Federated learning (FL) is widely used in Internet-of-Things (IoT) systems, but its distributed training process also exposes it to backdoor attacks. Existing studies mainly consider single-target or centralized multi-target settings, while coordinated distributed multi-target attacks remain underexplored. In practical IoT scenarios, one adversarial entity may control multiple distributed malicious clients and assign each client distinct triggers and target labels. Under this setting, existing distributed backdoor methods often fail to preserve the effectiveness of all backdoors because malicious updates conflict during aggregation. To address this issue, we propose a Distributed Multi-Target Backdoor Attack (DMBA) for FL. DMBA introduces a Backdoor Replay (BR) mechanism to reduce discrepancies among malicious gradients and a Channel-Frequency Composite Trigger (CFCT) strategy to improve trigger distinguishability and alleviate local interference. Experiments on multiple datasets show that DMBA ensures attack success rates above 80% for all implanted back-doors, whereas some baseline backdoors fall below 50% and may even approach 0.

Deep neural networks are vulnerable to adversarial examples which fool model predictions by adding imperceptible perturbations to natural examples. Adversarial training is effective in defending against adversarial attacks but faces a challenge with long-tailed data, where the over-compression of tail feature space undermines the reliability of defect classification models. To address this problem, we propose APRCB-AT, a novel adversarial training framework that integrates an adaptive perturbation radius and class-balanced loss. We find a positive correlation between the aggressiveness of adversarial examples and the perturbation radius within a certain range. Based on this, APRCB-AT assigns a larger perturbation radius in a certain range to the tail classes, while incorporating class-balanced loss with regularization to penalize the head classes. The experimental results show that APRCB-AT achieves 70.28% robust accuracy against adversarial attacks, surpassing existing methods, such as LSRG-DRW(67.22%) and REAT(68.06%).

Offline reinforcement learning (RL) heavily relies on the coverage of pre-collected data over the target policy’s distribution. Existing studies aim to improve data-policy coverage to mitigate distributional shifts, but overlook security risks from insufficient coverage, and the single-step analysis is not consistent with the multi-step decision-making nature of offline RL. To address this, we introduce the sequence-level concentrability coefficient to quantify coverage, and reveal its exponential amplification on the upper bound of estimation errors through theoretical analysis. Building on this, we propose the Collapsing Sequence-Level Data-Policy Coverage (CSDPC) poisoning attack. Considering the continuous nature of offline RL data, we convert state-action pairs into decision units, and extract representative decision patterns that capture multi-step behavior. We identify rare patterns likely to cause insufficient coverage, and poison them to reduce coverage and exacerbate distributional shifts. Experiments show that poisoning just 1% of the dataset can degrade agent performance by 90%. This finding provides new perspectives for analyzing and safeguarding the security of offline RL.

Backdoors on federated learning will be diluted by subsequent benign updates. This is reflected in the significant reduction of attack success rate as iterations increase, ultimately failing. We use a new metric to quantify the degree of this weakened backdoor effect, called attack persistence. Given that research to improve this performance has not been widely noted, we propose a Full Combination Backdoor Attack (FCBA) method. It aggregates more combined trigger information for a more complete backdoor pattern in the global model. Trained backdoored global model is more resilient to benign updates, leading to a higher attack success rate on the test set. We test on three datasets and evaluate with two models across various settings. FCBA's persistence outperforms SOTA federated learning backdoor attacks. On GTSRB, post-attack 120 rounds, our attack success rate rose over 50% from baseline. The core code of our method is available at https://github.com/PhD-TaoLiu/FCBA.

In this study, we present synchronous bilingual Connectionist Temporal Classification (CTC), an innovative framework that leverages dual CTC to bridge the gaps of both modality and language in the speech translation (ST) task. Utilizing transcript and translation as concurrent objectives for CTC, our model bridges the gap between audio and text as well as between source and target languages. Building upon the recent advances in CTC application, we develop an enhanced variant, BiL-CTC+, that establishes new state-of-the-art performances on the MuST-C ST benchmarks under resource-constrained scenarios. Intriguingly, our method also yields significant improvements in speech recognition performance, revealing the effect of cross-lingual learning on transcription and demonstrating its broad applicability. The source code is available at https://github.com/xuchennlp/S2T.

Due to the cache’s capacity of forwarding information, Named Data Networking (NDN) has become a promising networking architecture. Since distributed caching is susceptible to cache pollution attacks (CPAs), researchers pay more attention to CPAs detection and defense. The current detection schemes seriously rely on an assumption that the content popularity remains stable over time. However, the change in interests of legitimate users in the network is unavoidable, which makes content popularity change dynamically. Thus, it is difficult to detect CPAs based on a static content popularity distribution. To address this issue, we propose a novel scheme to detect CPAs by analysing latency instead of popularity. The proposed scheme constructs the probability transfer matrix based on the Markov process of contents transfer and detects CPAs by the convergence states of the matrix. Once a CPA is detected, the affected router recognizes the attack type and adopts a specific defense method according to the attack type. This defense method can improve the network Quality of Service (QoS) by leveraging particular methods for different routers rather than the broadcasted global method. Extensive simulations in ndnSIM show that our scheme can effectively detect CPAs with higher detection ratio and defense CPAs with acceptable impacts on the overall network in network scenarios with dynamically changing content popularity.

The vulnerability of Named Data Network (NDN) causes a series of privacy problems. Path inference provides network privacy protection and network security improvement in NDN via obtaining the content transmission paths. The traditional path inference methods ignore the influence of in-network cache, leading to a lacking consideration of the confusion issue between content sources and cached copies. To facilitate the effectiveness and robustness of the path inference, we have proposed a novel method that reduces the influence of incomplete and misleading information. The proposed method consists of two parts. First, the hybrid attack method based on hop count variation is used for collecting complete request feedback information and merging all information to infer paths. Second, the benefit-driven evolutionary game is adopted to promote the cooperation of the timing attack and the cache pollution attack. The combination of the two types of attacks alleviates the restriction of in-network cache. The simulations in ndnSIM indicate that our proposed method infers paths with higher precision, recall, and F1-Score compared to other advanced methods.

Few-shot website fingerprinting (FSWF) attacks attempt to identify whether the users have access to specific websites based on a few training data. Existing FSWF attack methods focus on adapting to variable network conditions in real scenarios. They use various techniques to transfer the model to adapt to test data which has a different distribution from training data. However, recent methods ignore the impact of pre-training data diversity on adaptability. The poor data diversity caused by the user-specific data crawl limits representation ability, and further hinders rapid adaptation to new network conditions. Due to the extreme Non-IId between multiple attackers’ datasets, it is not feasible to mix multiple datasets or perform traditional federated learning methods to improve representation ability. To address the issue, we propose a novel method based on a joint learning framework to achieve the collusion FSWF attacks. The proposed method fuses the feature spaces of multiple user-side attackers to enhance the representation ability of the local model, and constructs a virtual fusion center to mitigate the impact of Non-IID. It improves the adaptability under variable network conditions for the local attacker. This paper conducts comprehensive experiments to evaluate the performance of the proposed method in both closed-world and open-world settings. Compared with the state-of-the-art method, the proposed method improves the accuracy by up to 13.02% in the closed-world setting and the AUC by up to 0.085 in the open-world setting, respectively.

Anchor link prediction enhances the effectiveness of digital forensics through the identification of multiple social network users. The current methods based on deep learning are characterized by both the exaggerated similarity between adjacent nodes in the same latent space and the variation in the feature spaces caused by semantics. A novel approach is developed to fuse the semantic features of different networks in this paper. The proposed method is divided into two stages. Firstly, representation learning pays more attention to the influence of uncertainty on the equivalence of node network structure, and introduces the difference between adjacent nodes from the latent space. Secondly, a joint representation learning framework trains and exchanges the parameters depending on known anchor links. The joint representation learning framework injects fused features into the representation learning processes of different networks. The combination of enhanced discrimination and cross-network feature fusion reduces the feature space differences caused by the semantics of different social networks. This paper conducts comprehensive experiments on social networks in the real world. The outcome shows that the proposed approach is more efficient and robust compared to the existing state-of-theart methods.

Anchor link prediction exacerbates the risk of privacy leakage via the de-anonymization in edge computing. The predictive effect of traditional unsupervised learning methods is too dependent on user attributes and supervised learning methods are sensitive to network structure noise. Anchor link prediction methods based on graph embedding are restricted by the sparsity of the observable anchor links which can be used for training. To facilitate the effectiveness and robustness of the anchor link prediction, we have proposed a novel method which reduces the restrictions on the observable anchor links used for training. The proposed method consists of two phases. First, graph embedding based on network structural roles is used to generate the latent feature space, reconciling the distinction and similarity between nodes. Second, the supervised learning for optimizing the Wasserstein distance which estimates the minimum amount of work to change one distribution into the other. The combination of the reconciled latent feature space and the estimate for the amount of change alleviates the restriction on observable overlapping parts. Extensive experiments on real-life social networks have demonstrated that the proposed method significantly outperforms the state-of-the-art methods in terms of both precision and robustness.

Federated learning is designed to train models in a distributed scheme while keeping the clients' data stored locally. The aggregation server only receives local models from clients and does not require clients to upload their local data, in which way it protects the clients' privacy. However, federated learning is vulnerable. The federated learning models are sensitive to poisoning attacks. Existing data poisoning attack methods assume that the attacker and the client have the same data distribution and data volume, which is unpractical. In this paper, we first propose a privacy inference-based poisoning data generation method, FLPDG. FLPDG changes the relationship between data and labels, and uses the data of benign clients to launch poisoning attacks. This method relies on the global model of an iterative update to obtain the data and labels of benign clients. Second, a privacy inference-based data poisoning attack model Poi_PDG is proposed. This model uses the FLPDG method to launch a data poisoning attack under conditions of insufficient original data volume of the attacker. Meanwhile, a defense method PDG_DF is proposed for Poi_PDG. It splits the image data into variance regions and utilizes GANs to hide the visual features of each image region. It controls the degree of feature hiding by setting different thresholds to keep the classification features of the image while ensuring the accuracy of the training model. Finally, several experiments are conducted to evaluate the proposed attack and defense methods, and the experimental results indicate the effectiveness of the methods.

Facing the explosive growth of data, the introduction of cloud computing in the Space-Air-Ground Integrated Internet of Things (SAGIIoT) can solve the problem of limited computing power of the terminals. At the same time, data security on the cloud is also a focus that cannot be ignored. Secure outsourcing computing is helpful in improving privacy preserving. Due to the wide applicability of$K $-means clustering, outsourcing computing for$K $-means has become a major research hotspot in industry and academia. Most of the existing work on outsourcing$K $-means clustering is based on homomorphic encryption, which has a high computational overhead due to the mathematical puzzles’ nature of homomorphic encryption. In addition, the high computational overhead of designing a verification algorithm based on homomorphic encryption is unacceptable. To address the above issues, we design a${K}$-means clustering outsourcing algorithm by sparse matrix transformation, which can verify the deceptive behavior of cloud while achieving high efficiency. In this article, we theoretically prove the accuracy, security, efficiency, and verifiability of the proposed algorithm. Extensive experiments indicate that our algorithm is efficient.

Anchor link prediction exacerbates the risk of privacy leakage via the de-anonymization of social network data. Embedding-based methods for anchor link prediction are limited by the excessive similarity of the associated nodes in a latent feature space and the variation between latent feature spaces caused by the semantics of different networks. In this article, we propose a novel method which reduces the impact of semantic discrepancies between different networks in the latent feature space. The proposed method consists of two phases. First, graph embedding focuses on the network structural roles of nodes and increases the distinction between the associated nodes in the embedding space. Second, a federated adversarial learning framework which performs graph embedding on each social network and an adversarial learning model on the server according to the observable anchor links is used to associate independent graph embedding approaches on different social networks. The combination of distinction enhancement and the association of graph embedding approaches alleviates variance between the latent feature spaces caused by the semantics of different social networks. Extensive experiments on real social networks demonstrate that the proposed method significantly outperforms the state-of-the-art methods in terms of both precision and robustness.

Anchor link prediction across social networks plays an important role in multiple social network analysis. Traditional methods rely heavily on user privacy information or high-quality network topology information. These methods are not suitable for multiple social networks analysis in real-life. Deep learning methods based on graph embedding are restricted by the impact of the active privacy protection policy of users on the graph structure. In this paper, we propose a novel method which neutralizes the impact of users’ evasion strategies. First, graph embedding with conditional estimation analysis is used to obtain a robust embedding vector space. Secondly, cross-network features space for supervised learning is constructed via the constraints of cross-network feature collisions. The combination of robustness enhancement and cross-network feature collisions constraints eliminate the impact of evasion strategies. Extensive experiments on large-scale real-life social networks demonstrate that the proposed method significantly outperforms the state-of-the-art methods in terms of precision, adaptability, and robustness for the scenarios with evasion strategies.

There is a new cache pollution attack in the information-centric network (ICN), which fills the router cache by sending a large number of requests for nonpopular content. This attack will severely reduce the router cache hit rate. Therefore, the detection of cache pollution attacks is also an urgent problem in the current information center network. In the existing research on the problem of cache pollution detection, most of the methods of manually setting the threshold are used for cache pollution detection. The accuracy of the detection result depends on the threshold setting, and the adaptability to different network environments is weak. In order to improve the accuracy of cache pollution detection and adaptability to different network environments, this paper proposes a detection algorithm based on gradient boost decision tree (GBDT), which can obtain cache pollution detection through model learning. Method. In feature selection, the algorithm uses two features based on node status and path information as model input, which improves the accuracy of the method. This paper proves the improvement of the detection accuracy of this method through comparative experiments.

As an innovative strategy, edge computing has been considered a viable option to address the limitations of cloud computing in supporting the Internet-of-Things applications. However, due to the instability of the network and the increase of the attack surfaces, the security in edge-assisted IoT needs to be better guaranteed. In this paper, we propose an intelligent intrusion detection mechanism, FedACNN, which completes the intrusion detection task by assisting the deep learning model CNN through the federated learning mechanism. In order to alleviate the communication delay limit of federal learning, we innovatively integrate the attention mechanism, and the FedACNN can achieve ideal accuracy with a 50% reduction of communication rounds.

The rapid development in network technology has resulted in the proliferation of Internet of Things (IoT). This trend has led to a widespread utilization of decentralized data and distributed computing power. While machine learning can benefit from the massive amount of IoT data, privacy concerns and communication costs have caused data silos. Although the adoption of blockchain and federated learning technologies addresses the security issues related to collusion attacks and privacy leakage in data sharing, the “free-rider attacks” and “model poisoning attacks” in the federated learning process require auditing of the training models one by one. However, that increases the communication cost of the entire training process. Hence, to address the problem of increased communication cost due to node security verification in the blockchain-based federated learning process, we propose a communication cost optimization method based on security evaluation. By studying the verification mechanism for useless or malicious nodes, we also introduce a double-layer aggregation model into the federated learning process by combining the competing voting verification methods and aggregation algorithms. The experimental comparisons verify that the proposed model effectively reduces the communication cost of the node security verification in the blockchain-based federated learning process.

In social networks, some private information, such as the personal name, age gender, the number of friends, can be obtained by others. This paper defines a combination degree-neighborhood label matching attack model based on group maps obtained from multi-social networks. Based on the heuristic combination degree attack model, the target combination degree and neighborhood labels are used as the background knowledge of the attacker to obtain the candidate vertices set. The singularity of the sensitive label matching results will expose the sensitive information of the vertex being attacked. In order to solve this privacy attack, this paper proposes a group graph sensitive label generalization L diversity algorithm. This algorithm reduces the probability of sensitive labels being identified by designing a group map sensitive label generalization tree. According to the background knowledge, the number of sensitive labels in the candidate set and the number of sensitive labels obtained by matching are not less than L, so as to protect the sensitive information of the attacked target. The algorithm was evaluated by using three sets of data with different ratios. The experiment results show that the privacy protection algorithm effectively prevents sensitive label privacy attacks consisting of combination degree-domain label matching and better maintains the availability of graph data.

In this issue, we consider the location-aware targeted influence blocking maximization (LTIBM) problem, which plays a very important role in viral marketing and rumor control. LTIBM aims to find a set of positive seeds in a given social network to block the influence propagation of negative seeds over the targeted nodes located in a given region and having a preference on a given topic set as much as possible. We devise a simulation-based greedy algorithm based on monotone and submodular characteristics of influence function under the homogeneous independent cascade model. To improve the efficiency of the greedy algorithm, we propose LTIBM-H, a heuristic algorithm based on QT-tree and maximum influence arborescence (MIA). Experimental results show that the proposed LTIBM-H algorithm can achieve matching the blocking effect to the greedy algorithm and often performs better in terms of effectiveness than other baseline algorithms, while LTIBM-H is four orders of magnitude faster than the greedy algorithm.

More data and applications are moving to the cloud, which presents many new security risks. Malware is one of the most significant threats to cloud computing. In this paper, we explore to employ virtual machine introspection(VMI) and memory forensics analysis(MFA) techniques to detect malware running in guest virtual machines. Our scheme differs from existing malware detection methods based on virtualization technology in three aspects. First, this paper combines VMI with MFA to extract multiple type features in the guest virtual machine at the same time. Our scheme can effectively minimize the data acquisition overhead. Second,compared with single dynamic feature or multiple static feature detection methods, our data acquisition method employs dynamic multiple type features, and effectively promotes the ability of sophisticated malware detection. Finally,we use AdaBoost ensemble learning method and combination strategy of voting to improve the accuracy and generalization ability of the overall classifier. The experimental results based on a lot of real-world malware show that our scheme can achieve a detection accuracy of 0.9975. Our approach can improve virtual machines security, and further effectively enhance the security of cloud computing environment.

While mobile networks provide many opportunities for people, they face security problems huge enough that a firewall is essential. The firewall in mobile networks offers a secure intranet through which all traffic is handled and processed. Furthermore, due to the limited resources in mobile networks, the firewall execution can impact the quality of communication between the intranet and the Internet. In this paper, a performance evaluation mathematical model for firewall system of mobile networks is developed using queuing theory for a multihierarchy firewall with multiple concurrent services. In addition, the throughput and the package loss rate are employed as performance evaluation indicators, and discrete‐event simulated experiments are conducted for further verification. Lastly, experimental results are compared to theoretically obtained values to identify a resource allocation scheme that provides optimal firewall performance and can offer a better quality of service (QoS) in mobile networks.

Human behavior data is the basis of behavior analysis, and usually we need to collect large quantities of data before analysis. Most existing data collection methods are labor intensive works in which the volunteers need to be asked to behave naturally under the monitoring of researchers. Identity identification can be used in passive data collection of human behavior analysis systems in big data. Previous researches show the sensing potential of WiFi signals in a device-free passive manner. It is confirmed that human's gait is unique from each other like fingerprint and iris. As a result, researchers start to explore the ability of WiFi in human identification. However, the identification accuracy of existing approaches is not satisfactory in practice. In this paper, we present Wii, a device-free WiFi-based Identity Identification approach utilizing human's gait based on Channel State Information (CSI) of WiFi signals. Principle Component Analysis (PCA) and low pass filter are applied to remove the noises in the signals. We then extract entities' gait features from both time and frequency domain. Based on these features, Wii realizes identity identification through a Support Vector Machine (SVM) with Radial Basis Function (RBF) kernel. It is implemented using commercial WiFi devices and evaluated in a typical indoor scenario. The results indicate that Wii achieves high identification accuracy with low computational cost and has the potential to work in human behavior analysis systems.

Web Security Gateway (WSG) is a new type of network security product that maintains the security of trusted networks. In this paper, a WSG model for evaluating WSG performance is presented. This paper advances discussion of previous studies on series services under multiple service windows. The proposed model consists of a two-stage queuing system. The first stage is a network layer simulation. The second stage is thus similar to a parallel hyper-Erlang distribution model. The results of a simulation test verified the feasibility and performance of the proposed model.

Occupancy information is one of the most important privacy issues of a home. Unfortunately, an attacker is able to detect occupancy from smart meter data. The current battery-based load hiding (BLH) methods cannot solve this problem. To thwart occupancy detection attacks, we propose a framework of battery-based schemes to prevent occupancy detection (BPOD). BPOD monitors the power consumption of a home and detects the occupancy in real time. According to the detection result, BPOD modifies those statistical metrics of power consumption, which highly correlate with the occupancy by charging or discharging a battery, creating a delusion that the home is always occupied. We evaluate BPOD in a simulation using several real-world smart meter datasets. Our experiment results show that BPOD effectively prevents the threshold-based and classifier-based occupancy detection attacks. Furthermore, BPOD is also able to prevent nonintrusive appliance load monitoring attacks (NILM) as a side-effect of thwarting detection attacks.

Intrusion detection plays a critical role in security of people's possessions. Approaches such as video-based, infrared-based, RFID, UWB, etc. can provide satisfying detection accuracy. However, they all require specialized hardware deployment and strict using conditions which hinder their wide deployment. Beyond communication, WLANs can also act as generalized sensor networks and there are several researches working on motion detection via WLAN due to its advantages in deployment flexibility, coverage, and cost efficiency. Nevertheless, they are unsuitable for intrusion detection as none of them can accurately detect human motion when the moving speed is very slow. This paper proposes SIED as an accurate method for Speed Independent device-free Entity Detection which is suitable for intrusion detection even when the entity's moving speed is very slow. The influence becomes much smaller when the entity is moving with a very slow speed. Previous methods have the limitations in that their performance downgrades sharply when the entity's moving speed is very slow. Recently, it has been shown that Channel State Information (CSI) at PHY layer of wireless network has the potential to detect moving entities more accurately. In this paper we leverage CSI of 802.11n wireless network and probability technique to detect entities of different moving speeds. SIED captures the variance of variances of amplitudes of each CSI subcarrier, and combines Hidden Markov Model (HMM) to make entity detection a probability problem. We implement SIED using commercial WiFi devices and evaluate our method using two typical testbeds and show that SIED can achieve an average detection accuracy of greater than 98% under different entity moving speed.