watermarks robustly from the viewport projection. The watermark in the spherical domain can protect not only the 360 VR video but also its corresponding viewports. The experimental results show that the embedded watermarks are reliably extracted both from the spherical and the viewport projections of the 360 VR video, and the robustness of the proposed scheme to various copyright attacks is significantly better than that of the competing planar-domain approaches when detecting the watermark from viewport projection.

Although deep neural networks have demonstrated exceptional performance in substantial computer vision tasks, they can be easily confused by carefully generated adversarial examples. Via a novel technique we call activation visualization, the particular characteristics of adversarial examples are analyzed in this paper. Observing that the dominant features of adversarial examples are distributed over a high-dimensional space, we propose a defense framework named RegionSparse that projects the images into a low-dimensional space to remove the influence of the adversarial perturbations on the performance of deep neural networks. In RegionSparse, after training a robust global dictionary, the region where pixels are highly related to classification is firstly located by an object localization mechanism, then the sparse coding is performed on the located object region, together with a perturbation suppression for the remaining region. Extensive experiments on ImageNet dataset for gray-box, black-box, and transferred attacks are performed and the results show that RegionSparse can eliminate up to 90% attacks delivered by strong attacks including Momentum Iterative Fast Gradient Sign Method and Carlini-Wagner's L2attack.

As an essential part to intelligently fine-grained scheduling, planning and maintenance in smart grid and energy internet, short-term load forecasting makes great progress recently owing to the big data collected from smart meters and the leap forward in machine learning technologies. However, the centralized computing topology of classical electric information system, where individual electricity consumption data are frequently transmitted to the cloud center for load forecasting, tends to violate electric consumers' privacy as well as to increase the pressure on network bandwidth. To tackle the tricky issues, we propose a privacy-preserving framework based on the edge-fog-cloud continuum for smart grid. Specifically, 1) we gravitate the training of load forecasting models and forecasting workloads to distributed smart meters so that consumers' raw data are handled locally, and only the forecasting outputs that have been protected are reported to the cloud center via fog nodes; 2) we protect the local forecasting models that imply electricity features from model extraction attacks by model randomization; 3) we exploit a shuffle scheme among smart meters to protect the data ownership privacy, and utilize a re-encryption scheme to guarantee the forecasting data privacy. Finally, through comprehensive simulation and analysis, we validate our proposed privacy-preserving framework in terms of privacy protection, and computation and communication efficiency.

Bring-Your-Own-Device (BYOD) has been widely viewed as a definite trend among enterprises in which employees bring and use their personal smartphones for work. Despite the perceived opportunities of increasing productivity and reducing costs, BYOD raises severe security and privacy concerns: the corporate networks and data are directly exposed to malware apps running on the personal smartphones. This highlights the necessity for performing real-time mobile malware detection in BYOD environments. Deep learning seems to be a natural choice to handle such detection, due to its state-of-the-art detection effectiveness. However, deep learning inference is usually too computationally complex for resource-constrained smartphones, and the communication overhead of cloud-based inference may be unacceptable. As a result, it is hard to seek the tradeoff between the real-time demand and optimality of detection accuracy. In this paper, we tackle this issue by proposing an endedge coordinated inference approach that can support highlyaccurate and average latency guaranteed malware detection. Our proposed approach integrates the early-exit and model partitioning methods to allow fast, correct and smartphonelocalized inference to occur frequently. Extensive evaluations are carried out, demonstrating that our proposed approach offers a good compromise between detection accuracy and efficiency.

Linear cryptanalysis is an important evaluation method for cryptographic primitives against key recovery attack. In this paper, we revisit the Walsh transformation for linear correlation calculation of modular addition, and an efficient algorithm is proposed to construct the input-output mask space of specified correlation weight. By filtering out the impossible large correlation weights in the first round, the search space of the first round can be substantially reduced. We introduce a concept of combinational linear approximation table (cLAT) for modular addition with two inputs. When one input mask is fixed, another input mask and the output mask can be obtained by the Splitting-Lookup-Recombination approach. We first split the n-bit fixed input mask into several subvectors and then find the corresponding bits of other masks, and in the recombination phase, pruning conditions can be used. By this approach, a large number of search branches in the middle rounds can be pruned. With the combination of the optimization strategies and the branch-and-bound search algorithm, we can improve the search efficiency for linear characteristics on ARX ciphers. The linear hulls for SPECK32/48/64 with a higher average linear potential ( ALP ) than existing results have been obtained. For SPARX variants, an 11-round linear trail and a 10-round linear hull have been found for SPARX-64 and a 10-round linear trail and a 9-round linear hull are obtained for SPARX-128. For Chaskey, a 5-round linear trail with a correlation of 2−61 has been obtained. For CHAM-64, 34/35-round optimal linear characteristics with a correlation of 2−31/2−33 are found.

Inference based on deep learning models is usually implemented by exposing sensitive user data to the outside models, which of course gives rise to acute privacy concerns. To deal with these concerns, Donget al.recently proposed an approach, namely the dropping-activation-outputs (DAO) first layer. This approach was claimed to be a non-invertible transformation, such that the privacy of user data could not be compromised. However, In this paper, we prove that the DAO first layer, in fact, can generally be inverted, and hence fails to preserve privacy. We also provide a countermeasure against the privacy vulnerabilities that we examined.

The development of Internet of Things (IoT) dramatically facilitates the integration of computing systems with the physical world. However, as IoT devices are more easy to compromise than desktop computers, cybercriminals have founded IoT-based botnets to launch Distributed Denial of Service (DDoS) attacks with unprecedented traffic volume. To mitigate the damages associated with these attacks, the detection of IoT-based botnet has to preempt the command and control (C&C) communication to prevent the delivery of the attack codes. Motivated by the extensively implementation of domain generation algorithm in botnets, in this article, we propose ConnSpoiler, a lightweight system that detects IoT-based botnets by identifying the stream of algorithmically generated domains (AGDs) in a fast way. ConnSpoiler only needs negligible system resources to take effect and thus can execute well on the resource-restraint IoT devices. By outfitting a powerful statistical algorithm, i.e., threshold random walk, ConnSpoiler has a high probability (about 94%) of detecting infection before the compromised devices connect C&C servers, which can help to prevent the succeeding attacks. Moreover, ConnSpoiler only requires the benign domains to take effect and therefore does not need extra effort to label malicious samples for training phase. We evaluate ConnSpoiler based on real-world DNS traffics collected from two different large ISP networks and show that it accurately identifies devices that are compromised by unknown botnets.

Auditing security compliance with respect to security standards and policies becomes increasingly important in clouds for ensuring the transparency and accountability of a cloud provider to its tenants. However, security auditing in clouds encounters various challenges in the scalability and response time due to the large-scale cloud size and the high operational complexity. Existing approaches cannot verify the legitimacy of user requests in proper response time at runtime for a large cloud. To this end, this paper proposes a novel security auditing framework named Deep lEArning-based user-level Proactive Security auditing (DeaPS) for clouds, which leverages the Long Short-Term Memory (LSTM) neural network to automatically learn user behavior patterns from historical events and notify for possible critical events causing violations. Our solution implements the costly verification in advance for reducing the runtime response time to a realistic level. We evaluate our approach by integrating DeaPS into OpenStack and extensive experiments show that DeaPS exhibits excellent performance in large-scale clouds and outperforms other existing security auditing methods.

Insider threats gradually exert great influence in cy-bersecurity, causing a significant loss to organizations or companies. However, whatever the form of the threat is, insiders have to conduct the unauthorized activities through the communication traffic, such as controlling the victim systems and unauthorizedly requesting the resources. Moreover, as one of the most fundamental intranet resources, bandwidth is frequently targeted by insider attackers for sabotage to traffic communication and service delivery of the network. In this paper, we present a user-level full-lifecycle security management scheme for intranet traffic from anomaly detection to mitigation execution in an online manner. This scheme dynamically monitors abnormal users that deviate from normal behavior patterns through bidirectional Gated Recurrent Unit (bi-GRU) based online unsupervised log parser, then adaptively adjusts the traffic scheduling policy according to the adequate consideration of network security, network performance and user requirements by using deep Reinforcement Learning (RL) method for online decision-making. Extensive experimental evaluations show that our scheme can stably maintain the high performance of traffic scheduling and effectively mitigate multifarious traffic threats. Our work is a valuable step towards designing self-adaptive intranets that learn to enhance security management by themselves with high scalability and deployability.

Trust mechanisms based on trusted authorities are used to establish trust between two unfamiliar entities in conventional authentication, which cannot meet the higher security requirements of emerging technologies for authentication. Blockchain achieves a decentralized trust mechanism between a group of mutually distrusted nodes without a trusted authority by means of data encryption, timestamp and consensus mechanisms. In this work, we therefore propose an innovative blockchain-based authentication method with one-time password (OTP). Specifically, we propose an OTP scheme and apply it to the blockchain, where blockchain serves as an OTP verifier. We then develop a list of security criteria to analyze the security of the proposed authentication protocol. This paper shows that our scheme can resist replay attacks, brute force attacks, OTP forgery attacks and so on. We finally compare the proposed OTP scheme with the representative OTP schemes according to the designed evaluation criteria and conduct a series of comparative experiments on Hyperledger Fabric. The results show that the proposed scheme is superior in terms of performance and security.

Recent studies have shown that co-resident attacks have aroused great security threat in cloud. Since hardware is shared among different tenants, malicious tenants can launch various co-resident attacks, such as side channel attacks, covert channel attacks and resource interference attacks. Existing countermeasures have their limitations and can not provide comprehensive defense against co-resident attacks. This paper combines the advantages of various countermeasures and proposes a complete co-resident threat defense solution which consists of co-resident-resistant VM allocation (CRRVA), analytic hierarchy process-based threat score mechanism (AHPTSM) and attack-aware VM reallocation (AAVR). CRRVA securely allocates VMs and also takes load balance and power consumption into consideration to make the allocation policy more practical. According to the intrinsic characteristics of co-resident attacks, AHPTSM evaluates VM’s threat score which denotes the probability that a VM is suffering or conducting co-resident attacks based on analytic hierarchy process. And AAVR further migrates VMs with extremely high threat scores and separates VM pairs which are likely to be malicious to each other. Extensive experiments in CloudSim have shown that CRRVA can greatly reduce the allocation co-resident threat as well as balancing the load for both CSPs and tenants with little impact on power consumption. In addition, guided by threat score distribution, AAVR can effectively guarantee runtime co-resident security by migrating high threat score VMs with less migration cost.

Software-Defined Networking (SDN) controllers are deployed as a cluster in large-scale networks, for high availability and scalability. A device (e.g., switch) is connected to multiple controllers in cluster while assigned with one master controller, and the controller with master role is only permitted to manage the device. Therefore, the master controller assignment can affect the network throughput seriously. In this paper, we first study the assignment strategy of two well-known SDN platforms (i.e., OpenDayLight and Open Network Operating System) as well as the assignment mechanisms in the existing literature, analyzing their deficiencies. Meanwhile, we present an adaptive controller assignment so as to maximize the network throughput under the fluctuating network environment. We formulate this problem as an Integer Linear Program and propose an algorithm based on Ant Colony optimization to solve it efficiently. Extensive experiments validate that our solution is superior to other existing mechanisms in terms of network throughput and load balancing.

Aiming at various issues, such as performance, load-balance and robustness, of Software-Defined Networking (SDN) controller cluster, several deployment strategies have been proposed. As a supplement, our work delivers a cost-effective multicontroller deployment strategy in terms of economic benefits. In this paper, we formalize a Cost-optimized Multi-Controller Deployment (CO-MCD) problem by concerning the controller's deployment cost and the network service's delay cost. Meanwhile, we characterize the communication between the controller and the switch using a queuing system, and design an algorithm to determine the optimal deployment in the specified environment. Simulation results validate that the proposed algorithm can always obtain a deployment solution which minimizes the economic cost of the entire SDN network. This work aims at being a useful primer to provide insights regarding the economic impact of multiple controllers in the realistic SDN deployment, so that network owners plan their investments more rationally.

Mobile edge computing (MEC), as an enabling paradigm allowing mobile services to be performed at the edge of mobile networks, has been commonly expected to facilitate real-time services and offload backhaul traffic. However, the security issues raised by innovations in networks always hesitate us to embrace their benefits, and so MEC does. DDoS attacks originating from massive mobile devices in 5G, especially the vulnerable internet-of-things devices, is an unavoidable threat for MEC services. Unfortunately, each MEC node is hard to maintain service availability in the wake of DDoS attacks, owing to its limited hardware resources for defense. Therefore, we are motivated to study the cooperative defense against DDoS attacks for MEC, where defense resources can be shared between MEC nodes. Specifically, we purpose a global orchestration (GO) mechanism to make full use of freedom degrees of cooperative defense with respect to the number of cooperative participants and the amount of cooperative resource. In the process of designing the GO mechanism, we formulate the optimization problem of GO, and exploit the bucket elimination algorithm to find the optimal solutions. To make the GO mechanism practically usable, we propose a scalable approximation algorithm based on the mini-bucket elimination algorithm. Extensive simulations are carried out to validate the effectiveness of GO.

In the Software-Defined Networking (SDN) philosophy, a logically centralized control plane remotely instructs distributed network devices to forward packets over control channel. However, such a centralized design makes the control channel a bottleneck due to bandwidth fatigue, which seriously threatens the availability of the network. Existing studies economize control channel bandwidth at the expense of the visibility of most or mice flows. This paper presents EFastLane, an effective solution for bandwidth-efficient flow setup without sacrificing any flow visibility. EFastLane advocates that the controller only sends the entire processing rule to an intermediate switch of the forwarding path, and then the rule is installed bidirectionally along the path. Besides, we propose two algorithms to locate such a target switch to ensure that all switches on the path install the forwarding rule before receiving packets. Simulation experiments validate that EFastLane saves over 70% of control channel bandwidth compared to traditional SDN, while introducing the negligible flow setup time and traffic delay.

Band selection refers to finding the most representative channels from hyperspectral images. Usually, certain objective functions are designed and combined via regularization terms. Owing to the parameters independence and the optimal solutions, multi-objective based methods have presented promising performance. However, the characteristics of the hyperspectral band selection problem make its range to be discrete. In this case, recently proposed weighted Tchebycheff based multi-objective band selection methods could only reach the weakly Pareto optimal, which would result in non-unique solutions. In this paper, we improve the decomposition process of the multi-objective based band selection method via a boundary intersection approach. Compared with weighted Tchebycheff decomposition, the proposed method is able to change the shape of the contour lines between Pareto Front and the ideal point, and this approach is particularly suitable for discrete-range problems. The effectiveness of our improvement is demonstrated by comparison experiments.

In remote sensing images, detecting aeroplanes of special shapes is difficult due to limited number of samples. Without enough training samples, most supervised learning based algorithms will fail. Focusing on the specially-shaped aeroplanes in high-resolution optical remote sensing imagery, this paper presents a single-sample approach. The proposed approach takes one sample as input and directly searches for similar matches from the image. Unlike the supervised learning algorithms which extracts information from positive and negative samples, the hyperspectral algorithm estimates the statistics of background by analyzing the global information of the target image, needless to provide negative samples. Furthermore, this algorithm tries to find a hyperplane projected on which the background is compressed while the target is preserved, making it more data-adaptive than the conventional similarity measurements. Experiments on real data have presented the robustness of the proposed method.

Sparse representation based hyperspectral unmixing methods have attracted increasing investigations during the past decade. Recently, multiple signal classification (MUSIC) algorithm has been verified effective in reducing the mutual coherence of the spectral library. However, the popular pre-pruning strategy by MUSIC cannot guarantee that the end-members exactly exist in the selected spectral subset when the image noise is serious. In this paper, we propose a new sparse unmixing method for hyperspectral images via integrating the pruning operation into the optimization process. The projection of the library is represented by an objective function in the proposed method. To avoid the manually settings of regularization parameters, we develop a new multi-objective based method where reconstruction error, sparsity error and the projection function are considered as three parallel objectives that could be optimized simultaneously. Experimental results have indicated the superiority of the proposed method, especially in high-noise conditions.

Recent research has shown that co-resident attacks can lead to cross-tenant information leakage in cloud. Existing solutions for mitigating co-resident attacks generally require significant changes to hypervisors, guest OSes or hardwares, even periodic VM migration, which might not be immediately applied to cloud datacenters. In this paper, we propose the security-aware VM placement approach (SecVMP) to minimize co-residency and mediate conflicts between tenants for proactively mitigating co- resident attacks in cloud. The proposed approach also takes load balancing and power consumption into consideration to make itself more practical. It consists of three main parts: the security-aware co-resident rules (SecCRRs), the security-aware VM allocation algorithm (SecVMA) and the security-aware VM migration algorithm (SecVMM). The SecCRRs are based on characteristics of known co-resident attacks and stipulate which pairs of VMs are conflicting so that they shouldn't be co-resident on the same physical server. The SecVMA proactively avoids being co-resident with conflicting tenants and minimizes co-residency as much as possible when launches VMs. And the SecVMM migrates conflicting VMs and separates conflicting tenants in time. Experimental results in CloudSim show that the SecVMA can greatly reduce the co-residency of cloud with little impact on load balancing and power consumption. In addition, triggered by the SecCRRs, the SecVMM can separate conflicting tenants with fewer migrations to reduce co-resident risk.

Table-overflow occurs frequently in switches due to the limited flow table storage space, which has seriously hampered thelarge-scale deployment of SDN. To mitigate this threat, we introduce a new, proactive switch space intelligent management framework that can react actively before table- overflow occurs. Our solution, referred to as CPD, contains three modules: flow table status collection, flow prediction and proactive flow entry deletion. To preserve the function of switch, flow prediction predicts the number of new flows entering the network based on historical information, and proactive flow entry deletion evaluates the likelihood of table-overflow according to the predicted value and the state of current flow table gathered by status collection and proactively deletes the flow entries through a flow-based eviction algorithm. Further, we determine the best sampling period by solving a multi-objective optimization problem. Finally, the simulation results show that CPD can prevent about 90% of table-overflow and significantly reduce over 20% occurrences of table-miss compared with existing mechanisms.

5G mobile edge computing (MEC), which pushes mobile services to the edge, has been recognized as an effective solution to enhance mobile users' quality of service, as well as to tackle the backhaul bottleneck. Although its architecture and service related techniques have drawn sufficient attentions, solutions to the security defense are still open. Therefore, we are motivated to propose a cooperative defense (CODE) framework against DDoS attacks for MEC by leveraging network function virtualization and software-defined networking architectures; in the framework, MEC nodes owning spare defense resource are orchestrated to help MEC nodes whose incoming traffic overwhelms their self-defense capability. To explore the elasticity space of CODE among multiple MEC nodes and develop an online resource management method of such CODE, we formulate the multi-requester multi-provider resource management problem for CODE, jointly considering the defense resource usage efficiency and the fairness of CODE participants. To balance complexity and performance for the formulated problem, we are motivated by online combinatorial auctions, and propose an online algorithm that has a provable performance guarantee. Finally, extending the MEC simulation platform that is used in our previous work, we validate the effectiveness in terms of resilient defense capability, fairness and computation efficiency.

The cooperative Mobile Edge Computing (MEC) caching, where caches are distributed in 5G edge to cooperatively bring popular contents closer to mobile users, is concocted with high expectation on improving users' quality of experience. However, the system may suffer severely from the pollution attack. By frequently requesting unpopular contents, the adversaries in the pollution attack can break the patterns of content popularity. As a result, the cooperative content placement decision making, which relies on the popularity pattern, will be misguided to store unpopular contents instead of popular ones. In this paper, we explore the behaviors and threats of pollution attack in cooperative MEC caching. Starting with single-inject pollution where adversaries concentrate the attack on a single cache, we study the behaviors and diffusion features from both statistical experiments and theoretical analysis. Results show that the unpopular contents can fill into caches within 2 hops from attackers, and caches within 3 hops suffer from severe damage. Meanwhile, we also derive the upper boundary of the attack damage and present a boundary analysis. Furthermore, in multiple-inject pollution where the adversaries launch multiple single-inject pollution simultaneously, we evaluate the threat when attackers intelligently organize the attacks. It is found that the adversaries can cause 44% severer damage and cover 45% more area after intelligently organizing the attacks. Our work in this paper provides a fundamental basis for countermeasures designing.

Moving Target Defense(MTD) provides a promising solution to reduce the chance of weakness exposure by constantly changing the target's attack surface. Though lots of MTD technologies have been researched to defend network attacks, there is little systematic study on security assessment of MTD. This paper proposes a novel method to quantify the security of MTD system which based on three factors: Vulnerability Entropy, Attack Entropy and Attenuation Entropy. This assessment model provides a theoretical and practical guidance for building MTD system and improving MTD technology.

SCADA (Supervisory Control and Data Acquisition) system is the core component of industrial and critical infrastructure, and cyber security of SCADA system has become the key consideration of system managers and engineers. Therefore, a great many of standards, guidelines and best practices have been developed to give reference of SCADA system cyber security, hoping to provide some instructions for system managers. Unfortunately, there is little consensus on what to do. Whats worse, it is difficult to choose the right one for a particular industrial scene. These standards are usually long and complex texts, whose reading and understanding often takes much time and effort. We provide a comprehensive and structured review of SCADA cyber security standards, guidelines and best practices with three dimensions: release time, geographic location and intended audience. Finally, we use the theory of defense-in-depth as a reference to evaluate these standards. It is concluded that no standard performs better than others on all the criteria and that we should integrate different standards to apply them to a specific industrial scene.

In this paper, we design a game theoretical framework for improving the Quality of Service (QoS) in cooperative RAN caching. Considering the cooperation under both single cell transmission and joint transmission, the QoS metric is uniformly quantified as the total content delivery time. Although the formulated cooperative content placement problem is proved NP-hard, noticing the local cooperative characteristics, we transform the problem into Local Altruistic Gaming where the Nash Equilibrium (NE) can be guaranteed and distributive algorithms such as Spatial Adaptive Play (SAP) are applicable. Then, two distributed learning algorithms are proposed, where the former overcomes the execution difficulties over tremendous action set in traditional SAP, and the latter further accelerate the convergence by reducing the number of additional suboptimal NEs brought by the former. To further improve the computation efficiency, an updating scheme is constructed to enable parallel updating in the proposed algorithms. Finally, based on a real-world LTE traffic dataset, the performance of the proposed algorithms and the updating scheme have been validated.

This paper addresses mainly the problem of private data protection in network communication against eavesdropping attacks. As this kind of attacks is stealthy and untraceable, it is barely detectable for those feature detection or static configuration based passive defense approaches. We propose a Moving Target Defense(MTD) method by utilizing the protocol customization ability of Protocol-Oblivious Forwarding (POF). The novel full protocol stack randomization MTD can greatly increase the difficulty of implementing network eavesdropping attack and protect the privacy of the network communication process.