Most adversarial attacks on point clouds perturb a large number of points, causing widespread geometric changes and limiting applicability in real-world scenarios. While recent works explore sparse attacks by modifying only a few points, such approaches often struggle to maintain effectiveness due to the limited influence of individual perturbations. In this paper, we propose SCP, a sparse and cooperative perturbation framework that selects and leverages a compact subset of points whose joint perturbations produce amplified adversarial effects. Specifically, SCP identifies the subset where the misclassification loss is locally convex with respect to their joint perturbations, determined by checking the positive-definiteness of the corresponding Hessian block. The selected subset is then optimized to generate high-impact adversarial examples with minimal modifications. Extensive experiments show that SCP achieves 100% attack success rates, surpassing state-of-the-art sparse attacks, and delivers superior imperceptibility to dense attacks with far fewer modifications.

Mixture-of-Experts (MoE) architecture with experts parallelism scales LLMs efficiently by activating only a subset of experts per input, avoiding proportional training costs. However, the intensive and heterogeneous communication substantially hinders the efficiency and scalability of MoE training in the resource-constrained scenario. Existing communication compression techniques fall short in MoE training due to: (i) Intensive training amplifies compression overhead, compromising training efficiency; (ii) Accumulated compression errors propagate through the network, degrading training quality. In this paper, we propose RCMoE, a communication-efficient Random Compression framework for MoE training with two core modules: (1) Local-Stochastic Quantization compresses the all-to-all communication by stochastically quantizing each row of the expert's intermediate computing results in parallel, effectively improving the compression efficiency and reducing compression error; (2) Probabilistic Thresholding Sparsification compresses the all-reduce communication by probabilistically sampling large gradients at high probability, thereby reducing the computational complexity and maintaining the convergence efficiency. Experiments on four typical MoE training tasks prove that RCMoE achieves higher 5.9x-8.1x total communication compression ratios and 1.3x-10.1x training speedup compared with the state-of-the-art compression techniques while maintaining the MoE training accuracy.

Adversarial training is often modeled as a two-player zero-sum game, relying on strong assumptions that limit its practical guidance. In this paper, we instead analyze the interactions between training samples and show that even the fundamental objective—minimizing training loss—may not converge. To address this, we propose AT-Field, an adversarial training framework guided by sample-wise game-theoretic relationships. Specifically, we prove that training samples across different batches can form a none-potential game, where gradient descent induces cyclic behaviors, preventing convergence. By strategically searching and grouping these samples within the same batch, AT-Field transforms none-potential games into exact potential games, which are more effectively optimized using gradient-based methods. Experiments demonstrate that AT-Field integrates seamlessly with existing adversarial training techniques, enhancing both accuracy and robustness.

For complex networks such as the Web, the Network Dismantling (ND) problem, which asks for the minimum-cost removal of nodes that destroys the giant connected component in the network, is significant in system robustness and misinformation containment. In this paper, we propose a heuristic algorithm, IG+, which is based on reverse dismantling and incorporates novel optimizations. Besides, we design two dynamic algorithms, CCRT-ins and CCRT-rem, employing tree-like indexes to update dismantling results efficiently. Experiments show that our methods outperform state-of-the-art approaches in both effectiveness and efficiency, and can dismantle 10-million-scale networks at arbitrary granularity in a few minutes.

Timely detection of anomalous interactions between different entities is crucial for the security and stability of Web-related systems, which are often modeled as streaming graphs. A major issue in existing anomaly detection methods is the effectiveness and the corresponding interpretability. In this paper, we introduce a novel evaluation metric, namely Interaction Willingness, to measure the propensity for entity interactions. Based on this metric, we design two efficient anomaly detection algorithms, AnoD and AnoC, tailored for real-time detection of two prevalent types of anomalous interaction behaviors (dense-type and counter-type), respectively. Notably, we adopt and extend the High-Order Count-Min Sketch for the implicit storage of streaming graphs, and ensure that both the space cost and the time cost of AnoD and AnoC for processing each new streaming edge remain constant and user-controllable, with estimation guarantees. Experimental evaluations on 6 real-world datasets demonstrate that the proposed evaluation metric and the associated algorithms achieve superior detection performance with clear interpretability for detecting anomalous interactions.

Encrypted traffic classification (ETC) serves as a pivotal research for network measurement and Quality of Service (Qos) management. However, most current approaches rely on statically frozen parameters after training, which remain vulnerable to performance degradation under real-world distribution shifts caused by dynamic network conditions, protocol evolution, and adversarial obfuscation techniques. To tackle this issue, we propose NetTTT, an online self-adaptation framework that, for the first time, performs continuous representation refinement directly at inference time without retraining. NetTTT leverages a hierarchical feature encoder that constructs structured dual-modal representations from raw traffic sessions, coupled with a pre-trained generic representation extractor to powerfully capture transferable contextual features, and incorporates a lightweight test-time adapter, which performs unsupervised, layer-wise parameter optimization during inference to dynamically align traffic representations with test-time traffic distributions, followed by a compact classifier that efficiently maps adapted representations to traffic categories. Experimental results reveal the superiority of NetTTT across diverse ETC tasks and few-shot scenarios with accuracy improvements ranging from 0.08% to 9.34% over state-of-the-art baselines. By enabling dynamic adaptation without labeled data or full model retraining, NetTTT bridges the gap between static pre-trained models and evolving encrypted traffic, highlighting its potential to provide a practical solution for real-world scenarios.

Federated Learning (FL) has emerged as a promising paradigm for collaborative model training without direct data sharing. However, existing frameworks struggle to simultaneously ensure strong privacy protection and robustness against adversarial participants. Although numerous robust aggregation algorithms can mitigate Byzantine attacks and various privacy-preserving approaches have been developed, achieving both objectives within a unified and efficient framework remains challenging. To address this compatibility gap, we propose a generic privacy-preserving FL framework that integrates robust aggregation with cryptographic privacy guarantees through secure two-party computation under a dual-server architecture. In our design, two non-colluding servers collaboratively execute the aggregation process, ensuring that sensitive client updates remain confidential while enabling the application of diverse robust aggregation strategies. The framework is flexible and supports a wide range of aggregation algorithms, such as Multi-Krum, Median, and Mean-based methods, under a consistent security model. We implement the proposed system and conduct extensive experiments on multiple benchmark datasets. Experimental results demonstrate that our framework preserves the effectiveness of existing robust aggregation algorithms while maintaining acceptable runtime and communication overhead compared with standard FL baselines. These findings confirm that the proposed approach provides a practical and balanced solution to the dual challenges of privacy protection and Byzantine robustness, offering a versatile foundation for secure and trustworthy FL in adversarial environments.

Transformer-based foundation models are becoming deeper and larger. For fast training, their billions of parameters (tensors) are split onto parallel tasks running on many modern yet expensive accelerators. To amortize the huge hardware investments, it is cost-effective to share the aggregated resources among multi-tenants. However, resource contention yields the heavy straggling problem. Existing works feature contributions for the traditional data parallelism. They cannot work well for the new tensor parallelism, due to the dependency among split tensors. This paper is the first attempt on accelerating heterogeneous tensor parallelism. We summarize specific challenges, including the very frequent synchronizations and the heavy tensor computation workloads. Our solution is to resize dimensions of parameters on demand, to quickly and dynamically balance workloads. The accuracy loss is reduced through priority resizing. We also migrate workloads between tasks, without any loss of accuracy. The most efficient communication primitives are selected and then scheduled in a non-redundant manner, to reduce the runtime latency. Our final hybrid solution is built on top of resizing and migration. By studying the tradeoff between accuracy and efficiency, it can smartly hit the “sweet spot”. Extensive experiments validate the effectiveness of our proposals.

This paper investigates the differential privacy consensus problem for a class of multiagent systems under dynamic topologies. To meet the requirements of power consumption, a random communication strategy is proposed in which each agent sends data to its neighbors with different probabilities. For analyzing the effect of time-varying topology and coupling strength among agents on system performance, a necessary and sufficient condition for almost sure convergence of differential privacy consensus systems is established. Furthermore, the convergence rate and convergence accuracy of the system are also studied. By formulating the communication costs and topological characteristics as a constrained problem, a convex optimization algorithm for fast convergence of the differential privacy consensus system is proposed. In addition, the differential privacy of the agents is analyzed, and the optimal noise parameters that achieve a trade-off between convergence accuracy and privacy levels are derived. A numerical simulation is presented to demonstrate the effectiveness of the developed approach.

Traditional threat detection systems fail in dynamic cloud-native environments, primarily due to their inability to resolve the inherent trade-off between high-precision detection of known attacks and high-recall discovery of unknown threats. We proposeD3-Guard, a novel framework whose core innovation lies in a dynamic decision fusion engine grounded in Bayesian decision-theoretic principles. The engine computes a similarity score from incoming requests to serve as a real-time contextual signal for the Bayesian framework. It then uses this signal directly as a dynamic weight to adaptively integrate the outputs of two complementary expert models: a high-precision LSTM-Attention for known patterns and a high-recall Isolation Forest for unknown threats. This synergy, reinforced by an adaptive update layer to mitigate concept drift, achieves F1-scores of 99.39% and 96.97% on the CSIC 2010 and ATRDF 2023 datasets, respectively. Performance tests in a production-like environment further validate its practical viability, demonstrating that this high accuracy is achieved with a justifiable overhead. An ablation study highlights the pivotal role of our fusion engine: disabling it results in a sharp decline in F1-score on the ATRDF 2023 dataset, from 96.97% to 11.47%, demonstrating its critical contribution to robust, dual-model collaboration.

Since the proliferation of image manipulation methods, effective image manipulation localization (IML) in scenarios with post-processing operations gradually becomes a core challenge. For a long time, IML either relies on strongly semantically related features, resulting in semantic relevance bias in the localization results, or only uses a single semantic-agnostic space feature, which is unable to maintain effective localization capabilities after image post-processing operations. Inspired by this, we propose a novel mixture semantic-agnostic manipulation trace robust localization network (MSFT-Net), which specifically utilizes mixture semantic-agnostic information to achieve effective and robust IML. The MSFT-Net introduces two new modules, the mixture shared manipulation trace enhancement module (MISE) and the Multiscale Feature Association Module (FAM). MISE dynamically links multiple semantic-agnostic feature extractors using a sparsity-enhanced mixture of shared experts, enabling the extraction of diverse manipulation features for accurate localization. Furthermore, keeping the high resolution of the localization features is very important in the mask prediction stage. Therefore, FAM outputs high-resolution fused manipulation features by using the correlation of features at the same level and the spatial context information from different levels. This further improves the effectiveness of IML in post-processing scenarios. Comprehensive experiments on five datasets demonstrate that our model significantly improves both in localization accuracy (average F1 score and IoU increasing by over 9.9% and 4.0%) and robustness. The codes will be made available.

In the battlefield of cyberspace, sophisticated attackers often operate by following meticulously designed cyber kill chains, enabling them to maintain a persistent presence within victim systems while evading conventional detection mechanisms. Traditional honeypot-based deception defenses aim to uncover such threats by luring attackers into exposing their malicious activities through decoy systems. However, advanced attackers are frequently able to identify and avoid these traps, making it increasingly challenging to detect and engage them effectively. To overcome this challenge, this study proposes a novel defensive paradigm named as the safeguard mode, which emphasizes the covert identification of attackers rather than solely preventing initial breaches. By proactively recognizing potential threats in a hidden manner, victim systems can be better protected through early threat intelligence. Based on the propsoed safeguard mode concept, we propose$Honey^{4}$, abbreviated as$H^{4}$, a comprehensive framework designed to systematically entrap advanced threats.$H^{4}$comprises four core components: Honeypoint, Honeyproxy, Honeytrace, and Honeycenter, which work in concert to deceive, monitor, and analyze attacker behavior. Furthermore, we explore how Artificial Intelligence Generated Content (AIGC) techniques can enhance$H^{4}$'s capabilities, particularly as attackers themselves begin to leverage AI-driven tactics. The practical efficacy of the proposed safeguard mode and the$H^{4}$framework has been validated through being deployed in real scenarios including the 19th Asian Games and the Canton Fairs, and$H^{4}$has successfully captured a significant number of threatening IP addresses and malicious behavioral patterns, generating actionable cyber threat intelligence that fundamentally safeguards system defense.

Semantic attacks against programmable logic controllers (PLCs) can manipulate industrial control systems (ICS) and cause severe physical disruptions. Yet, existing researches remain target-agnostic: they fail to determine which PLC tags or physical process are most critical when exploited. To address this gap, we introduce Pipe4Stux, a semantic attack framework that generates effective PLC attacks by leveraging key tags and attack semantics. Pipe4Stux first constructs a lexical analyzer and a data dependency subgraph from PLC control logic to capture the relationships among variables and operations. It then identifies key tags, derives semantic features of attacks, and formulates process-aware attack tactic. Experimental results on real ICS scenarios demonstrate that Pipe4Stux enables destructive and reproducible semantic attacks, revealing the potential impact of compromising key tags during specific physical states. In addition, Pipe4Stux provides the first systematic and objective methodology for semantic attack synthesis, offering practical guidance for defense. The framework can also support continuous monitoring by designating key tags as alarm variables within supervisory control and data acquisition (SCADA) systems such as WINCC.

The wide deployment of Face Recognition (FR) systems poses privacy risks. One countermeasure is adversarial attack, deceiving unauthorized malicious FR, but it also disrupts regular identity verification of trusted authorizers, exacerbating the potential threat of identity impersonation. To address this, we propose the first double identity protection scheme based on traceable adversarial watermarking, termed DIP-Watermark. DIP-Watermark employs a one-time watermark embedding to deceive unauthorized FR models and allows authorizers to perform identity verification by extracting the watermark. Specifically, we propose an information-guided adversarial attack against FR models. The encoder embeds an identity-specific watermark into the deep feature space of the carrier, guiding recognizable features of the image to deviate from the source identity. We further adopt a collaborative meta-optimization strategy compatible with sub-tasks, which regularizes the joint optimization direction of the encoder and decoder. This strategy enhances the representation of universal carrier features, mitigating multi-objective optimization conflicts in watermarking. Extensive experiments on two large-scale facial datasets demonstrate that DIP-Watermark achieves significant attack success rates and traceability accuracy on state-of-the-art FR models and commercial APIs. It also exhibits superior robustness against a wide range of real-world simulated distortions, outperforming existing privacy protection methods based on adversarial attacks, deep watermarking, or their simple combination. Our work potentially opens up new insights into proactive protection for FR privacy.

This paper investigates the security issues of average consensus networks under false data injection attacks. Based on the intention of the adversary, the attacks to be analyzed are categorized into two types. The first type aims to rapidly diverge the states of the agents to infinity, while the second type seeks to mislead the states towards the adversary's desired values. A necessary and sufficient condition for Type-I attacks to bypass the anomaly detector is first provided, along with an algorithm for generating false data sequences. Subsequently, for Type-II attacks, the closed-form expression of the optimal attack is derived using dynamic programming. To address these vulnerabilities, a watermarking-based data transmission strategy is proposed by resorting to cryptographic pseudo-random sequences. The influence of watermarking parameters on detection performance across different attack scenarios is analyzed. It is demonstrated that the proposed strategy can effectively assist the anomaly detector in identifying stealthy attacks or mitigate its impact on the consensus networks by adjusting the watermarking parameters. A numerical example is provided to demonstrate the validity of the developed results.

This paper investigates the resilience of autonomous vehicle (AV) systems against denial-of-service (DoS) attacks in networked environments. While existing fuzzy logic and event-triggered control strategies have been explored, they often fail to address the combined impacts of DoS-induced network congestion, data compression errors, and the lack of real-time adaptive learning to maintain stability. To address this gap, this work introduces a unified, resilient fuzzy logic control framework. First, a Performance Error Estimation (PEE) framework is developed to quantify control performance degradation under cyber threats, informing the design of the resilient controller. Second, an efficient data compression scheme is designed to mitigate DoS-induced network overload, ensuring reliable real-time data transmission under constrained bandwidth. Third, an Intelligent Event-Triggered Fuzzy Controller (IETFC) is proposed, which adaptively optimizes its triggering threshold via a mini-batch machine learning algorithm, effectively balancing communication efficiency with system robustness. Validation on the CarSim–Simulink platform demonstrates that the proposed framework successfully mitigates DoS-induced performance degradation, enhances triggering sparsity, and improves computational efficiency. By embedding learning-based adaptation into fuzzy control, this work provides a scalable and secure solution for AV systems operating under adversarial conditions.

Advanced Persistent Threats (APTs) have become one of the most prominent cybersecurity risks globally. The external network-facing (ENF) services (e.g., e-commerce platforms) within a system are particularly vulnerable, as they are directly exposed to the Internet and often serve as the primary targets for attackers. By deploying deception resources to protect these ENF services, defenders can detect threats early, block potential attacks, and enhance overall system resilience. However, most existing studies on cyber deception strategies assume simultaneous moves by both attacker and defender. Furthermore, few works have considered the evolution of the system state resulting from APT attacks on the ENF services. To address these limitations, this paper proposes a Cyber Deception Stackelberg Markov Game (CDSMG) for protecting ENF services, which dynamically captures state transitions and accurately characterizes the strategic interactions between defenders and APT attackers. In CDSMG, the defender acts as the leader, who proactively selects a subset of services to deploy the deception resources based on the current system state, while the APT attacker plays as the follower, making a best response which incorporates the defender’s policy into its own strategy. To overcome the challenge of the combinatorial optimization problem of selecting a subset of services, we propose a revised version of the PPO algorithm by using no-replacement sampling to select multiple services at once, thereby significantly reducing the action space size. Finally, experimental results demonstrate that our approach effectively defends against APT attacks. It not only outperforms several baseline methods but also exhibits better scalability and robustness under varied model parameter settings.

Accurately mapping unstructured cyber threat intelligence (CTI) text to standardized attack technique remains a fundamental yet challenging task, especially under zero-shot conditions where no labeled instances from the target CTI corpus are available. Existing methods largely rely on heuristic rules or supervised classifiers,and therefore often generalize poorly to unseen, long-tail, or evolving techniques. To overcome these limitations, we proposeRA-CTI—a retrieval-augmented Large Language Model (LLM) framework for contextual technique identification, which instantiates a generalRetrieve–Expand–Inferparadigm for aligning free-form text with ontology-driven label inventories. RA-CTI reformulates MITRE ATT&CK Techniques Identification as semantic evidence discovery and reference-guided reasoning through three coordinated stages: (1) dense semantic retrieval for efficient candidate acquisition, (2) query expansion and evidence aggregation to improve recall and coverage, and (3) reference-aware inference that conditions on explicit MITRE ATT&CK technique definitions for fine-grained semantic alignment. An optional weakly supervised retriever adaptation module further improves domain robustness without relying on manual annotations. Extensive experiments on multiple public CTI datasets show that RA-CTI consistently outperforms state-of-the-art baselines in zero-shot multi-label identification, achieving better efficiency–accuracy trade-offs and strong generalization across LLM backbones. Code is available athttps://anonymous.4open.science/r/APT9-RA-CTI.

While face recognition (FR) models have brought remarkable convenience in face verification and identification, they also pose substantial privacy risks to the public. Existing facial privacy protection schemes usually adopt adversarial examples to disrupt face verification of FR models. However, these schemes often suffer from weak transferability against black-box FR models and permanently damage the identifiable information that cannot fulfill the requirements of authorized operations such as forensics and authentication. To address these limitations, we proposeErasableMask, a robust and erasable privacy protection scheme against black-box FR models. Specifically, via rethinking the inherent relationship between surrogate FR models, ErasableMask introduces a novel meta-auxiliary attack, which boosts black-box transferability by learning more general features in a stable and balancing optimization strategy. It also offers a perturbation erasion mechanism that supports the erasion of semantic perturbations in protected face without degrading image quality. To further improve performance, ErasableMask employs a curriculum learning strategy to mitigate optimization conflicts between adversarial attack and perturbation erasion. Extensive experiments on the CelebA-HQ and FFHQ datasets demonstrate that ErasableMask achieves the state-of-the-art performance in transferability, achieving over72%mean confidence in commercial FR systems. Moreover, ErasableMask also exhibits outstanding perturbation erasion performance, achieving over90%erasion success rate.

Network scanning is crucial for both network management and cybersecurity. However, due to the vast address space of IPv6, brute-force scanning is infeasible. Seed-based target generation algorithms have recently attracted considerable research attention. However, existing target generation algorithms lack a deeper exploration of patterns, leading to poor capture of dense regions and consequently low hitrate. To address this issue, we propose 6Global, a dynamic IPv6 active address scanning method assisted by global perspective. 6Global first performs rapid clustering of seed addresses based on their descriptive attributes. Then, for each cluster, patterns are generated in a bottom-up manner based on entropy, using subranges to represent patterns and resulting in denser patterns. Finally, dynamic scanning is conducted using these patterns. During scanning, the reward of each pattern is dynamically adjusted based on its active density and global statistics, which enhances the capability in capturing dense regions. Experimental results on six seed datasets show that 6Global overall outperforms seven baseline methods and demonstrates significant advantages across multiple datasets.

Imperceptible adversarial attacks on 3D point clouds rely on effective constraints. While manifold constraints have notable advantages over Euclidean ones, the global parameterization used in current methods often fails to fully preserve manifold properties. In this paper, we propose to constrain lattice-based barycentric coordinates during attacks from a local parametric perspective to ensure imperceptibility. Specifically, we utilize a permutohedral lattice to partition point clouds into multiple cells, and then extract barycentric coordinates for each point within these cells, forming a local parametric representation of the point clouds. By enforcing local parametric constraints that minimize the displacement of barycentric coordinates, we largely preserve the manifold properties, ultimately leading to improved imperceptibility. Extensive experiments validate that integrating these local parametric constraints into conventional adversarial attacks yields superior imperceptibility, outperforming state-of-the-art methods.

Deep neural networks (DNNs) often exhibit out-of-distribution (OOD) overconfidence, producing overly confident predictions on OOD samples. We attribute this issue to the inherent over-complexity of DNNs and investigate two key aspects: capacity and nonlinearity. First, we demonstrate that reducing model capacity through knowledge distillation can effectively mitigate OOD overconfidence. Second, we show that selectively reducing nonlinearity by removing ReLU operations further alleviates the issue. Building on these findings, we present a practical guide to model simplification, combining both strategies to significantly reduce OOD overconfidence. Extensive experiments validate the effectiveness of this approach in mitigating OOD overconfidence and demonstrate its superiority over state-of-the-art methods. Additionally, our simplification strategies can be combined with existing OOD detection techniques to further enhance OOD detection performance.

Mixture-of-Experts (MoE) model parallelism is prevalent in training Large Language Models (e.g., ChatGPT). However, the intensive all-to-all collective communication of the MoE layer’s intermediate computing results substantially degrades MoE training efficiency. In this paper, we propose BirdMoE, a novel load-aware communication compression technique with Bi-random quantization for MoE training with two core modules. Specifically, BirdMoE employs a lightweight Random Quantization (RQ) with expectation invariance property to efficiently map the floating-point intermediate computing results into integers while maintaining the MoE training quality. Additionally, BirdMoE utilizes a Mixed Precision (MP) strategy to dynamically balance the communication loads among expert nodes, significantly improving all-to-all communication efficiency for the MoE training system. Experiments on four typical MoE training tasks demonstrate that BirdMoE achieves higher $4.06 \times- 10.44 \times$ total communication compression ratios and $1.18 \times-5.27 \times$ training speedup compared with the state-of-the-art compression techniques while maintaining the MoE training quality.

In masked language model (MLM) based attacks, candidate adversarial examples are flexibly generated according to the context, but how to balance imperceptibility and success rate of attacks remains a challenge. To pursue imperceptibility, external semantic constraints are imposed on candidate word generation, whereby the search space is restricted and the success rate of attacks drops. To address this issue, a semantically improved adversarial attack denoted by SAM-CP is proposed by generating high-quality candidate words satisfying the semantic constraints. In particular, linguistic constraints are adopted so that high-quality semantic candidates can be generated as fine-tuning labels instead of manual annotation. Extensive experimental results on three open-source datasets demonstrate that SAM-CP significantly improves the semantic consistency of generated adversarial samples by adopting different MLMs, respectively. Without compromising text quality, the ability to use SAM-CP to deceive state-of-the-art text classifiers is evaluated. Visit https://github.com/HugoTianX/SAM-CP for details and codes.

Adversarial attacks on point clouds are crucial for assessing and improving the adversarial robustness of 3D deep learning models. Traditional solutions strictly limit point displacement during attacks, making it challenging to balance imperceptibility with adversarial effectiveness. In this paper, we attribute the inadequate imperceptibility of adversarial attacks on point clouds to deviations from the underlying surface. To address this, we introduce a novel point-to-surface (P2S) field that adjusts adversarial perturbation directions by dragging points back to their original underlying surface. Specifically, we use a denoising network to learn the gradient field of the logarithmic density function encoding the shape’s surface, and apply a distance-aware adjustment to perturbation directions during attacks, thereby enhancing imperceptibility. Extensive experiments show that adversarial attacks guided by our P2S field are more imperceptible, outperforming state-of-the-art methods.

Cross-domain recommendation (CDR) has demon-strated to be an effective solution for alleviating the user cold-start issue. By leveraging rich user-item interactions available in a richly informative source domain, CDR could improve the recommendation performance for cold-start users in the target domain. Previous CDR approaches mostly adhere the Embedding and Mapping (EMCDR) paradigm, which learns a user-shared mapping function to transfer users' preference from the source domain to the target domain, neglecting users' personalized preference. Recent CDR approaches further leverage the meta-learning paradigm, considering the CDR task for each user independently and learning user-specific mapping functions for each user. However, they mostly learn representations for each user individually, which ignores the common preference between different users, neglecting valuable information for CDR. In addition, all these approaches usually summarize the user's preference into an overall representation, which can hardly capture the user's multi-interest preference. To this end, we propose a personalized multi-interest modeling framework for CDR to cold-start users, termed as NF-NPCDR. Specifically, we propose a personalized preference encoder that enhances the neural process (NP) with the normalizing flow (NF) to convert the Gaussian (unimodal) distribution to a multimodal distribution, providing a novel way to capture the user's personalized multi-interest preference. Then, we propose a common preference encoder with a preference pool to capture the common preference between different users. Furthermore, we introduce a stochastic adaptive decoder to incorporate both the personalized and common preference for cold-start users, adaptively modulating both preference for better recommendation. Experimental evalu-ations demonstrate that NF-NPCDR outperforms previous SOTA approaches in five benchmark CDR scenarios.

The Cyber Threat Intelligence (CTI) marketplace is an emerging platform for CTI service requesters to countermeasure advanced cyber attacks, where CTI service providers are employed on payment. To create a trustworthy CTI marketplace environment, consortium-blockchain-based CTI service platforms have been widely proposed, where the blockchain system becomes the third role, crucially impacting the CTI service quality. How to sustainably promote CTI service quality in this tripartite marketplace is a challenging issue, which has not been well investigated in the literature. In this study, we propose a two-stage tripartite dynamic game-model-based incentive mechanism, where the participation incentives of the three parties are promoted under the constraints of Individual Rationality (IR) and Incentive Compatibility (IC). The sustainability of CTI service is quantitatively formalized through the CTI market demand, which impacts the future profits of the three parties. The Nash equilibrium of the proposed incentive mechanism is solved, where the CTI requester offers an optimal price to achieve effective defense against cyber attacks, and the blockchain platform and CTI service providers collaboratively contribute high-quality CTI services. Empirical experimental results show that the higher the quality of CTI services provided in the marketplace, the greater the market demand for CTI, resulting in a sustainable CTI marketplace.

The proliferation of Internet of Things (IoT) devices, now numbering in the tens of billions, has exposed new attack surfaces due to their heterogeneous network architectures and vast numbers of distributed endpoints. The offensive-defensive dynamics of Advanced Persistent Threats (APTs) in IoT environments exhibit unique complexities, including enhanced stealth capabilities and prolonged attack lifecycles. This paper introduces APT-KG2QA, a knowledge graph-driven framework for generating specialized question-answering datasets. This methodology addresses two critical challenges in deploying large language models (LLMs) for cybersecurity applications: mitigating inherent biases in attack behavior recognition and overcoming logical reasoning deficits stemming from limited access to high-quality, domain-specific training data. The methodology utilizes a systematic conversion mechanism to translate the APT KG data into a hierarchical instruction template framework, which incorporates a hybrid prompt template engine with adversarial augmentation modules to produce domain-adaptive fine-tuning data. Low-Rank Adaptation (LoRA) technique facilitates parameter-efficient fine-tuning for four basic models. Experimental results indicate that the models enriched with domain knowledge achieve average increases of 577.7% in BLEU-4 and 623.7% in ROUGE-L measures, showing significant enhancements noted in output correctness, relevance, and comprehensibility. This study explores cross-modal integration pathways between KGs and LLMs, confirming the effectiveness of structured knowledge infusion in improving cybersecurity analysis capabilities, thus offering methodological guidance for developing intelligent security defense systems.

The Internet of Things (IoT) has seen widespread adoption across various industries, enabling cross-domain collaboration among devices. As IoT platforms that allows users to manage and control these IoT devices become more prevalent, the security of IoT platforms has come under increasing scrutiny. Cryptographic authentication (CA) has long been central to authentication but poses significant security risks, such as account takeovers due to credential disclosure. Unfortunately, CA remains the most popular authentication scheme for IoT platforms, making them vulnerable to phishing attacks. Current enhancement methods, including challenge-response protocols and multifactor authentication, are still susceptible to phishing and face usability issues, particularly in cross-domain collaboration systems where seamless access is crucial. This article introduces Context-Auth, a context-aware authentication scheme designed to defend against phishing attacks without requiring tamper-resistant hardware. By incorporating contextual features of authentication and access behaviors into access credentials, Context-Auth detects inconsistencies and strengthens security. Our scheme addresses cross-domain verification challenges in federated identity scenarios, ensuring high usability across various devices without additional requirement from users. We conducted comprehensive evaluations and security analyses to ensure that Context-Auth remains secure against potential threats. The experimental results demonstrate that Context-Auth effectively defends against phishing attacks launched by real-world phishing kits.

In Internet of Things (IoT) environments, anti-honeypot attacks exploit device fingerprinting and protocol analysis to undermine traditional honeypot defenses. To this end, we introduce an innovative camouflage defense strategy that dynamically disguises real IoT devices as honeypots to mislead attackers and protect critical resources. We formalize this adversarial interaction using a multi-stage signaling game, thereby reversing the information asymmetry that typically favors attackers. Subsequently, we theoretically derive the Perfect Bayesian Nash Equilibrium (PBNE) under conditions of incomplete information, thereby confirming the existence of mixed-strategy equilibria. Practically, we develop a signaling game-based multi-agent proximal policy optimization (SG-MAPPO) algorithm by integrating LSTM networks and Bayesian updates, overcoming challenges in long-term strategy optimization and parameter convergence in dynamic IoT contexts. Experiments demonstrate that SG-MAPPO outperforms existing methods in terms of training efficiency and stability, effectively reducing attacker success rates and providing a proactive defense solution for IoT security.

As a key technology in the Industrial Internet of Things (IIoT), the combination of digital twins (DTs) and mobile-edge computing (MEC) facilitates edge intelligence in B5G, while the application of generative artificial intelligence (GenAI) further enhances edge intelligence in resource allocation. However, the DT-enabled MEC system in IIoT faces the challenges of the low latency and high reliability demands, especially with the massive increase in machine-type communication devices and limited wireless and computing resources. To reduce overall delay and ensure high-reliability communication among machine devices (MDs), we propose an optimization problem for joint wireless and MEC computation resource allocation (JWMC-RA), which is shown to be NP-hard and intractable. Thereafter, the original problem is decomposed into two stages, i.e., machine-to-machine (M2M) links clustering, and MEC computing resource management. In the step of M2M links clustering, a heuristic clustering scheme via spectrum radius (HCS-SR) is presented for reducing interference of MDs which uses GenAI and graph theory. In the step of MEC computing resource management, the initial optimization problem is transformed to a convex problem, then, the optimal task offloading ratios of MD and MEC computing resource allocation are obtained. Finally, simulations show that the JWMC-RA scheme can reduce the overall delay and ensure the communication reliability.

With the increasing complexity of urban transportation systems and the growing demand for dynamic, real-time responsiveness, Time-Dependent Minimum Travel Time Queries (TD-MTTQs) in time-dependent road networks have become a core challenge in intelligent transportation system research. To address the trade-offs between preprocessing complexity and query efficiency in existing index-based methods for large-scale road network applications, this paper proposes a 4-hop index method, TD-TNR-CH. The core methodology involves establishing local indexes from each node to its nearest critical nodes (named transit nodes) through strategic critical node selection, while simultaneously constructing query tables associated with candidate sets between these critical nodes. This architecture enables rapid computation of medium-to-long distance queries through efficient index lookups, while ensuring high responsiveness for short-distance queries via TCH-based local searches. Extensive experimental results on large-scale real-world road networks demonstrate that our method exhibits superior scalability, achieving query efficiency of up to 103 times that of the fastest existing algorithms. Furthermore, it shows exceptional stability across queries of varying distances. Additionally, leveraging its parallelized architecture, TD-TNR-CH requires only approximately 30 minutes of preprocessing time for large-scale networks, significantly outperforming comparable methods in terms of preprocessing efficiency.

Log data is a vital resource for capturing system events and states. With the increasing complexity and widespread adoption of modern software systems and IoT devices, the daily volume of log generation has surged to tens of petabytes, leading to significant collection and storage costs. To address this challenge, lossless log compression has emerged as an effective solution, enabling substantial resource savings without compromising log information. In this paper, we first conduct a characterization study on extensive public log datasets and identify four key observations. Building on these insights, we propose LogLite, a lightweight, plug-and-play, streaming lossless compression algorithm designed to handle both TEXT and JSON logs throughout their life cycle. LogLite requires no predefined rules or pre-training and is inherently adaptable to evolving log structures. Our evaluation shows that, compared to state-of-the-art baselines, LogLite achieves Pareto optimality in most scenarios, delivering an average improvement of up to 67.8% in compression ratio and up to 2.7X in compression speed.

This study examines denial-of-service (DoS) attacks on networked control systems (NCSs) caused by TCP/IP protocol vulnerabilities. It conducts a comprehensive analysis of hacker tactics to uncover vulnerability exploitation techniques. The research reformulates the performance error estimation (PEE) problem, framing it as a search for ellipsoid constraints within a P-dependent set. It introduces the Higher-Order Weight Method (HOWM) to optimize sampling intervals, leveraging the unique properties of zero-order-hold sampling. The study enhances Lyapunov-Krasovskii functions (LKFs) using HOWM by partitioning integral terms and applying quadratic scaling to optimize control algorithms, thereby reducing conservatism and tightening upper-bound criteria. Additionally, it proposes an innovative Integral Event-Triggered Control (IETC) strategy for estimating system performance errors. The performance of the proposed control algorithm is rigorously evaluated through simulations on a complex 2-degree of freedom (DoF) helicopter system (HS). Note to Practitioners—This study delves into DoS attacks targeting NCSs due to TCP/IP protocol vulnerabilities. It conducts an exhaustive analysis of hacker tactics to uncover techniques used to exploit these vulnerabilities. The research reframes the PEE problem, casting it as a search for ellipsoid constraints within a P-dependent ellipsoidal set. It introduces the HOWM to optimize sampling intervals, leveraging the unique characteristics of zero-order-hold sampling. Additionally, the study enhances LKFs using HOWM by partitioning integral terms and applying quadratic scaling to optimize the control algorithm, thereby reducing conservatism and tightening upper-bound criteria. Moreover, it introduces an innovative IETC strategy for estimating system performance errors. The performance of the proposed control algorithm is rigorously evaluated through simulations on a complex 2-DoF HS.

Graph partitioning as a classic NP-complete problem, is the most fundamental procedure that needs to be performed before parallel computations. Partitioners can be divided into vertex- and edge-based approaches. Recently, both approaches are employing a streaming heuristic to find approximate solutions. It is lightweight in space and time complexities, but suffers from suboptimal partitioning quality, especially for directed graphs where the explicit knowledge provided for heuristic is limited. This paper thereby proposes new heuristics for not only vertex-based but also edge-based partitioning. They improve quality by additionally utilizing implicit knowledge, which is embedded in the local streaming view and the global graph view. Memory reduction techniques are presented to extract this knowledge with negligible space costs. That preserves the lightweight advantages of streaming partitioning. Besides, we study parallel acceleration and restreaming, to further boost the partitioning efficiency and quality. Extensive experiments validate that our proposals outperform the state-of-the-art competitors.

Access Control Lists (ACLs) are crucial for ensuring the security and integrity of modern cloud and carrier networks by regulating access to sensitive information and resources. However, previous software and hardware implementations no longer meet the requirements of modern datacenters. The emergence of FPGA-based SmartNICs presents an opportunity to offload ACL functions from the host CPU, leading to improved network performance in datacenter applications. However, previous FPGA-based ACL designs lacked the necessary flexibility to support different rulesets without hardware reconfiguration while maintaining high performance. In this paper, we propose HACL, a heterogeneous and adaptive architecture for decision-tree-based ACL engine on FPGA. By employing techniques such as tree decomposition and recirculated pipeline scheduling, HACL can accommodate various rulesets without reconfiguring the underlying architecture. To facilitate the efficient mapping of different decision trees to memory and optimize the throughput of a ruleset, we also introduce a heterogeneous framework with a compiler in CPU platform for HACL. We implement HACL on a typical SmartNIC and evaluate its performance. The results demonstrate that HACL achieves a throughput exceeding 260 Mpps when processing 100K-scale ACL rulesets, with low hardware resource utilization. By integrating more engines, HACL can achieve even higher throughput and support larger rulesets.

Container technology in cloud computing has improved resource utilization and deployment efficiency, but it also introduces new security risks. Excessive privileges in containerized environments can allow attackers to exploit insufficiently restricted system calls, potentially leading to container escapes and other attacks. Some system calls are only necessary during the initialization phase of a container, and allowing them during runtime can increase the risk of exploitation. This paper proposes a Phase-based System Call Filtering (PSF) method to minimize system call permissions during the runtime of cloud containers. The PSF method builds a comprehensive whitelist of system calls during the initialization phase to cover all necessary calls for containerized applications. During runtime, a refined, phase-specific system call whitelist is enforced, dynamically adjusting privileges based on the functions encapsulated within the container. Additionally, we introduce a container phase recognition algorithm to distinguish between initialization and runtime phases, supporting the generation of phase-specific system call lists. Experimental results show that the proposed method enhances runtime system call restrictions, minimizes privileges, and improves the overall security of cloud containers.

The adoption of container-based cloud computing services has been prevalent, especially with the introduction of Kubernetes, which enables the automated deployment, scaling, and administration of applications in containers, hence boosting the popularity of containers. As a result, researchers have placed greater emphasis on container runtime security, notably investigating the efficacy of traditional techniques such as Capabilities, Seccomp, and Linux security modules in guaranteeing container security. However, due to the limitations imposed by the container environment, the results have been unsatisfactory. In addition, eBPF-based solutions face the problem of being unable to quickly load policies and affect real-time operations when faced with newer kernel vulnerabilities. This paper investigates the limitations of existing container security mechanisms. Additionally, it examines the specific constraints of these mechanisms in Kubernetes environments. The paper classifies container monitoring and obligatory access control into three distinct categories: system call access control, LSM hook access control, and kernel function access control. Therefore, we propose a technique for regulating container access with a variety of granularity levels. This technique is executed using eBPF and is tightly integrated with Kubernetes to collect relevant meta-information. In addition, we suggest implementing a consolidated routing method and employing function tail call chaining to overcome the limitation of eBPF in enforcing mandatory access control for containers. Lastly, we conducted a series of experiment to verify the effectiveness of the system's security using CVE-2022-0492 and to benchmark the system that had BPFGuard enabled. The results indicate that the average performance loss increased merely by 2.16%, demonstrating that there are no adverse effects on the container services. This suggests that greater security can be achieved at a minimal cost.

Temporal dynamic graphs (TDGs), representing the dynamic evolution of entities and their relationships over time with intricate temporal features, are widely used in various real-world domains. Existing methods typically rely on mainstream techniques such as transformers and graph neural networks (GNNs) to capture the spatiotemporal information of TDGs. However, despite their advanced capabilities, these methods often struggle with significant computational complexity and limited ability to capture temporal dynamic contextual relationships. Recently, a new model architecture called mamba has emerged, noted for its capability to capture complex dependencies in sequences while significantly reducing computational complexity. Building on this, we propose a novel method, TDG-mamba, which integrates mamba for TDG learning. TDG-mamba introduces deep semantic spatiotemporal embeddings into the mamba architecture through a specially designed spatiotemporal prior tokenization module (SPTM). Furthermore, to better leverage temporal information differences and enhance the modeling of dynamic changes in graph structures, we separately design a bidirectional mamba and a directed GNN for improved spatiotemporal embedding learning. Link prediction experiments on multiple public datasets demonstrate that our method delivers superior performance, with an average improvement of 5.11% over baseline methods across various settings.

Ethereum, as one of the most active cryptocurrency trading platforms, has garnered significant academic interest due to its transparent and accessible transaction data. In recent years, phishing scams have emerged as a serious criminal activity on Ethereum. Although most studies model Ethereum account transactions as networks and analyze them using traditional machine learning or network representation learning techniques, these approaches often rely solely on the latest static transaction records or use manually designed features while neglecting transaction histories, thus failing to fully capture the dynamic interactions and potential trading patterns between accounts. This article introduces an innovative multiperspective cascaded dynamic graph neural network model named DMPCG, which extracts phishing transaction data from authoritative databases like blockchain explorers to construct transaction network graphs. The model elevates the analysis from the microscopic features of nodes to the macroscopic dynamics of the entire network, integrating the attributes of static snapshot graphs with the evolution of dynamic trading networks, significantly enhancing the accuracy of phishing detection. Experimental results demonstrate that the DMPCG method achieves an impressive precision of 92.6% and an F1-score of 90.9%, outperforming existing baseline models and traditional subgraph sampling techniques.

This article addresses the challenge of achieving asymptotic stability in nonlinear networked control systems (NNCSs) amid denial-of-service (DoS) attacks, particularly under constrained communication resources. We begin by establishing a practical DoS attack model using the NSL-KDD dataset, which provides a realistic depiction of DoS attack dynamics based on real-world data. We then introduce the iterative shrinkage-thresholding algorithm (ISTA) to supervise the adaptive event-triggered controller (AETC), ensuring that system parameters are adjusted effectively while conserving communication resources. We develop an enhanced data compression mechanism to further mitigate the impact of DoS attacks on communication servers. Additionally, we construct an asymmetric Lyapunov-Krasovskii function (LKF) to rigorously verify the asymptotic stability of NNCSs. Finally, we empirically validate the effectiveness of our proposed AETC using an autonomous vehicle (AV) model.

The landscape of cyberattacks has transcended from mere hobbyist pursuits of cybercriminals to a lucrative business model, facilitating their sustenance. Concurrently, the cybercrime market has evolved into a complex ecosystem. Empowered by this environment, cybercriminal tactics have evolved from simple, isolated activities to intricate and coordinated cyberattacks. In this article, we reveal a new type of cyberattack paradigm termed Attack Exploiting Cyber-arms Industry (AECI), which, despite its potential for severe impact, requires less investment and entails fewer obstacles and risks compared to traditional methods. However, this type of attack is still neglected by security researchers and communities and this is the first work focusing on this type of attack. To elucidate AECI, we provide an overview of the cyber-arms industry and introduces an attack model. The model dissects each phase of AECI to illuminate its operational mechanics and strategic imperatives. Furthermore, to assess its potential impact, a mathematical model is proposed to estimate the scale of infection attributable to AECI. Through analysis of a specific attack case, our findings demonstrate that AECI can generate significant impacts within a brief timeframe, akin to the magnitude observed with the Mirai botnet. The proposed model is demonstrated to prove instrumental in effectively analyzing AECI and providing accurate estimations of its infection scale.

As blockchain technology matures and its adoption grows across multiple industries, there is a growing need to make the data stored on blockchains adaptable to prevent misuse. However, such modifications can lead to inconsistency when interacting with other blockchains, necessitating the preservation of dynamic data consistency during these cross-chain interactions. While ensuring consistency for static data is relatively straightforward, doing so for dynamic data with efficiency remains a significant challenge. In response, we propose an efficient dynamic cross-chain data consistency auditing model (EDCA), which artfully integrates an advanced gamma multi-signature approach (AGMS) with the designed dynamic Merkle hash tree (D-MHT) to facilitate effective auditing of the consistency for dynamic data in cross-chain interaction. EDCA can produce relatively small auditor states while maintaining the storage proof. Moreover, EDCA is proven to satisfy strong security and privacy guarantees, tag unforgeability, and proof unforgeability. Experimental evaluations confirm that EDCA has high computational and communicational efficiency and can retain a small and relatively constant auditing overhead.

For security enhancement and privacy preservation, homomorphic encryption is deployed to facilitate computations among cipher texts. To transmit extra data over a cipher text without affecting its usage, lossless data hiding in encrypted domain (LDH-CT) has been developed by exploiting randomness introduced in homomorphic encryption such as Paillier, BGN and NTRU schemes. As data extraction without decrypting the cipher text has been accomplished with BGN, how to retrieve the hidden data after decryption remains unexplored. In this article, an LDH-CT scheme named random number replacement and partitioning (RNRP) is proposed to achieve the versatility of data embedding by designing two algorithms, namely the random number replacement algorithm to embed data to be extracted after decryption and the cipher value selection algorithm to embed data to be extracted without decryption. For the first time, confidential information to be extracted after decryption can be hidden into a cipher text by a third party without knowing its plain text. For convenience in performance evaluation, the proposed algorithms and the RNRP scheme are applied to a set of test images. Experimental results and comparisons with the state-of-the-art schemes demonstrate the better applicability of the proposed scheme, such as in saving bandwidth.

This paper addresses the critical challenge of achieving asymptotic stability in networked control systems (NCSs) under denial-of-service (DoS) attacks, focusing on maintaining security and stability within bandwidth-constrained environments. First, we construct a practical attack model using the NSL-KDD dataset to provide a realistic representation of DoS attack dynamics, capturing key attributes such as attack duration and frequency. Then, an iterative shrinkage-thresholding algorithm (ISTA) is introduced to supervise the adaptive sampled-data controller (ADSC), dynamically optimizing the sampling period to enhance control performance while minimizing communication overhead. To further mitigate the impact of DoS attacks, we propose a novel data compression mechanism that adapts to varying network conditions, ensuring efficient bandwidth utilization and preserving critical control data fidelity. In addition, the stability of the NCS is rigorously verified through Lyapunov-Krasovskii functions (LKFs), demonstrating robust system behavior even under adverse network conditions. Finally, the effectiveness and practicality of the proposed approach are validated through experimental studies on a 2-degree-of-freedom (2-DoF) helicopter system, confirming its capability to ensure stability, optimize communication efficiency, and mitigate the effects of DoS attacks in real-world scenarios.

The continuous evolution of malware is posing a serious threat to personal privacy, enterprise data security, and global network infrastructure. For example, attackers can use phishing emails, botnets, etc. to induce victims to execute malware for nefarious purposes such as stealing sensitive information. Therefore, it is significant to develop effective and efficient methods to detect malware. Towards this, most state-of-the-art methods are focused on learning-based method. In order to adapt to the characteristics of sample scarcity and dynamic evolution of malware detection tasks, few-shot class incremental learning has been proposed as an efficient pairwise solution. Nevertheless, they still face two major challenges: 1) Catastrophic Forgetting: the erosion of existing knowledge by newly acquired knowledge during incremental learning. 2) Decision boundary confusion: after continuous multiple incremental sessions, the discriminative ability of the classification model is weakened. To address the above challenges, we propose a new Malware detection framework based on Few-Shot Class Incremental Learning, MalFSCIL, which utilizes a decoupled training strategy combined with a variational autocoder to mitigate catastrophic forgetting, and designs a dynamic boundary delineation method based on class prototyping to achieve accurate delineation of incremental decision boundaries. Extensive experimental results show that the proposed method outperforms the state-of-the-art techniques in malware detection and classification with high classification accuracy with open-source dataset and Internal enterprise dataset.

Model Inversion Attacks (MIAs) pose a certain threat to the data privacy of learning-based systems, as they enable adversaries to reconstruct identifiable features of the training distribution with only query access to the victim model. In the context of deep learning, the primary challenges associated with MIAs are suboptimal attack success rates and the corresponding high computational costs. Prior efforts assumed that the expansive search space caused these limitations, employing generative models to constrain the dimensions of the search space. Despite the initial success of these generative-based solutions, recent experiments have cast doubt on this fundamental assumption, leaving two open questions about the influential factors determining MIA performance and how to manipulate these factors to improve MIAs. To answer these questions, we reframe MIAs from the perspective of information flow. This new formulation allows us to establish a lower bound for the error probability of MIAs, determined by two critical factors: (1) the size of the search space and (2) the mutual information between input and output random variables. Through a detailed analysis of generative-based MIAs within this theoretical framework, we uncover a trade-off between the size of the search space and the generation capability of generative models. Based on the theoretical conclusions, we introduce the Query-Efficient Model Inversion Approach (QE-MIA). By strategically selecting an appropriate search space and introducing additional mutual information, QE-MIA achieves a reduction of$60\%\sim 70\%$in query overhead while concurrently enhancing the attack success rate by$5\%\sim 25\%$.

In the era where Web3.0 values data security and privacy, adopting groundbreaking methods to enhance privacy in recommender systems is crucial. Recommender systems need to balance privacy and accuracy, while also having the ability to overcome cold start problems. The Differential Trust Mechanism (DTM) introduced in this paper is such an approach. The DTM provides a unique use of Gaussian distributions in modeling trust relationships within data, offering a novel way to balance recommendation accuracy with user privacy. This mechanism innovatively applies differential privacy principles, using Gaussian noise addition to protect individual user data from inference attacks, while maintaining the integrity and utility of the overall dataset. Unlike traditional anonymization techniques that often compromise data utility or vulnerability to reverse engineering, DTM provides a robust solution by dynamically adjusting privacy levels based on the trustworthiness of data requests. By combining DTM with existing mainstream recommendation algorithms, the prediction accuracy of MAE and RMSE increases by at least 6.60% and 2.69%, respectively. This dual benefit positions DTM as a significant advancement in secure data processing, especially relevant for online businesses and platforms where personalized recommendations are crucial yet privacy concerns are paramount.

The widespread adoption of face manipulation systems has brought entertainment and convenience to users while posing significant challenges to media forensics. Conventional active defense strategies typically generate adversarial images by introducing perturbations into the original images. When adversarial images undergo facial manipulation, they often exhibit distortions or speckle artifacts, which helps reduce the dissemination of forged content on social media platforms. Nevertheless, the widespread dissemination of degraded images may contribute to facial stigmatization. Furthermore, conventional perturbation techniques are vulnerable to failure under JPEG compression and various image processing operations on OSN platforms. To address these challenges, we introduce a robust and unstigmatized imperceptible perturbation (RUIP) method designed to counteract face manipulation. First, RUIP utilizes an end-to-end adversarial training framework to generate robust and imperceptible perturbations. Second, to mitigate facial stigmatization, we incorporate both pixel-level and feature-level guidance losses during training, ensuring that the output images remain visually natural and closely aligned with the original images. Finally, we develop a novel module, the Flexible Random Enhancement Generator (FREG), to simulate complex JPEG compression and diverse image processing operations on OSN platforms, enhancing the model’s robustness against perturbations. Extensive qualitative and quantitative experiments demonstrate that the proposed method effectively defends against face manipulation attacks while preserving the visual quality of facial images under JPEG compression and other image processing operations on OSN platforms. We propose an effective and unstigmatized defense algorithm to safeguard privacy and maintain the stability of the social media ecosystem.Code is available athttps://github.com/silencecmsj/RUIP.

Reinforcement Learning (RL), particularly deep reinforcement learning (DRL) has shown significant potential in addressing optimal attack path discovery problems (OAPDPs) for cybersecurity. However, existing approaches often oversimplify the causal relationships among attack actions, limiting their applicability to complex systems. This study pioneers DRL-based solutions for OAPDPs in Intelligent Transportation Systems (ITS), specifically targeting scenarios where attack actions exhibit disjunctive, conjunctive, and hybrid causal relationships. We propose TTCRT, a novel attack pattern template that formalizes attack logic through vector-compatible representations of OAPDP-related attack components, while developing a refinement method for TTCRT-derived attack patterns and presenting a rigorous framework for formalizing OAPDPs as Markov Decision Processes (MDPs) based on refined attack patterns. Through extensive experiments, we demonstrate TTCRT’s capability to rigorously capture disjunctive, conjunctive, and hybrid causal relationships among attack actions, achieving semantic equivalence with Logical Attack Graphs (LAGs) while resolving their implementation bottlenecks in highly complex systems like ITS environments. The framework seamlessly integrates with established DRL algorithms to accurately identify optimal attack paths in an intelligent traffic management system. These findings establish TTCRT as a foundational framework for RL-driven OAPDP resolution in ITS and other sophisticated systems with complex attack dynamics.

Data privacy protection legislation around the world has increasingly enforced the “right to be forgotten” regulation, generating a surge in research interest in machine unlearning (MU), which aims to remove the impact of training data from machine learning models upon receiving revocation requests from data owners. There exist two major challenges for the performance of MU: the execution efficiency and the inference interference. The former requires minimizing the computational overhead for each execution of the MU mechanism, while the latter calls for reducing the execution frequency to minimize interference with normal inference services. Nowadays most MU studies focus on the sample-level unlearning setting, leaving the other paramount feature-level setting under-explored. Adapting these existing techniques to the latter turns out to be non-trivial. The only known feature-level work achieves anapproximateunlearning guarantee, but suffers from degraded model accuracy and still leaves the inference interference challenge unsolved. We are therefore motivated to propose FELEMN, the first FEature-Level Exact Machine uNlearning method that overcomes both of the above-mentioned hurdles. For the MU execution efficiency challenge, we explore the impact of different feature partitioning strategies on the preservation of semantic relationships for maintaining model accuracy and MU efficiency. For the inference interference challenge, we propose two batching mechanisms to combine as many individual unlearning requests to be processed together as possible, while avoiding potential privacy issues coming with falsely postponing unlearning requests, which is grounded on theoretical analysis. Experiments on five real datasets show that our FELEMN outperforms up-to-date competitors with up to$3\times$speedup for each MU execution, and 50% runtime reduction by mitigating inference interference.

Most of the current federated learning schemes aimed at safeguarding privacy exhibit vulnerability to collusion attacks and lack a verification mechanism for participants to consolidate the aggregation results of the parameter server, leading to privacy breaches for users and inaccurate model training outcomes. In order to address these issues, we propose a verifiable privacy-preserving federated learning framework against collusion attacks. Primarily, the federated learning scheme is reconfigured utilizing the ElGamal encryption algorithm, which effectively safeguards the data privacy of participants in scenarios involving collusion between certain participants and servers. Additionally, the introduction of the assistant server can realize the joint decryption of the gradient ciphertext by the non-collusive parameter server and assistant server, which can effectively resist the internal attack of a single parameter server model in the process of data upload. Third, this scheme designs a verification mechanism that enables participants to effectively verify the accuracy and integrality of the parameter server's aggregated results, preventing the parameter server from returning incorrect aggregation results to participants. Experimental results and performance analysis demonstrate that our proposed scheme not only fortifies security measures but also upholds the precision of model training, surpassing the security and correctness of many existing methodologies.

In the forthcoming era of 6G networks, delay-sensitive applications for Internet of Things (IoT) are poised to become the prevailing services with ultra-reliable and low-latency (URLLC) requirements. Unlike traditional video caching, IoT-based edge caching faces unique challenges due to diverse data types, update frequencies, and computational needs, requiring integrated storage and computational resource management. To support the more stringent requirements for these innovative applications, mobile edge computing (MEC) is introduced to enhance the service reliability of delay-sensitive applications in the 6G era. However, task offloading, as an indispensable procedure in MEC, would encounter many challenges, such as network jitter and resource insufficiency, possibly leading to unpredictable queuing delays and other negative issues. To ensure reliable services in a dynamical MEC environment, the caching-enabled MEC network has emerged as a novel architecture, placing computing and storage resources in the edge network. In this paper, we investigate the caching-enabled MEC to support reliable task offloading for delay-sensitive applications, with a focus on IoT scenarios. In our system model, we formulate the task process as a two-hop tandem queuing system with limited capacity, including task transmission and computation queues. The Martingale theory is leveraged to analyze the delay violation probability in this system, demonstrating how the offloading and caching decisions affect the end-to-end (E2E) delay. Besides, task offloading and resource allocation policies are integrated to reduce high system costs, including energy consumption and cache resource rental costs. Based on the delay analysis of martingale theory, we propose an advanced deep reinforcement learning (DRL) algorithm called Dynamic Request Aware Soft Actor-Critic (DRA-SAC) algorithm to achieve minimal system costs by obtaining the optimal task offloading and resource allocation policies, including caching and computation resources. We conduct some illustrative studies to evaluate the proposed scheme. The algorithm we have put forward outperforms benchmark algorithms regarding both cache hit ratio and system cost.

With the rapid development of mobile and wireless technologies, the industrial manufacturing sector has entered the era of automation. The proliferation of mobile devices, sensor networks, and remote monitoring systems enables factory equipment to be more flexibly connected and controlled. However, the trend towards industrial networks also brings new challenges. Industrial control systems (ICSs) and programmable logic controllers (PLCs) are more susceptible to hacker attacks and interference. Honeypoints have been developed to protect ICSs from addressing these threats, including potential internal attacks. Honeypoints are active deception systems that mitigate the limitations of conventional defense mechanisms, which successfully entice and neutralize internal enemies. This paper presents the PLC Honeypoint Communication Generator (PHCG), enhancing honeypoint protective capabilities in industrial IoT systems. Using an automated construction process, PHCG provides a convenient and efficient deployment method, ensuring quick and effective functioning. The functionality of a PLC relies on a data generation model trained on PLC response data. This model allows PHCG to imitate genuine PLC responses accurately when given authorized commands. The experimental results illustrate the adaptability of information produced by PHCG in different communication processes, with satisfactory timescales for both model training and data generation.

Convolutional neural networks, in which each layer receives features from the previous layer(s) and then aggregates/abstracts higher level features from them, are widely adopted for image classification. To avoid information loss during feature aggregation/abstraction and fully utilize lower layer features, we propose a novel decision fusion module (DFM) for making an intermediate decision based on the features in the current layer and then fuse its results with the original features before passing them to the next layers. This decision is devised to determine an auxiliary category corresponding to the category at a higher hierarchical level, which can, thus, serve as category-coherent guidance for later layers. Therefore, by stacking a collection of DFMs into a classification network, the generated decision fusion network is explicitly formulated to progressively aggregate/abstract more discriminative features guided by these decisions and then refine the decisions based on the newly generated features in a layer-by-layer manner. Comprehensive results on four benchmarks validate that the proposed DFM can bring significant improvements for various common classification networks at a minimal additional computational cost and are superior to the state-of-the-art decision fusion-based methods. In addition, we demonstrate the generalization ability of the DFM to object detection and semantic segmentation.

Learning-based systems have been proved to be vulnerable against model inversion attacks (MIAs), where attackers steal private information of training data by querying the target model using synthetic samples. To alleviate the urgent threat introduced by MIAs, existing advancements are proposed to increase the attack overhead by limiting the information available. Although these methods successfully reduced the attack success rate (ASR) for a one-time inversion attempt, they usually compromise the usability of the protected model. More importantly, existing MIA defense methods fail to capture attack attempts, which can lead to persistent threats to data privacy. To bridge this gap, we propose Neural Honeypoint, an active defense framework against MIAs. The key insight is that MIA attackers will make a series of forward steps in the feature space while benign users will not. Motivated by the observation, defenders can deploy active defense devices (honeypoints) on critical paths to capture attack behaviors. Specifically, Neural Honeypoint first models the attackers' capabilities from the frequency domain and designs specialized honeypoints for protected classes in the training dataset. Subsequently, it deploys these honeypoints into the protected model via backdoor-like model fine-tuning. Then, defenders can distinguish model inversion examples by comparing the similarity of input features with deployed honeypoints. Experiments show that Neural Honeypoint reduces the ASRs of advanced MIAs to 0%~2%. Furthermore, it can effectively capture inversion queries, which helps defenders to detect and block attacks in time.

In recent years, supply chain attacks have garnered significant attention from both enterprises and the security community due to their profound impact. Numerous studies have begun to focus on supply chain attacks targeting legitimate software. However, it is noteworthy that supply chain attacks also occur within cybercrime markets, which differ substantially from those observed in lawful markets. These attacks not only facilitate widespread infections but also enable the rapid formation of substantial botnets. Regrettably, supply chain attacks within cybercrime markets have largely been overlooked by the security community, resulting in a notable deficiency in systematic methodologies for comprehending such assaults. This work concentrates on a specific supply chain attack observed in cybercrime markets, denoted as the “Nigrita Attack.” To facilitate a systematic understanding of this attack, we introduce a model designed to delineate its propagation dynamics. Additionally, we propose an approach for forecasting its future threat potential. To assess the efficacy of the proposed propagation model and the accuracy of the method for predicting the scale of infections, we assembled a dataset comprising more than 40,342 distinct malware samples and conducted a comprehensive series of analyses. Empirical results substantiate both the effectiveness of the proposed propagation model and the precision of the approach in estimating the magnitude of potential infections.

Federated tree-based models are popular in many real-world applications owing to their high accuracy and good interpretability. However, the classical synchronous method causes inefficient federated tree-based model training due to tree node dependencies. Inspired by speculative execution techniques in modern high-performance processors, this paper proposes FTSeir, a novel and efficient speculative federated learning system. Instead of simply waiting, FTSeir optimistically predicts the outcome of the prior tree node. By resolving tree node dependencies with a neural network-based split point predictor, the training tasks of child tree nodes can be executed speculatively in advance via separate threads. This speculation enables cross-layer concurrent training, thus significantly reducing the waiting time. Furthermore, we propose an eager verification mechanism to promptly identify mispredictions, thereby reducing wasted computing resources. On a misprediction, an incomplete rollback is triggered for quick recovery by reusing the output of the mis-speculative training, which reduces computational requirements. We implement FTSeir and evaluate its efficiency in a real-world federated learning setting with six public datasets. Evaluation results demonstrate that FTSeir achieves up to 3.45× and 3.60× speedup over the state-of-the-art gradient boosted decision trees and random forests implementations, respectively.

Malware acts as a critical component in network attack and defense mechanisms. As Malware threats become more complex and diverse, timely automatic malware classification is urgently needed. Take into account consistency and system resource consumption, malware classification should be compatible with other security devices to form a comprehensive defense system. In this study, we proposed a pre-training framework Proof based on malware behavior captured by honeypoints inside and outside the protected system, and the framework’s performance in the categorization of malware families is evaluated. To describe malware, a structure called Malware Behavior Instruction Set (MBIS) was designed by selecting a subset of all behaviors captured by honeypoints. Then, a self-supervised pre-training model MBI2vec is applied to learn the internal mode of malicious code to guide the downstream malicious code family classification model composed of Bi-LSTM and attention mechanism. Finally, the Proof framework evaluated 846 malware samples from five malware families that we captured in the real world, and the f1-score of the classification result was 0.9554.

PLC, as the core of industrial control systems, has been turned into a focal point of research for attackers targeting industrial control systems. However, current researched methods for attacking PLCs suffer from issues such as lack of precision and limited specificity. This paper proposes a novel attack method called AOIFF. Specially, AOIFF extracts the binary control logic code from a running PLC and reverses the binary code into assemble code. And then awareness of industrial field information is extracted from assemble code. Finally, it is based on awareness that attack code is generated and injected into a PLC, which can disrupt the normal control logic and then launch precise attacks on industrial control systems. Experimental results demonstrate that AOIFF can effectively perceive information in industrial field and initiate precise and targeted attacks on industrial control systems. Additionally, AOIFF achieves excellent results in the reverse engineering of binary code, enabling effective analysis of binary code.

Adversarial attacks on 3D point clouds often exhibit unsatisfactory imperceptibility, which primarily stems from the disregard for manifold-aware distortion, i.e., distortion of the underlying 2-manifold surfaces. In this paper, we develop novel manifold constraints to reduce such distortion, aiming to enhance the imperceptibility of adversarial attacks on 3D point clouds. Specifically, we construct a bijective manifold mapping between point clouds and a simple parameter shape using an invertible auto-encoder. Consequently, manifold-aware distortion during attacks can be captured within the parameter space. By enforcing manifold constraints that preserve local properties of the parameter shape, manifold-aware distortion is effectively mitigated, ultimately leading to enhanced imperceptibility. Extensive experiments demonstrate that integrating manifold constraints into conventional adversarial attack solutions yields superior imperceptibility, outperforming the state-of-the-art methods.

Deep neural networks (DNNs) often display overconfidence when encountering out-of-distribution (OOD) samples, posing significant challenges in real-world applications. Capitalizing on the observation that responses on convolutional kernels are generally more pronounced for in-distribution (ID) samples than for OOD ones, this paper proposes the COnvolutional REsponse-based Score (CORES) to exploit these discrepancies for OOD detection. Initially, CORES delves into the extremities of convolutional responses by considering both their magnitude and the frequency of significant values. Moreover, through backtracking from the most prominent predictions, CORES effectively pinpoints sample-relevant kernels across different layers. These kernels, which exhibit a strong correlation to input samples, are integral to CORES's OOD detection capability. Comprehensive experiments across various ID and OOD settings demonstrate CORES's effectiveness in OOD detection and its superiority to the state-of-the-art methods.

Reparameterization techniques have demonstrated their efficacy in improving the efficiency of deep neural networks. However, their application has been largely confined to single-input network structures, leaving multi-input ones, commonly encountered in real-world applications, largely unexplored. In this paper, we formulate reparameterization head (RepHead), the first framework designed to introduce reparameterization into multi-input neural networks. RepHead compresses multiple inputs into a single input and employs reconstruction operations to recover them, thereby transforming multi-input networks into single-input, multibranch architectures, thereby enabling the application of reparameterization. We demonstrate the usage of RepHead in both image and point cloud domains. Extensive experimental results validate that the integration of RepHead substantially reduces computational overhead and memory requirements while maintaining minimal performance loss.

Adversarial attacks on point clouds are crucial for assessing and improving the adversarial robustness of 3D deep learning models. Despite leveraging various geometric constraints, current adversarial attack strategies often suffer from inadequate imperceptibility. Given that adversarial perturbations tend to disrupt the inherent symmetry in objects, we recognize this disruption as the primary cause of the lack of imperceptibility in these attacks. In this paper, we introduce a novel framework, symmetry-aware imperceptible adversarial attacks on 3D point clouds (SymAttack), to address this issue. Our approach starts by identifying part- and patch-level symmetry elements, and grouping points based on semantic and Euclidean distances, respectively. During the adversarial attack iterations, we intentionally adjust the perturbation vectors on symmetric points relative to their symmetry plane. By preserving symmetry within the attack process, SymAttack significantly enhances imperceptibility. Extensive experiments validate the effectiveness of SymAttack in generating imperceptible adversarial point clouds, demonstrating its superiority over the state-of-the-art methods.

Language models have shown vulnerability against backdoor attacks, threatening the security of services based on them. To mitigate the threat, existing solutions attempted to search for backdoor triggers, which can be time-consuming when handling a large search space. Looking into the attack process, we observe that poisoned data will create a long-tailed effect in the victim model, causing the decision boundary to shift towards the attack targets. Inspired by this observation, we introduce LT-Defense, the first searching-free backdoor defense via exploiting the long-tailed effect. Specifically, LT-Defense employs a small set of clean examples and two metrics to distinguish backdoor-related features in the target model. Upon detecting a backdoor model, LT-Defense additionally provides test-time backdoor freezing and attack target prediction. Extensive experiments demonstrate the effectiveness of LT-Defense in both detection accuracy and efficiency, e.g., in task-agnostic scenarios, LT-Defense achieves 98% accuracy across 1440 models with less than 1% of the time cost of state-of-the-art solutions.

The swift advancement of internet technologies has had profound effects across numerous sectors, placing cyber-security at the forefront of concerns for both businesses and governmental entities. Effective cyber risk assessment enables network administrators to reinforce their defenses against cyber attacks and mitigate potential losses. However, traditional approaches frequently concentrate solely on either the difficulty of exploiting vulnerabilities or the topological characteristics of attack graphs, and some require knowledge of the attacker’s strategies. This paper presents a novel network risk assessment methodology that combines cooperative game theory with host probabilistic Attack Graphs, called RACAG. Specifically, Shapley values is employed to provide an unbiased assessment the impact of each attack path on the overall system damage. The proposed approach does not require any prior knowledge of the attacker’s strategies but instead focuses on comprehensive analysis of the exploit difficulty of vulnerabilities in the system, asset information, network topology, and attacker traces detected by defense mechanisms. Experiments have confirmed the efficacy of RACAG in assessing the impact of attack paths on overall system damage. Additionally, they highlighted its advantages in the deployment of deceptive resources, enhancement of risk perception abilities, and its support for defenders in analyzing smokescreen tactic.

The encryption of traffic data offers a means of protecting the security of data and the private information of the public. However, this same technology also presents a potential avenue for attackers to conceal malicious activities. Attackers hide malicious behaviour in encrypted traffic data to bypass detection by firewalls or early intrusion detection systems (IDS). In order to cope with malicious encrypted traffic, traffic attack detection is classified into active and passive detection depending on the way it is handled. Active detection is mainly based on searchable traffic detection and traffic plaintext data parsing. Measures such as analysing controllable transmission protocols and trusted execution environments are used to ensure both attack detection efficiency and privacy security. Passive detection focuses on feature construction of encrypted traffic. The characterisation of traffic data from multiple perspectives, including channel and context, is achieved through the utilisation of machine learning or deep learning models, thereby facilitating the generation of accurate prediction outcomes. This paper offers an overview of existing detection methods from two distinct vantage points: active attack detection and passive attack detection. Finally, the paper presents a summary of the strengths and weaknesses of existing methods and suggests potential avenues for future research.

In the scenario of smart healthcare applications, the Internet of Medical Things (IoMT) devices, equipped with limited resources, would offload numerous computation-heavy tasks to an edge server through 5G networks. However, IoMT devices should usually move around different diagnostic areas in smart healthcare systems, leading to the dynamics of the uplink channel quality. Moreover, the burst generation of a substantial number of tasks from IoMT devices can result in congestion within the computing queue of the edge server. And, heterogeneous services in IoMT devices make it hard to collect global information for a central controller to get the optimal optimization for all IoMT devices. So, how to determine task offloading among IoMT devices in a distributed scenario of smart healthcare applications should be considered appropriately and comprehensively. In this paper, we investigate task offloading in mobile edge computing (MEC) through wireless networks. To improve the utilization of wireless resources, non-orthogonal multiple access (NOMA) is adopted in 5G networks. We first formulate the mobility of IoMT devices as a Hidden Markov Model (HMM) and the problem of task offloading policy as a distributed Partial Markov Decision Process (Dec-POMDP). Then, we propose a mobility-aware method based on Multi-agent reinforcement learning for task offloading in 5G NOMA-enabled networks. In our approach, task offloading scheduling for each IoMT device in NOMA-enabled 5G networks is considered to improve energy efficiency and guarantee service quality. Besides, the time complexity and the existence of a Nash equilibrium for our proposed Dec-POMDP method are theoretically derived. Simulations are conducted to show that our algorithm outperforms other alternative methods in energy consumption under the delay constraint.

Federated learning (FL) is widely studied for local privacy protection, and it involves exchanging model parameters rather than raw data among clients. However, gradient attacks (GAs) make a malicious client or parameter server of FL infer the local data of other clients only based on the model parameters exchanged. In FL frameworks and processes, it is important to understand the features that provide heuristic information for inferring raw data, as well as how best to defend against GAs. The academic community is currently investigating this problem. In this study, we demonstrate that the labels of input samples play a key role in the success of GAs. We analyze the rank of the coefficient matrix of the nonhomogeneous linear equation of gradients and input samples and propose an approach that performs special operations on the repetition and order of labels. The approach achieves a better defense effect against GAs without using a differential privacy (DP) framework. Our experimental results show that GAs fail (i.e., without leaking any valid information about local data) during the entire training process of a deep convolutional network in FL, and the accuracy of the network is less affected than that of DP. The code is available athttps://github.com/zhaohuali/Label-based-Defense.

With the proliferation of cloud applications, users are increasingly uploading their private images to cloud servers to avail of supplement storage capacity. Unlike image encryption, the thumbnail-preserving technique is a way of safeguarding image privacy and preserving the outline of the thumbnail image, which allows authorized users to recognize it according to prior knowledge of the original image. Recently, several thumbnail-preserving encryption schemes have been proposed. But most of the sum-preserving encryption-based schemes are irreversible, thereby significantly constraining their practical value. Moreover, the visual quality of images generated with the reversible thumbnail-preserving schemes is unsatisfactory, leading to a decrease in the user recognition accuracy. In light of the aforementioned, this paper presents an efficient scheme to generate thumbnail-preserving images with high visual naturalness, while ensuring reversibility of the process. Experimental results demonstrate that the proposed scheme achieves significant visual quality improvements compared to state-of-the-art schemes. Furthermore, the thumbnail-preserving images generated by using our scheme can effectively evade detection by confusing malicious attackers.

There is a large amount of real-time data, e.g., measurement data and instructions, among controllers, sensors, and actuators in the Industrial IoT. These data are vulnerable to unauthorized access and tampering. In addition, once the controller is controlled by malicious code, it may send out dangerous instruction that is not compliant with the preset control process, which seriously interferes with the industrial control process. To achieve correct and undisturbed control based on real-time data, we propose an attribute-based access control (ABAC) method for real-time data in the Industrial IoT to mitigate unauthorized access and tampering, and noncompliant operation. First, we analyze the abnormal behaviors of real-time data interaction in the Industrial IoT. Second, we propose the multilevel hash identity authentication method to identify and block unauthorized access and tampering with real-time data. And, then we model the timing relationship and task logic relationship of the control process into the attribute fields of the ABAC method to identify and block noncompliant operations. Further, we design an access control module and display the lightweight deployment under the availability constraints of the control service. Finally, the proposed access control method is analyzed, proved, and experimented with. The results show that the proposed method can prevent unauthorized and noncompliant behaviors to field real-time data, meanwhile, it has a controllable delay and better scalability.

Spatial crowdsourcing (SC) has become a fundamental and emerging technology in Metaverse, facilitating the creation of immersive experiences through location-based services. In these systems, a central SC server leverages SC workers who physically travel to task locations to gather spatiotemporal environment data. However, conventional SC systems face two significant challenges: (1) the SC server, functioning as a centralized authority, can sometimes be unreliable, either due to intentional or unintentional misconduct, (2) to ensure efficient task assignment and validation, the location privacy of tasks and workers is openly accessible. In this study, we formally define location privacy preserved proof generation and verification problem (LP-PGVP) within an SC task matching scenario, with the aim to the above two challenges. Our proposed solution is a blockchain-based SC system (BlockSC), which provides a decentralized platform for task requesters and workers in the Metaverse context through calling smart contracts. We also introduce a ciphertext-based task matching scheme where task location access is granted only to eligible workers executing a task, benefiting from the design of geographic coordinate transformation and bilinear mapping methodology. To further demonstrate the task matching scheme’s operation and impact, we present an easy-to-understand case study. Our evaluation findings confirm that the proposed system effectively maintains location privacy for both SC workers and task requesters, without a considerable sacrifice in task matching efficiency.

The biclique is a fundamental model of bipartite cohesive subgraphs. To analyze a bipartite graph, many existing works seek the maximum biclique, that is, the biclique with the largest number of edges. However, our finding is that the most similar biclique (i.e., the biclique whose vertices are the most similar to each other) can be a good alternative for understanding the network. Using the model, we can detect meaningful communities with high similarity and avoid unnecessary searches based on vertex similarity. In particular, we aim to find (i) local most similar biclique : the biclique that contains a query node q and the similarity between vertices is the highest, and (ii) global most similar biclique : the biclique with the highest similarity between vertices. Despite the NP-hardness of the problems, this paper presents two efficient algorithms, Mosib and Mosib-GloApp. Specifically, our Mosib is an exact algorithm for the most similar biclique search. The algorithm incorporates three novel graph reduction rules that can reduce the size of the bipartite graph while preserving the most similar biclique, as well as two similarity-first search rules that can prioritize the bicliques with high similarity in the search. These techniques can significantly improve the practical efficiency of the algorithm. Meanwhile, our Mosib-GloApp is an approximate algorithm that adopts a novel MinHash-based dividing method, and it can further improve the efficiency of the global most similar biclique search. We experimentally evaluate our algorithms on real-world networks, and show that the most similar biclique models can find meaningful results while being computed efficiently.

There are two fundamental problems in regular simple path queries (RSPQs). One is the reachability problem which asks whether there exists a simple path between the source and the target vertex matching the given regular expression, and the other is the enumeration problem which aims to find all the matched simple paths. As an important computing component of graph databases, RSPQs are supported in many graph database query languages such as PGQL and openCypher. However, answering RSPQs is known to be NP-hard, making it challenging to design scalable solutions to support a wide range of expressions. In this paper, we first introduce the class of transitive restricted expression , which covers more than 99% of real-world queries. Then, we propose an efficient algorithm framework to support both reachability and enumeration problems under transitive restricted expression constraints. To boost the performance, we develop novel techniques for reachability detection, the search of candidate vertices, and the reduction of redundant path computation. Extensive experiments demonstrate that our exact method can achieve comparable efficiency to the state-of-the-art approximate approach, and outperforms the state-of-the-art exact methods by up to 2 orders of magnitude.

Competitive viral marketing considers the product competition of multiple companies, where each user may adopt one product and propagate the product to other users. Existing studies focus on a traditional seeding strategy where a company only selects seeds from the users with no adopted product to maximize its influence (i.e., the number of users who will adopt its product). However, influential users are often rare, and the gain from traditional seeding will degrade as the number of seeds increases. Therefore, in this paper, we study the promisingcounteringstrategy which is to counter some users who initially use other products s.t. they will turn to adopting the target product and recommending it to others. We propose the problem ofinfluence countering: given a graph, a budgetb, a target companyCt, and a setSof the seeds adopting different companies (where each seed adopts one company), we counterbusers inSwho do not adoptCtto turn to adoptCts.t. the expected number of users who eventually adoptCtin the influence diffusion is maximized. Following existing studies, we formalize the diffusion process by the Multi-Campaigner Independent Cascade model. We prove the influence countering problem is #P-complete and its influence computation is #P-hard. Then, we propose two novel algorithmsMICandMIC+to address the problem. In general,MICestimates seed influence by its empirical average influence in multiple graph samplings, whileMIC+improvesMICby reducing the cost of influence estimation and the required number of samples. Given pre-setεandl, both algorithms return a (1 -ε)-approximate solution with at least 1 -n-lprobability. We also design an index forMIC+to efficiently process graphs that are frequently updated. The experiments on 8 real-world datasets show that our algorithms are efficient in practice while offering strong result quality.

Continuously monitoring structural patterns in streaming graphs is a critical task in many real-time graph-based applications. In this paper, we study the problem of time-constrained continuous subgraph matching (shorted as TCSM) over streaming graphs. Given a query graph Q with timing order constraint and a data graph stream G , TCSM aims to report all incremental matches of Q in G for each update of G , where a match should obey both structure constraint (i.e., isomorphism) and timing order constraint of Q. Although TCSM has a wide range of applications, such as cyber-attack detection and credit card fraud detection, we note that this problem has not been well addressed. The state-of-the-art bears the limitations of high index space cost and intermediate result maintenance cost. In this paper, we propose TC-Match, an effective approach to TCSM. First, we design a space and time cost-effective index CSS, which is essentially a k -partite graph structure where a node corresponds to an edge in G. By carefully creating links between nodes, we can encapsulate into CSS the partial embedding and timing order information between edges in G. We theoretically show that CSS has polynomial space and construction time complexities. Second, based on the property of CSS, we develop an efficient incremental matching algorithm with an effective node merging optimization. Extensive experiments show that TC-Match can achieve up to 3 orders of magnitude query performance improvement over the baseline methods, and meanwhile the memory consumption is reduced by 48.7%-86.7%.

The impressive development of facial manipulation techniques has raised severe public concerns. Identity-aware methods, especially suitable for protecting celebrities, are seen as one of promising face forgery detection approaches with additional reference video. However, without in-depth observation of fake video’s characteristics, most existing identity-aware algorithms are just naive imitation of face verification model and fail to exploit discriminative information. In this article, we argue that it is necessary to take both spatial and temporal perspectives into consideration for adequate inconsistency clues and propose a novel forgery detector named SpatioTemporal IDentity network (STIDNet). To effectively capture heterogeneous spatiotemporal information in a unified formulation, our STIDNet is following a knowledge distillation architecture that the student identity extractor receives supervision from a spatial information encoder (SIE) and a temporal information encoder (TIE) through multiteacher training. Specifically, a regional sensitive identity modelling paradigm is proposed in SIE by introducing facial blending augmentation but with uniform identity label, thus encourage model to focus on spatial discriminative region like outer face. Meanwhile, considering the strong temporal correlation between audio and talking face video, our TIE is devised in a cross-modal pattern that the audio information is introduced to supervise model exploiting temporal personalized movements. Benefit from knowledge transfer from SIE and TIE, STIDNet is able to capture individual’s essential spatiotemporal identity attributes and sensitive to even subtle identity deviation caused by manipulation. Extensive experiments indicate the superiority of our STIDNet compared with previous works. Moreover, we also demonstrate STIDNet is more suitable for real-world implementation in terms of model complexity and reference set size.

Although recommender systems have achieved considerable success, sometimes it is difficult to convince users due to the failure to explain the recommendation results. For this reason, explainable recommender systems have drawn a lot of attention in recent years. Among explainable recommendation models, the meta-path-based model plays a significant role because it can reason over the path connecting a user–item pair to achieve explainability. However, it is difficult for the meta-path-based model to achieve such a common explanation in collaborative filtering as “a user similar to you has purchased item$A$” because there is no such meta-path. In this article, we contribute a new model named collaborative meta-path modeling for explainable recommendation (COMPER). It models the similarity of user pairs and item pairs through rating information and constructs collaborative meta-paths for explainability. In addition, we design an attention mechanism to aggregate different paths connecting the target user and the target item. Moreover, the information of the subgraph composed of all paths connecting the target user and the target item is integrated for rating prediction. Extensive experiments on five real-world datasets demonstrate that COMPER achieves good performance in a variety of scenarios, achieving improvements over several baselines.

Nested named entity recognition (Nested NER) in knowledge graph (KG) aims at obtaining all meaningful entities, including nested entities for sentences in longer text region. Those obtained entities are to facilitate downstream applications, such as relation extraction, entity resolution, and coreference resolution. This task, however, is challenging not only because of the demand to detect the boundary of the entity but also due to the complexity of those hierarchically nested entities. Since a substantial amount of work has been made to Flat NER (or Nested NER), a few of them can explicitly acquire the position of the entity and utilize the grammatical construction of text. In this work, we propose PANNER, a POS-aware Nested NER model, to solve all the above issues. Specifically, we first construct a heterogeneous graph by introducing the part-of-speech (POS) information of the word. Second, we design a dilated random walk (DRW) algorithm based on a grammatical path to sample a fixed size of neighbors for each node. Third, we aggregate the message from different types of neighbors through an attention mechanism. Finally, we use a bidirectional decoding module to recognize and categorize all the flat and nested entities based on the node embedding in a layer-wise manner. Our extensive experiments show the effectiveness of PANNER in both flat and nested NER.

This article presents a comprehensive solution to mitigate network congestion in T-S fuzzy networked control systems caused by denial-of-service (DoS) attacks and quality-of-service (QoS) queuing mechanisms. We develop a novel data compression mechanism to alleviate network congestion and use a mini-batch descent gradient algorithm to optimize trigger thresholds, thereby reducing bandwidth usage. In addition, we introduce asymmetric Lyapunov–Krasovskii functions to decrease the number of decision variables, which improves the reliability and robustness of the control algorithm. Finally, we propose an intelligent event-triggered controller supervised by mini-batch machine learning and validate it on the joint CarSim–Simulink platform. Experimental results demonstrate that our approach reduces the sensitivity of autonomous vehicle systems to network fluctuations while ensuring system stability under network congestion caused by DoS attacks.

This study investigates the stability problem in nonlinear networked control systems (NCSs). First, innovative compression rules are introduced to mitigate network congestion and bandwidth utilization issues stemming from quality of service (QoS) queuing mechanisms and denial of service (DoS) attacks. We develop an intelligent trigger controller supervised by a mini-batch machine learning (MBML) algorithm to optimize network bandwidth utilization. Furthermore, we formulate more generalized Lyapunov-Krasovskii functions (LKFs) to simplify mathematical derivations, and we employ appropriate integral inequalities to minimize constraints. Finally, experimental evaluations are conducted on an autonomous vehicle (AV) using the joint CarSim-Simulink platform to verify the effectiveness of the proposed intelligent trigger controller.

Lossless data hiding in ciphertexts (LDH-CT) is to perform data embedding without changing their plaintexts, which can be used to transmit extra data in the applications of homomorphic encryption at little cost. In this paper, two LDH-CT algorithms named Polynomial Encoding (PE) and Polynomial Modulation (PM) are proposed for the “N-th Degree Truncated Polynomial Ring Unit” (NTRU) scheme, respectively. In the PE algorithm, a polynomial is encoded according to a string of bit values and further used to encrypt a plain-text polynomial. After decrypting the ciphertext, the encoded polynomial can be retrieved so that dozens of bit values can be extracted from it. Moreover, the PE algorithm can be combined with a polynomial partitioning strategy to achieve data extraction before decryption as well. In applying the PM algorithm, no parameter setting of an NTRU cryptosystem is changed while a cipher-text polynomial is generated by selectively sampling a polynomial to match the to-be-hidden value. Furthermore, the data hidden with the PM algorithm can be pre-chosen to be extracted without decryption or after decryption, and in each case up to 10 bit values can be hidden into one cipher-text polynomial. The proposed algorithms and schemes are implemented and compared with several schemes developed for NTRU, BGN, LWE and Paillier encryption. Experimental results and performance evaluations demonstrate the efficacy and superiority of the proposed algorithms and schemes.

This short paper addresses the challenge of network congestion in T-S fuzzy networked control systems (NCSs) caused by denial of service (DoS) attacks and quality of service (QoS) queuing mechanisms. Firstly, a novel signal compression mechanism is introduced to mitigate network congestion. The trigger threshold is optimized using a mini-batch descent gradient algorithm, effectively reducing bandwidth utilization. Furthermore, tailored Lyapunov-Krasovskii functions (LKFs) are established for the system, and we propose an intelligent event-triggered controller (IETC) under machine learning supervision. Finally, the effectiveness of the proposed approach is demonstrated through rigorous verification on the joint CarSim-Simulink platform.

The Internet now plays a pivotal role in the social and economic landspace, providing individuals and businesses with access to essential daily services and tasks. However, it has also become a breeding ground for conflicts. Advanced Persistent Threats (APTs) pose a formidable chanllenge when directed at organizations and governments, exposing the entire network to substantial security risks. Employing network fornesics for attributing cyber-attacks and acquiring timely, credible forensic results is a fundamental challenge in maintaining cyber security. This paper introduces a Deep Learning-based network forensics framework for digitally identifying and tracking network attacks, providing a comprehensive overview of the network forensics process. Specifically, we extract network traffic and employ encryption to ensure the integrity and security of data. Subsequently, we apply feature filtering techniques to retain essential traceability information, and Deep Learning model parameters are automatically optimized using hyperparameter optimization techniques. Lastly, we develop a Multi-Layer Perceptual Deep Neural Network (MLP DNN) model with perceptual capabilities for detecting anomalous events within the network. We evaluated the framework’s effectiveness using the UNSW-NB15 dataset. The experiments demonstrate that the proposed framework is applicable to APT attack forensics scenarios. In comparison to other AI methods, our framework excels in discovering and tracking network attack events with high performance.

In this paper, we study the problem of maximal biclique enumeration on large signed bipartite graphs. Given a signed bipartite graph$G=(U,V,E,s)$, a parameter$\theta \in [0.5, 1.0]$, our goal is to efficiently enumerate all maximal$\theta$-bicliques in$G$, where a maximal$\theta$-biclique$B(L,R)$is a complete subgraph of$G$with (1) the proportion of positive neighbors for each vertex in$B$is at least$\theta$, and (2)$B$is not contained in another biclique$B^{\prime }$, while$B^{\prime }$also satisfies (1). This problem has many applications, such as biclustering for genes, recommendation of similar groups, collaboration in communities, etc. However, it is computationally challenging due to its #P-completeness. Besides, we prove that even determining the maximality of a$\theta$-biclique is NP-hard. To the best of our knowledge, there is no efficient and scalable solution to this problem in the literature. In this paper, we first propose a branch-and-bound framework, namely${\sf MSiBE}$, which enumerates all maximal$\theta$-bicliques in a depth-first manner. Then, we develop three effective optimizations to improve the performance of${\sf MSiBE}$. (1) The local information of each search space is utilized to enhance the pruning capacity. (2) When expanding the partial biclique, we always focus on the side with fewer candidates first, by which fruitless search branches can be skipped early. (3) We implement${\sf MSiBE}$with efficient array reordering techniques and set intersection strategy. To further accelerate the computation, we introduce useful graph reduction techniques. Comprehensive performance studies on 10 real datasets demonstrate that our proposals can significantly outperform the baseline methods by up to 3 orders of magnitude.

The complexity and ongoing evolution of Advanced Persistent Threats (APTs) compromise the efficacy of conventional cybersecurity measures. Firewalls, intrusion detection systems, and antivirus software, which are dependent on static rules and predefined signatures, are increasingly ineffective against these sophisticated threats. Moreover, the use of system audit logs for threat hunting involves a retrospective review of cybersecurity incidents to reconstruct attack paths for attribution, which affects the timeliness and effectiveness of threat detection and response. Even when the attacker is identified, this method does not prevent cyber attacks. To address these challenges, we introduce ThreatInsight, a novel early-stage threat detection solution that minimizes reliance on system audit logs. ThreatInsight detects potential threats by analyzing IPs captured from HoneyPoints. These IPs are processed through threat data mining and threat feature modeling. By employing fact-based and semantic reasoning techniques based on the APT Threat Intelligence Knowledge Graph (APT-TI-KG), ThreatInsight identifies and attributes attackers. The system generates analysis reports detailing the threat knowledge concerning IPs and attributed attackers, equipping analysts with actionable insights and defense strategies. The system architecture includes modules for HoneyPoint IP extraction, Threat Intelligence (TI) data analysis, attacker attribution, and analysis report generation. ThreatInsight facilitates real-time analysis and the identification of potential threats at early stages, thereby enhancing the early detection capabilities of cybersecurity defense systems and improving overall threat detection and proactive defense effectiveness.

As a fundamental graph problem, community search is applied in various areas, e.g., social networks, the world wide web, and biology. A common requirement from real applications is to return a community with a bounded size while most existing solutions do not constrain community size. Recent studies on size-constrained community search still have some critical issues, e.g., the existence of a better cohesiveness objective, some queries returning empty results, and inefficiency on partial queries. Thus, in this paper, we study the size-constrained truss community search (STCS). Given a graph$G$, a query vertex$q$, and size constraint$[l,h]$, the STCS problem aims to find a subgraph containing$q$with the largest min-support among all connected subgraphs having at least$l$and at most$h$vertices. We prove the STCS problem is NP-hard and APX-hard unless P = NP. An effective heuristic is proposed to quickly find a high-quality initial result. Then, a branch and bound algorithm is introduced to find the exact result, with novel optimizations, e.g., budget-cost-based bounding and branching strategies. Extensive experiments verify that the community quality returned by our algorithm is better and our algorithm is faster by up to 5 orders of magnitude, compared with the state-of-the-art.

Within the paradigm of forthcoming 6G network infrastructures, unmanned aerial vehicles (UAVs), functioning as principal conveyances, are projected to emerge as pivotal enablers in the nascent domain of the low-altitude economy. UAVs are poised to embrace various innovative applications, including latency-sensitive and compute-intensive services. However, UAVs are constrained by their energy capacity and computational resources, rendering them insufficient for fulfilling the increasingly rigorous service demands in the future. To address these challenges, our investigation focuses on the innovative UAV-based Vehicular Edge Computing (UVEC) framework, incorporating Vehicular Edge Computing (VEC) in UAV systems to bolster service reliability. A UAV can enhance its mission duration by dynamically selecting suitable vehicles for computation offloading and adaptively adjusting the task offloading ratio between vehicles and the edge server. By integrating vehicle selection and task offloading scheduling in the UVEC framework, we investigate the optimization of energy efficiency while satisfying the statistical delay and the buffer constraints for UAVs. To deal with the proposed problem, a distributed algorithm is designed by jointly considering the vehicle selection for task offloading radio to vehicles and the edge server. The stochastic network calculus (SNC) is employed to derive performance bounds for the statistical delay and constraints, enabling robust analysis and optimization of network performance. After that, we leverage linear transformation techniques to reformulate the original problem into a linear framework, enabling the application of the Alternating Direction Method of Multipliers (ADMM) algorithm to efficiently solve the transformed problem. Theoretical analysis and simulation results show that our algorithm converges while effectively satisfying service reliability constraints within the desired targets, outperforming benchmark schemes in terms of efficiency while meeting task delay and error-rate bounded constraints.

In the field of smart justice, handling legal cases through artificial intelligence technology is a research hotspot. Traditional judgment prediction methods are mainly based on feature models and classification algorithms. The former is difficult to describe cases from multiple angles and capture the correlation information between different case modules, while requires a wealth of legal expertise and manual labeling. The latter is unable to accurately extract the most useful information from case documents and produce fine-grained predictions. This article proposes a judgment prediction method based on tensor decomposition with optimized neural networks, which consists of OTenr, GTend, and RnEla. OTenr represents cases as normalized tensors. GTend decomposes normalized tensors into core tensors using the guidance tensor. RnEla intervenes in a case modeling process in GTend by optimizing the guidance tensor, so that core tensors represent tensor structural and elemental information, which is most conducive to improving the accuracy of judgment prediction. RnEla consists of the similarity correlation Bi-LSTM and optimized Elastic-Net regression. RnEla takes the similarity between cases as an important factor for judgment prediction. Experimental results on real legal case dataset show that the accuracy of our method is higher than that of the previous judgment prediction methods.

Graph classification plays an important role in a wide range of applications from biological prediction to social analysis. Traditional graph classification models built on graph kernels are hampered by the challenge of poor generalization as they are heavily dependent on the dedicated design of handcrafted features. Recently, graph neural networks (GNNs) become a new class of tools for analyzing graph data and have achieved promising performance. However, it is necessary to collect a large number of labeled graph data for training an accurate GNN, which is often unaffordable in real-world applications. Therefore, it is an open question to build GNNs under the condition of few-shot learning where only a few labeled graphs are available. In this article, we introduce a new Structure-aware Prototypical Neural Process (SPNP for short) for a few-shot graph classification. Specifically, at the encoding stage, SPNP first employs GNNs to capture graph structure information. Then, SPNP incorporates such structural priors into the latent path and the deterministic path for representing stochastic processes. At the decoding stage, SPNP uses a new prototypical decoder to define a metric space where unseen graphs can be predicted effectively. The proposed decoder, which contains a self-attention mechanism to learn the intraclass dependence between graphs, can enhance the class-level representations, especially for new classes. Furthermore, benefited from such a flexible encoding-decoding architecture, SPNP can directly map the context samples to a predictive distribution without any complicated operations used in previous methods. Extensive experiments demonstrate that SPNP achieves consistent and significant improvements over state-of-the-art methods. Further discussions are provided toward model efficiency and more detailed analysis.

Federated learning offers a partial safeguard for participants’ data privacy. Nevertheless, the current absence of an efficient privacy-preserving federated learning technology tailored for the Internet of Things (IoT) poses a challenge. Numerous privacy-preserving federated learning frameworks have been proposed, primarily relying on homomorphic cryptosystems, yet their suitability for IoT remains limited. Furthermore, the application of federated learning in IoT confronts two significant obstacles: mitigating the substantial communication costs and communication failure rates, and effectively discerning and utilizing high-quality data while discarding low-quality data for collaborative modeling purposes. In order to address these challenges, this paper introduces a privacy-preserving optimal aggregation federated learning framework that relies on the utilization of the multi-key EC-ElGamal cryptosystem (MEEC) and the federated sum optimization algorithm (FSOA), which are characterized by their lightweight nature and fair properties. The proposed MEEC approach aims to tackle the issue of multi-key collaborative computing within the context of federated learning, thereby resulting in reduced communication costs and enhanced communication efficiency. This is achieved through the leverage of the EC-ElGamal cryptosystem, which is known for its ability to generate short keys and ciphertexts. Furthermore, this paper presents a dynamic federated learning framework that incorporates user dynamic quit and join algorithms. The primary objective of this framework is to mitigate the adverse effects of communication failures and enhance power computation on IoT devices. Additionally, an FSOA is devised to ensure the acquisition of optimal training data, thereby preventing the inclusion of low-quality data in the training process. Subsequently, the proposed scheme undergoes rigorous security analysis and performance evaluation. The obtained results unequivocally demonstrate that our scheme outperforms existing solutions in terms of security, practicality, and efficiency with lower communication and computational costs.

As a programmable accelerator, SmartNIC provides more opportunities for algorithmic packet classification. Our aim in this work is to achieve both line-speed rule search and efficient rule update, two highly desired metrics for SDN data plane. We leverage the parallelism offered by the FPGA in SmartNIC following an algorithm/hardware co-design paradigm. Particularly, we first design an algorithm that constructs multiple trees for the rule set with a recursive rule sifting process. Unlike traditional space-cutting-based multi-tree construction, our rule sifting mechanism breaks the space constraints of rule-to-tree mapping and enables bounded height on each tree, thus providing the potential of bounded worst-case and line-speed performance. We then design a flexible hardware architecture with multiple systolic arrays that can be implemented in parallel on FPGA. Each systolic array works as a coarse-grained pipeline, and the multiple trees constructed earlier will be mapped onto these pipeline stages. This hardware-software mapping enables bounded worst-case rule searching. Additionally, incremental rule update is achieved simply by traversing the pipeline in one pass, with little and bounded impact on rule searching. Experimental results show that our design achieves an average classification throughput of 600.8/147.5 MPPS and an update throughput of 8.2/5.9 MUPS for 10k/100k-scale 5-tuple and OpenFlow rule sets.

Convolutional Neural Networks (CNNs) possess extensive applicability across diverse domains, particularly in the realm of image recognition. In light of the advent of machine learning as a service, the utilization of a well-trained CNN model by servers to execute image classification based on user queries has become a significant service, catering to a wide array of applications. Nevertheless, this convenience is accompanied by the inherent risk of data privacy and model privacy disclosure, which can have severe ramifications, particularly in the context of specialized scenarios like medical images and location images. Hence, how to perform classification for CNN with privacy protection emerges as a crucial research concern. Furthermore, the nonlinearity of CNN's activation function renders it unsuitable for homomorphic cryptosystems. In order to address these challenges, we put forth a privacy-preserving CNN classification scheme employing a distributed two trapdoors public-key cryptosystem (DT-PKC). Initially, we introduce a security protocol toolkit encompassing protocols for secure multiplication, secure activation function computing, and average pooling. In addition, we propose a novel continuous and derivative Tanhplus function as an approximation of the Relu function, aiming to enhance the accuracy of classification results. The secure activation function computing protocol utilizes the aforementioned Tanhplus function in conjunction with the proposed homogenization algorithm to compute the activation function. This protocol guarantees more precise and accurate output in the activation function calculation of CNN when operating under ciphertext. Furthermore, the adoption of the DT-PKC cryptosystem not only ensures privacy protection for CNN classification but also provides support for lightweight users and multiple keys. Finally, security analysis and performance evaluations demonstrate that the proposed scheme is secure, practicable, and efficient with high accuracy.

In recent years, deep learning (DL)-based methods have been widely used in code vulnerability detection. The DL-based methods typically extract structural information from source code, e.g., code structure graph, and adopt neural networks such as Graph Neural Networks (GNNs) to learn the graph representations. However, these methods fail to consider the heterogeneous relations in the code structure graph, i.e., the heterogeneous relations mean that the different types of edges connect different types of nodes in the graph, which may obstruct the graph representation learning. Besides, these methods are limited in capturing long-range dependencies due to the deep levels in the code structure graph. In this paper, we propose aMeta-path basedAttentionalGraph learning model for code vulNErability deTection, calledMAGNET. MAGNET constructs a multi-granularity meta-path graph for each code snippet, in which the heterogeneous relations are denoted as meta-paths to represent the structural information. A meta-path based hierarchical attentional graph neural network is also proposed to capture the relations between distant nodes in the graph. We evaluate MAGNET on three public datasets and the results show that MAGNET outperforms the best baseline method in terms of F1 score by 6.32%, 21.50%, and 25.40%, respectively. MAGNET also achieves the best performance among all the baseline methods in detecting Top-25 most dangerous Common Weakness Enumerations (CWEs), further demonstrating its effectiveness in vulnerability detection.

In this paper, we propose a simple and general lightweight approach named AL-RE2 for text matching models, and conduct experiments on three well-studied benchmark datasets across tasks of natural language inference and paraphrase identification. Firstly, we explore the feasibility of dimensional compression of word embedding vectors using principal component analysis, and then analyze the impact of the information retained in different dimensions on model accuracy. Considering the balance between compression efficiency and information loss, we choose 128 dimensions to represent each word and make the model params 1.6M. Finally, the feasibility of applying depthwise separable convolution instead of standard convolution in the field of text matching is analyzed in detail. The experimental results show that our model’s inference speed is at least 1.5 times faster and it has 42.76% fewer parameters compared to similarly performing models, while its accuracy on the SciTail dataset of is state-of-the-art among all lightweight models.

Adversarial attack on point clouds plays a vital role in evaluating and improving the adversarial robustness of 3D deep learning models. Current attack methods are mainly applied by point perturbation in a non-manifold manner. In this paper, we formulate a novel manifold attack, which deforms the underlying 2-manifold surfaces via parameter plane stretching to generate adversarial point clouds. First, we represent the mapping between the parameter plane and underlying surface using generative-based networks. Second, the stretching is learned in the 2D parameter domain such that the generated 3D point cloud fools a pretrained classifier with minimal geometric distortion. Extensive experiments show that adversarial point clouds generated by manifold attack are smooth, undefendable and transferable, and outperform those samples generated by the state-of-the-art non-manifold ones.

The burgeoning landscape of blockchain technology has made the security of deployed smart contracts an imperative concern. While existing security measures excel in pre-deployment testing, they fall short in protecting smart contracts once they are deployed, leaving them susceptible to malicious attacks. In this paper, we propose a novel Smart Contract Firewall framework designed to bridge this security gap. Functioning as a dynamic gateway, the framework employs real-time transaction inspection through adaptable filtering rules, enabling the identification and rollback of malicious transactions as they occur. Our empirical analysis demonstrates the framework's efficacy in mitigating a majority of existing vulnerabilities in the deployed smart contracts. Although the added layer of security comes at a cost, we prove that the increased gas expenses could be limited to 30 % -50 % for most transactions. This trade-off, we argue, is a small price to pay for significantly enhanced security.

Deep neural networks often exhibit the overconfidence issue when encountering out-of-distribution (OOD) samples. To address this, leveraging large-scale pre-trained models like CLIP has shown promise. While CLIP has the capability to encode a vast array of interconnected concepts, current OOD detection methods based on it primarily focus on ID categories and a limited set of OOD categories. In this paper, we propose a novel approach that harnesses the power of WordNet to fully exploit the rich knowledge encapsulated within CLIP, resulting in enhanced OOD detection performance. Our methodology involves constructing a word tree that includes both in-distribution (ID) words and a large set of semantically similar OOD words selected from WordNet. By matching a test image with the concepts of the words in the word tree using CLIP, we estimate the probability of the image being classified as either ID or OOD. Furthermore, we introduce a conditional random field model to effectively handle both the parent-child and the sibling-sibling conflicts in the concept matching results. Extensive experiments under various ID/OOD settings demonstrate the effectiveness of our approach and its superiority over state-of-the-art methods.

ICN treats contents as the first citizens and faces serious scalability challenges. Especially for the content provider mobility scenario, the routing updates and routing efficiency should be cost-effective. This paper proposes a hierarchical name-based routing (HNR) for provider mobility. HNR focuses on how to find the mobile provider when given an immutable provider name. Based on the idea that the content provider generally moves within a certain topology area over a period of time, HNR first divides the topology into multiple partition topologies, then it adopts a hierarchical routing scheme that combines two suitable routing schemes for local mobility and global mobility. The experiments by simulation demonstrate that HNR achieves a good tradeoff between efficiency and scalability.

Traditional static taint analysis based on bytecode analysis such as GadgetInspector to detect deserialization vulnerabilities always faced precision problems. For example, missing the fact that taints flowing to members in called methods, type confusion, and chaotic inheritance relationships when detecting deserialization vulnerabilities, which would lead to many error results. To alleviate these problems, this paper considers three measures of improving precision of detecting deserialization vulnerabilities, including cross-function members data flow tracking, local variables and arguments types inference, and call chain subject inference based on inheritance relationships.

The content security policy (CSP) is a World Wide Web Consortium (W3C) standard, designed to prevent and mitigate security vulnerabilities, such as cross-site scripting (XSS) attacks, data injection attacks, and clickjacking attacks on websites. In this article, we present a newly discovered front-end Web attack that uses the current object storage services vulnerability of cloud vendors to bypass CSP. We selected the object storage services from two cloud vendors with the most users, i.e., Google and Amazon, to conduct systematic and large-scale research and analysis. Three cyberspace search engines are used to retrieve data, from which we analyze the consequence and damage range of this security breach. We focus on reporting four key aspects of this security breach: 1) how to use object storage services to bypass CSP; 2) analysis on the existence of such vulnerability in real-world websites; 3) analysis on the existing security vulnerabilities in current object storage services; and 4) the new strategy on object storage services that we propose to use to eliminate the discovered security threat.

Generating an accurate machine learning (ML) model is of great importance for the Internet of Vehicles (IoV). However, obtaining such a model is challenging due to the fact that subgroups of in-network vehicles receive data from different resources. A worthwhile investment then would be identifying those groups before inferring models. Similarity metrics are widely used to distinguish different groups. However, the efficiency of most existing similarity measurements is at the cost of increased computational complexity and decreased accuracy, making them unsuitable for IoV’s stringent conditions. To address this issue, we propose a computationally efficient method to measure the similarity of different vehicles, where a simplified version of Earth mover’s distance (EMD) is adopted. This distance metric is then embedded into a distributed clustering algorithm to learn the global pattern for vehicular systems. Our algorithm’s overall performance is measured using an asynchronous message delay simulator. Compared to the best algorithm of the state of the art, our proposed algorithm converges slightly slower (by less than 1%) but improves the clustering accuracy by as much as 20% with synthetic data. Additionally, real-world data collected from vehicles validates the efficiency of our proposed algorithm.

Adversarial attacks have been successfully extended to the field of point clouds. Besides applying the common perturbation guided by the gradient, adversarial attacks on point clouds can be conducted by applying directional perturbations, e.g., along normal and along the tangent plane. In this article, we first investigate whether adversarial attacks with these two orthogonal directional perturbations are more imperceptible than that with the gradient-aware perturbation. Second, we investigate the deeper difference between adversarial attacks with these two directional perturbations, and whether they are applicable to the same scenarios. Third, based on the verification results that the above two directional perturbations have different sensitiveness to curvature, we devise a novel normal-tangent attack (NTA) framework with a hybrid directional perturbation scheme that adaptively chooses the direction according to the curvature of the local shape around the point. Extensive experiments on two publicly available data sets, e.g., ModelNet40 and ShapeNet Part, with classifiers in three representative networks, e.g., PointNet++, DGCNN, PointConv, validate the effectiveness of NTA, and the superiority to the state-of-the-art methods.

With the rapid development of Internet of Things, the amount and distribution of malware has greatly increased. Internet of Things platform needs new defense technologies to protect users from new the increasing number and complexity of malware. This article extracts import Dlls and import APIs from the original portable executable (PE) file, and uses heterogeneous graph to describe higher-level semantic relationship between two PE files. Besides this we construct four static features to comprehensively describe PE file. Based on ensemble learning we develop a model called robust ensemble model based on semantic feature fusion (REMSF) which fuses five features mentioned above. To evaluate REMSF, we collect 5370 executable PE files from the real world for series of experiments, in which REMSF’s detection accuracy can reach 99.07%.

In recent years, the increasing prevalence of Intelligent Transportation Systems with advanced technologies has led to the emergence of many targeted forms of malware such as ransomware, Trojans, viruses, and malicious mining programs. And malware authors use policies like category disguise or family obfuscation in malware components to evade detection, which poses a great security threat to enterprises, government agencies, and Internet users. In this paper, we propose a malware family classification approach based on multimodal fusion and weight self-learning. Firstly, multiple modalities of malware such as byte, format, statistic, and semantic are fused in various ways to generate effective features. And then, we creatively add a weight self-learning mechanism of malware families into the classification model, which works by continuously calculating log-loss based on the family label and the probabilities predicted by each feature. The approach proves to achieve excellent classification performance on highly imbalanced malware family datasets with high efficiency and small resource overhead, which helps to identify and classify malware families and enhance the efficiency of massive malware analysis in Intelligent Transportation Systems.

Collaborative machine learning, especially Federated Learning (FL), is widely used to build high-quality Machine Learning (ML) models in the Internet of Vehicles (IoV). In this paper, we study the performance evaluation problem in an inherently heterogeneous IoV, where the final models across the network are not identical and are computed on different standards. Previous studies assume that local agents are receiving data from the same phenomenon, and a same final model is fitted to them. However, this “one model fits all” approach leads to a biased performance evaluation of individual agents. We propose a general approach to measure the performance of individual agents, where the common knowledge and correlation between different agents are explored. Experimental results indicate that our evaluation scheme is efficient in these settings.

The continuous increase and spread of malware have caused immeasurable losses to social enterprises and even the country, especially unknown malware. Most existing methods use predefined class samples to train models, which cannot handle unknown malware detection. In this paper, we formalize unknown malware detection as a Few-Shot Learning problem. However, the existing model cannot dynamically adjust the model parameters according to the samples and does not deeply consider the influence of the correlation between samples, so it achieves sub-optimal performance. We propose a Dynamic Prototype Network based on Sample Adaptation for few-shot malware detection (DPNSA). Specifically, we use dynamic convolution to realize dynamic feature extraction based on sample adaptation. Secondly, we define the class feature (prototype) as the mean of the dynamic embedding of all malware samples of each class in the support set. Then, a dual-sample dynamic activation function is proposed, which uses the correlation of the dual-sample to reduce the impact of unrelated features between samples on the metric. Finally, we use the metric-based method to calculate the distance between the query sample and the prototype to realize malware detection. Experiments show that our method outperforms the existing few-shot malware detection models and achieves significant improvement.

Graph neural networks (GNNs) are popularly used to analyze non-Euclidean graph data. Despite their successes, the design of graph neural networks requires heavy manual work and rich domain knowledge. Recently, neural architecture search algorithms are widely used to automatically design neural architectures for CNNs and RNNs. Inspired by the success of neural architecture search algorithms, we present a graph neural architecture search algorithm GraphNAS that enables automatic design of the best graph neural architecture based on reinforcement learning. Specifically, GraphNAS uses a recurrent network as the controller to generate variable-length strings that describe the architectures of graph neural networks, and trains the recurrent network with policy gradient to maximize the expected accuracy of the generated architectures on a validation data set. Moreover, based on GraphNAS, we design a new GraphNAS++ model using distributed neural architecture search. Compared with GraphNAS that generates and evaluates only one candidate architecture at each iteration, GraphNAS++ generates a mini-batch of candidate architectures and evaluates them in a distributed computing environment until convergence. Experiments on real-world datasets demonstrate that GraphNAS can design a novel network architecture that rivals the best human-invented architecture. Moreover, GraphNAS++ can speed up the design process at least five times by using the distributed training framework with GPUs.

It is a challenge to generate an accurate machine learning model in a distributed network due to the increased concern in data privacy and high cost in gathering all raw data. This paper presents an adaptive asynchronous distributed clustering algorithm for agents in wireless network to learn the global models, while the privacy is protected. Moreover, the communication cost and clustering quality can be adaptively balanced. The proposed clustering algorithm does not require the number of clusters to be pre-defined. To improve the accuracy of the global model, we propose a bounding boxes based method to fully utilize the shape information of clusters. In addition, we consider different knowledge levels of agent and different requirements about the global model. In experiments on randomly generated network topologies, we demonstrate that methods which do more extensive clustering in each cycle, and which exchange descriptions of cluster shape and density instead of just centroids and data counts, achieve more consistent clustering, in significantly shorter elapsed time. We also show that the proposed methods can learn the same number of clusters as the ground truth when clusters are well separated from each other.

Open-source cyber threat intelligence (OSCTI) is becoming more influential in obtaining current network security information. Most studies on cyber threat intelligence (CTI) focus on automating the extraction of threat entities from public sources that describe attack events. The cybersecurity knowledge graph aims to change the expression of threat knowledge so that security researchers can accurately and efficiently obtain various types of threat information for preliminary intelligent decisions. The attribution technology can not only assist security analysts in detecting advanced persistent threats, but can also identify the same threat from different attack events. Therefore, it is important to trace the attack threat actor. In this study, we used the knowledge graph technology, considered the latest research on cyber threat attack attribution, and thoroughly examined key related technologies and theories in the process of constructing and applying the advanced persistent threat (APT) knowledge graph from OSCTI. We designed a cybersecurity platform named CSKG4APT based on a knowledge graph. Inspired by the theory of ontology, we constructed CSKG4APT as an APT knowledge graph model based on real APT attack scenarios. We then designed an APT threat knowledge extraction algorithm for completing and updating the knowledge graph using deep learning and expert knowledge. Finally, we proposed a practical APT attack attribution method with attribution and countermeasures. CSKG4APT is not a passive defense method in traditional network confrontation but one that integrates a large amount of fragmented intelligence and can actively adjust its defense strategy. It lays the foundation for further dominance in network attack and defense.

IoT data exchange plays a vital role in supporting various applications and services with massive IoT devices. However, the existence of malicious devices threatens the integrity and reliability of the exchanged data. Trust management has been used to mitigate the impact of malicious devices in centralized and decentralized architectures. However, most of these traditional trust management systems bear computation, storage, and communication challenges. In this study, we propose a semi-centralized trust management system architecture based on blockchain in both single and multiple domains. The IoT devices are centralized organized by cloud servers who coordinately sustain a rating data ledger within each domain based the proposed rotation based consensus protocol in a decentralized manner to support cross-domain data exchange. A computational trust model is proposed by aggregating the direct and indirect trust information, where we elaborately design decay function, recommendation credibility and adaptable weights so as to calculate the trust value of dynamic malicious devices. Finally, we evaluate the proposed system model in various situations through simulation based experiments and compare it with two classical models in the literature. The experimental results demonstrate the effectiveness of the proposed trust model in identifying malicious devices and mitigating the influence of malicious devices.

Cyber Threat Intelligence (CTI), which is knowledge of cyberspace threats gathered from security data, is critical in defending against cyberattacks.However, there is no open-source CTI dataset for security researchers to effectively apply enormous CTI information for security analysis in the field of threat intelligence, particularly in the field of Chinese threat intelligence. As a result, for network security research and development, this article constructed a Chinese CTI entity relationship dataset–CDTier, which includes: 1) A threat entity extraction dataset composed of 100 CTI reports, 3744 threat sentences and 4259 threat knowledge objects; 2) A dataset for entity relation extraction including 100 CTI reports, 2598 threat sentences and 2562 knowledge object relations. CDTier is, as far as we know, the first CTI dataset. On the CDTier, we trained 4 models for threat entity extraction and relation extraction using well-established and widely used deep learning methods in the NLP. The results showed that the model trained on CDTier extracts knowledge objects and their relationships described in threat intelligence more accurately. This significantly minimizes threat intelligence analysts’ work while assessing threat intelligence.

The rapid growth of air traffic flow has brought the airspace capacity close to saturation and, at the same time, has resulted in great stress for air traffic controllers. The 4- D trajectory-based operation system is an important solution to problems in the current civil aviation field. The system mainly relies on accurate 4-D trajectory prediction technology to share trajectory information among air traffic control, airlines, and aircraft to achieve coordinated decision-making between flight and control. However, due to the complexity of trajectory data processing, the current 4-D trajectory prediction technology cannot meet actual needs. Therefore, a data generation and prediction network model (DGPNM) is proposed. It integrates the Wasserstein generative adversarial networks with gradient penalty (WGAN-GP) and long-short-term memory (LSTM) neu-ral networks. With its outstanding performance, the LSTM neural network is utilized in both the generation module and the prediction module. The proposed model generates plenty of sample data to enlarge the train set, so overfitting could be reduced in the process of LSTM training. Experimental results prove that compared with other classical methods, the altitude prediction accuracy in the proposed model far exceeds that in current research results, which improves the prediction accuracy of the 4- D trajectory.

Summary With advanced information technologies and industrial intelligence, Industry 4.0 has been witnessing a large scale digital transformation. Intelligent transportation plays an important role in the new era and the classic vehicle routing problem (VRP), which is a typical problem in providing intelligent transportation, has been drawing more attention in recent years. In this article, we study multidepot VRP (MDVRP) that considers the management of the vehicles and the optimization of the routes among multiple depots, making the VRP variant more meaningful. In addressing the time efficiency and depot cooperation challenges, we apply the artificial bee colony (ABC) algorithm to the MDVRP. To begin with, we degrade MDVRP to single‐depot VRP by introducing depot clustering. Then we modify the ABC algorithm for single‐depot VRP to generate solutions for each depot. Finally, we propose a coevolution strategy in depot combination to generate a complete solution of the MDVRP. We conduct extensive experiments with different parameters and compare our algorithm with a greedy algorithm and a genetic algorithm (GA). The results show that the ABC algorithm has a good performance and achieve up to 70% advantage over the greedy algorithm and 3% advantage over the GA.

Smart Grid 2.0 is the energy Internet based on advanced metering infrastructure and distributed systems that require an instantaneous two-way flow of energy information. Edge computing benefits from its proximity to the servers and edge nodes of the smart grid distributed systems, which can provide efficient and low latency information transmission to the smart grid. With the massive number of Internet of Things being used, the amount of real-time power usage information generated by that represents a huge challenge for edge computing. To improve the efficiency of information transmission and processing in power systems, this article combines different deep learning algorithms with edge computing to analyze and process distributed renewable energy generation and consumer power data in smart microgrid. Experiments on two real-world datasets from China and Belgium show that the proposed framework can obtain satisfactory prediction accuracy compared to existing approaches.

As the basis for the static security of the power grid, power load forecasting directly affects the safety of grid operation, the rationality of grid planning, and the economy of supply–demand balance. However, various factors lead to drastic changes in short-term power consumption, making the data more complex and thus more difficult to forecast. In response to this problem, a new hybrid model based on variational mode decomposition and long short-term memory with seasonal factors elimination and error correction is proposed in this article. Comprehensive case studies on four real-world load datasets from Singapore and the United States are employed to demonstrate the effectiveness and practicality of the proposed hybrid model. The experimental results show that the prediction accuracy of the proposed model is significantly higher than that of the contrast models.

Existing malware classification methods cannot handle the open-ended growth of new or unknown malware well because it only focuses on pre-defined malware classes with sufficient training data. Due to the superiority of the visualization method, some researchers use it for solving few-shot malware classification. However, the malware images generated by existing visualization methods contain insufficient semantic information. At the same time, existing few-shot models tend to converge to sharp minima resulting in poor generalization performance. By synthesizing the observations, we think that accurate and effective few-shot malware classification methods are affected by generated malware images and classification models, which can be called data and model levels, respectively. To solve the above problems, we propose a novel method from the Data and Model levels, which is used to classify new or unknown malware well, called DMMal. More specifically, we propose a multi-channel malware image generation method based on multi-view so that malware images can contain more prosperous information at the data level. In addition, we investigated adaptive sharpness-aware minimization in a few-shot scenario from the perspective of model optimization at the model level to minimize the loss value and sharpness simultaneously. This enhances the generalization ability of the model and improves the ability of the model to classify new or unknown classes. Experiments on two few-shot malware classification datasets show that the method proposed can improve the performance of few-shot malware classification from the data and model levels.

Privacy is currently one of the most concerned issues in Cyberspace. Tor is the most widely used system in the world for anonymously accessing Internet. However, Tor is known to be vulnerable to end-to-end traffic correlation attacks when an adversary is able to monitor traffic at both communication endpoints. In this paper, we present a set of novel Trapper Attacks that can be used to deanonymize user activities by both AS-level adversaries and Node-level adversaries in a Tor network. First, AS-level adversaries can exploit the occasional failures of censored network to selectively control entry guards of the Tor users. Second, the adversaries can exploit poor reliability of the Tor communication (e.g., natural churn) to compromise the exiting nodes and the anonymous path. Once the adversaries gain control of the routes, they can identify and inspect any traffic entering and leaving the Tor network, consequently, deanonymize a Tor user’s activity in the network. To demonstrate the effectiveness and feasibility of this attacks, we implemented a tool that can launch the proposed Trapper Attacks to automatic reveal communication relationships between a Tor user and its destinations running on a live Tor network. We also present a formal analysis framework to evaluate the integrity of the Tor network. With this framework, we successfully obtained quantitative estimates of Tor’s security vulnerability. The proposed Trapper Attacks are also designed to scale up in real-world Tor networks. Namely, it allows an adversary to perform deanonymization in honey relays effectively, and compromise the anonymity of Tor clients in real time. Our experimental results show that the proposed attacks succeed in less than 40 seconds achieving a 100% accuracy rate and a false positive rate close to 0.

Overconfident predictions on out-of-distribution (OOD) samples is a thorny issue for deep neural networks. The key to resolve the OOD overconfidence issue inherently is to build a subset of OOD samples and then suppress predictions on them. This paper proposes the Chamfer OOD examples (CODEs), whose distribution is close to that of in-distribution samples, and thus could be utilized to alleviate the OOD overconfidence issue effectively by suppressing predictions on them. To obtain CODEs, we first generate seed OOD examples via slicing&splicing operations on in-distribution samples from different categories, and then feed them to the Chamfer generative adversarial network for distribution transformation, without accessing to any extra data. Training with suppressing predictions on CODEs is validated to alleviate the OOD overconfidence issue largely without hurting classification accuracy, and outperform the state-of-the-art methods. Besides, we demonstrate CODEs are useful for improving OOD detection and classification.

Most of the existing methods define recommendation as regression or classification for user-item interactions and apply discriminative models. However, recommender systems suffer from interaction data sparsity and data noise problems in reality. Recent Generative Adversarial Network-based recommender systems have the potential to solve the aforementioned problems. The negative sampling methods use the generator to collect effective signals from a large amount of unlabeled data to alleviate the data sparsity problem, while they suffer from sparse rewards in the policy gradient training process. The vector reconstruction methods generate user-related vectors for data augmentation to enhance robustness, but they lead to redundant calculation and only take the user as a condition and ignore information conveyed by items. To alleviate the limitations of these methods, we propose a novel framework termed Personalized Recommendation with Conditional Generative Adversarial Networks (PRGAN) to consider both the user and the item subset as conditions and formulate conditional rating vector generation as a user-item matching problem. The sparsity of conditional rating vectors can be controlled in our method, which simplifies the discriminator’s learning task. Experiments are conducted on four datasets to evaluate the effectiveness of the proposed framework.

Clients (e.g., browsers) and servers require public key certificates to establish secure connections. When a client accesses a server, it needs to check the signature, expiration time, and revocation status of the certificate to determine whether the server is reliable. The existing solutions for checking certificate status either have a long update cycle (e.g., CRL, CRLite) or violate clients' privacy (e.g., OCSP, CCSP), and these solutions also have the problem of trust concentration. In this paper, we present PROCESS, an online privacy-preserving on-chain certificate status service based on the blockchain architecture, which can ensure decentralized trust and provide privacy protection for clients. Specifically, we design Counting Garbled Bloom Filter (CGBF) that supports efficient queries and BlockOriented Revocation List (BORL) to update CGBF timely in the blockchain. With CGBF, we design a privacy-preserving protocol to protect clients' privacy when they check the certificate statuses from the blockchain nodes. Finally, we conduct experiments and compare PROCESS with another blockchain-based solution to demonstrate that PROCESS is suitable in practice.

Due to the wide deployment of deep learning applications in safety-critical systems, robust and secure execution of deep learning workloads is imperative. Adversarial examples, where the inputs are carefully designed to mislead the machine learning model is among the most challenging attacks to detect and defeat. The most dominant approach for defending against adversarial examples is to systematically create a network architecture that is sufficiently robust. Neural Architecture Search (NAS) has been heavily used as the de facto approach to design robust neural network models, by using the accuracy of detecting adversarial examples as a key metric of the neural network’s robustness. While NAS has been proven effective in improving the robustness (and accuracy in general), the NAS-generated network models run noticeably slower on typical DNN accelerators than the hand-crafted networks, mainly because DNN accelerators are not optimized for robust NAS-generated models. In particular, the inherent multi-branch nature of NAS-generated networks causes unacceptable performance and energy overheads.To bridge the gap between the robustness and performance efficiency of deep learning applications, we need to rethink the design of AI accelerators to enable efficient execution of robust (auto-generated) neural networks. In this paper, we propose a novel hardware architecture, NASGuard, which enables efficient inference of robust NAS networks. NASGuard leverages a heuristic multi-branch mapping model to improve the efficiency of the underlying computing resources. Moreover, NASGuard addresses the load imbalance problem between the computation and memory-access tasks from multi-branch parallel computing. Finally, we propose a topology-aware performance prediction model for data prefetching, to fully exploit the temporal and spatial localities of robust NAS-generated architectures. We have implemented NASGuard with Verilog RTL. The evaluation results show that NASGuard achieves an average speedup of 1.74× over the baseline DNN accelerator.

Due to the large success in object detection and instance segmentation, Mask R-CNN attracts great attention and is widely adopted as a strong baseline for arbitrary-shaped scene text detection and spotting. However, two issues remain to be settled. The first is dense text case, which is easy to be neglected but quite practical. There may exist multiple instances in one proposal, which makes it difficult for the mask head to distinguish different instances and degrades the performance. In this work, we argue that the performance degradation results from the learning confusion issue in the mask head. We propose to use an MLP decoder instead of the "deconv-conv" decoder in the mask head, which alleviates the issue and promotes robustness significantly. And we propose instance-aware mask learning in which the mask head learns to predict the shape of the whole instance rather than classify each pixel to text or non-text. With instance-aware mask learning, the mask branch can learn separated and compact masks. The second is that due to large variations in scale and aspect ratio, RPN needs complicated anchor settings, making it hard to maintain and transfer across different datasets. To settle this issue, we propose an adaptive label assignment in which all instances especially those with extreme aspect ratios are guaranteed to be associated with enough anchors. Equipped with these components, the proposed method named MAYOR achieves state-of-the-art performance on five benchmarks including DAST1500, MSRA-TD500, ICDAR2015, CTW1500, and Total-Text.

Strangers build trustworthiness through reputation systems in various online platforms. A reputation system collects history ratings from users about an object/entity and aggregates them as a reputation score, for the reference of future potential users. The reputation score can accurately reflect the real quality of this object if all the ratings are fairly provided, otherwise it may mislead other users if the ratings are unfairly provided. In order to mitigate the impacts of unfair ratings, many unfair rating detection models have been studied in the recent years, through identifying and filtering out the unfair ratings. In this work, we aim to investigate the existing unfair rating detection models considering realistic application settings in an interactive approach where the process of the unfair ratings detection is conducted by involving the interactions between the application system designer and these models. Based on this idea, we design a customized interactive system (CIS) which can satisfy the customized demands of application system designers through five customized functions, i.e., customized scenes, customized attack, customized model, customized metrics, and customized result presentations. After a series of interactions, application system designers can obtain the detection model that best fulfills their demands and the corresponding optimal parameters. To present the applicability of the proposed CIS system, we analyze several typical reputation models in our experiments and the experimental results indicate that our work can effectively bridge the existing unfair rating detection models with realistic applications.

The advent of neural network (NN) based deep learning, especially the recent development of the automatic design of networks, has brought unprecedented performance gains at heavy computational cost. On the other hand, in order to generate a new consensus block, Proof of Work (PoW) based blockchain systems routinely perform a huge amount of computation that does not achieve practical purposes but to solving a difficult cryptographic hash puzzle problem.In this study, we propose a new consensus mechanism, Proof of Learning (PoLe), which directs the computation spent for block consensus toward optimization of neural networks. In our design, the training and testing data are released to the entire blockchain network and the consensus nodes train NN models on the data, which serves as the proof of learning. As a core component of PoLe, we design a secure mapping layer (SML) to prevent consensus nodes from cheating, which can be straightforwardly implemented as a linear NN layer. When the consensus on the blockchain network is achieved, a new block is appended to the blockchain. We experimentally compare the PoLe protocol with Proof of Work (PoW) and show that PoLe can achieve a more stable block generation rate, which leads to more efficient transaction processing. Experimental evaluation also shows the PoLe can achieve a stable block generation rate without significantly sacrificing training performance.

Identification of anomaly and malicious traffic in the Internet-of-Things (IoT) network is essential for the IoT security to keep eyes and block unwanted traffic flows in the IoT network. For this purpose, numerous machine-learning (ML) technique models are presented by many researchers to block malicious traffic flows in the IoT network. However, due to the inappropriate feature selection, several ML models prone misclassify mostly malicious traffic flows. Nevertheless, the significant problem still needs to be studied more in-depth that is how to select effective features for accurate malicious traffic detection in the IoT network. To address the problem, a new framework model is proposed. First, a novel feature selection metric approach named CorrAUC is proposed, and then based on CorrAUC, a new feature selection algorithm named CorrAUC is developed and designed, which is based on the wrapper technique to filter the features accurately and select effective features for the selected ML algorithm by using the area under the curve (AUC) metric. Then, we applied the integrated TOPSIS and Shannon entropy based on a bijective soft set to validate selected features for malicious traffic identification in the IoT network. We evaluate our proposed approach by using the Bot-IoT data set and four different ML algorithms. The experimental results analysis showed that our proposed method is efficient and can achieve >96% results on average.

The model of k -core and its decomposition have been applied in various areas, such as social networks, the world wide web, and biology. A graph can be decomposed into an elegant k -core hierarchy to facilitate cohesive subgraph discovery and network analysis. As many real-life graphs are fast evolving, existing works proposed efficient algorithms to maintain the coreness value of every vertex against structure changes. However, the maintenance of the k -core hierarchy in existing studies is not complete because the connections among different k -cores in the hierarchy are not considered. In this paper, we study hierarchical core maintenance which is to compute the k -core hierarchy incrementally against graph dynamics. The problem is challenging because the change of hierarchy may be large and complex even for a slight graph update. In order to precisely locate the area affected by graph dynamics, we conduct in-depth analyses on the structural properties of the hierarchy, and propose well-designed local update techniques. Our algorithms significantly outperform the baselines on runtime by up to 3 orders of magnitude, as demonstrated on 10 real-world large graphs.

In recent years, the popularity of IoT (Internet of Things) applications and services has brought great convenience to people's lives, but ubiquitous IoT has also brought many security problems. Among them, advanced persistent threat (APT) is one of the most representative attacks, and its continuous outbreak has brought unprecedented security challenges for the large-scale deployment of the IoT. However, important research on analyzing the attribution of APT malware samples is still relatively few. Therefore, we propose a classification method for attribution organizations with APT malware in IoT using machine learning. It aims to mark the real attacking organization entities to better identify APT attack activity and protect the security of IoT. This method performs feature representation and feature selection based on APT behavior data obtained from devices in the Internet of Things and selects the features with a high degree of differentiation among organizations. Then, it trains a multiclass model named SMOTE-RF that can better deal with imbalance and multiclassification problems. Our experiments on real dynamic behavior data are combined to verify the effectiveness of the method proposed in this paper for attribution analysis of APT malware samples and achieve good performance. Our method could identify the organization behind complex APT attacks in IoT devices and services.

The root cause of the insecurity for smart devices is the potential vulnerabilities in smart devices. There are many approaches to find the potential bugs in smart devices. Fuzzing is the most effective vulnerability finding technique, especially the coverage-guided fuzzing. The coverage-guided fuzzing identifies the high-quality seeds according to the corresponding code coverage triggered by these seeds. Existing coverage-guided fuzzers consider that the higher the code coverage of seeds, the greater the probability of triggering potential bugs. However, in real-world applications running on smart devices or the operation system of the smart device, the logic of these programs is very complex. Basic blocks of these programs play a different role in the process of application exploration. This observation is ignored by existing seed selection strategies, which reduces the efficiency of bug discovery on smart devices. In this paper, we propose a contribution-aware coverage-guided fuzzing, which estimates the contributions of basic blocks for the process of smart device exploration. According to the control flow of the target on any smart device and the runtime information during the fuzzing process, we propose the static contribution of a basic block and the dynamic contribution built on the execution frequency of each block. The contribution-aware optimization approach does not require any prior knowledge of the target device, which ensures our optimization adapting gray-box fuzzing and white-box fuzzing. We designed and implemented a contribution-aware coverage-guided fuzzer for smart devices, called StFuzzer. We evaluated StFuzzer on four real-world applications that are often applied on smart devices to demonstrate the efficiency of our contribution-aware optimization. The result of our trials shows that the contribution-aware approach significantly improves the capability of bug discovery and obtains better execution speed than state-of-the-art fuzzers.

With the widespread use of online social networks, billions of pieces of information are generated every day. How to detect new topics quickly and accurately at such data scale plays a vital role in information recommendation and public opinion control. One of the basic research tasks of topic detection is how to represent a topic. The existing topic representation models do not focus on how to select better differentiated words to represent topics, are still computer-centered, and do not effectively combine human intelligence and artificial intelligence (AI). To solve these problems, this article proposes a word-distributed sensitive topic representation model (WDS-LDA) based on hybrid human-AI (H-AI). The basic idea is that the distribution of words within a topic or among different topics has a great influence on the selection of topic expression words. If a word is evenly distributed among all documents of a certain topic, it indicates that the word is the common word of all documents in the topic, and it is more suitable to represent this topic. If a word is more evenly distributed among various topics, it indicates that the word is a common word of all topics, and cannot be used for the purpose of distinguishing among topics, becoming less suitable to represent any topic. At the same time, the human cognitive ability and cognitive models are introduced into topic representation based on H-AI. We introduce the user's modification of topic expression words into the topic model representation so that the topic model can learn human wisdom and become more and more accurate. Therefore, three different weights are introduced: inside weight; outside weight; and manual adjustment weight. The inside weight describes the uniform distribution of a word in the given topic, the outside weight describes the uniform distribution of a word in all topics, and the manual adjustment weight reflects whether a word is suitable as a representative vocabulary in the past manual adjustment. Tests using Sina microblog's actual data sets show that the WDS-LDA algorithm makes the representative words more important, the distinction among different topic words higher, and effectively improves the precision of subsequent algorithms, such as topic detection and topic evolutionary analysis using the topic model.

Internet of Things (IoT) has become one of the fastest-growing technologies and has been broadly applied in various fields. IoT networks contain millions of devices with the capability of interacting with each other and providing functionalities that were never available to us before. These IoT networks are designed to provide friendly and intelligent operations through big data analysis of information generated or collected from an abundance of devices in real time. However, the diversity of IoT devices makes the IoT networks’ environments more complex and more vulnerable to various web attacks compared to traditional computer networks. In this article, we propose a novel ensemble deep learning based web attack detection system (EDL-WADS) to alleviate the serious issues that IoT networks faces. Specifically, we have designed three deep learning models to first detect web attacks separately. We then use an ensemble classifier to make the final decision according to the results obtained from the three deep learning models. In order to evaluate the proposed WADS, we have performed experiments on a public dataset as well as a real-word dataset running in a distributed environment. Experimental results show that the proposed system can detect web attacks accurately with low false positive and negative rates.

In softwarized industrial networking, honeypot identification is very important for both the attacker and the defender. Existing honeypot identification relies on simple features of honeypot. There exist two challenges: The simple feature is easily simulated, which causes inaccurate results, whereas the advanced feature relies on high interactions, which lead to security risks. To cope with these challenges, in this article, we propose a secure fuzzy testing approach for honeypot identification inspired by vulnerability mining. It utilizes error handling to distinguish honeypots and real devices. Specifically, we adopt a novel identification architecture with two steps. First, a multiobject fuzzy testing is proposed. It adopts mutation rules and security rules to generate effective and secure probe packets. Then, these probe packets are used for scanning and identification. Experiments show that the fuzzy testing is effective and corresponding probe packet can acquire more features than other packets. These features are helpful for honeypot identification.

Stemming from the recent evolutionary progress of the 5G wireless communication and Internet of Things (IoT) relevant technologies, the vision of the Internet of Connected Vehicles (IoCV) has become more apparent. On the basis of the state-of-the-art IoCV conceptual implementations, the location of the vehicles is one of the essential driven data of IoCV, and the location privacy issue needs to be taken into account. However, when scrutinizing into IoCV, noticeable challenges of location-aware scenario has raised. For IoCV, location is more than just query criteria like in Location-based Services (LBSs) of mobile Internet. It is also the underpinning data of various types of IoCV underlying mechanisms and functions. This difference makes preserving location privacy in IoCV quite different from the traditional privacy scenarios. In this paper, an overall analysis of end-user location privacy in IoCV was performed. To solve the location privacy dilemma, we proposed an intent prediction-based approach named LocJury, which benefits from the emerging concept of Intent-based Networking (IBN). LocJury provides location privacy by learning and estimate the intent of location access and will penalize those malicious location accesses. By simulating the conceptual IBN-based IoCV application scenario, which relies on the location accesses, the performance of LocJury is evaluated under various circumstances. The simulation result verified the effectiveness of our proposed method.

With the gradual advancement of smart city construction, various information systems have been widely used in smart cities. In order to obtain huge economic benefits, criminals frequently invade the information system, which leads to the increase of malware. Malware attacks not only seriously infringe on the legitimate rights and interests of users, but also cause huge economic losses. Signature-based malware detection algorithms can only detect known malware, and are susceptible to evasion techniques such as binary obfuscation. Behavior-based malware detection methods can solve this problem well. Although there are some malware behavior analysis works, they may ignore semantic information in the malware API call sequence. In this paper, we design a joint framework based on local and global features for malware detection to solve the problem of network security of smart cities, called LGMal, which combines the stacked convolutional neural network and graph convolutional networks. Specially, the stacked convolutional neural network is used to learn API call sequence information to capture local semantic features and the graph convolutional networks is used to learn API call semantic graph structure information to capture global semantic features. Experiments on Alibaba Cloud Security Malware Detection datasets show that the joint framework gets better results. The experimental results show that the precision is 87.76%, the recall is 88.08%, and the F1-measure is 87.79%. We hope this paper can provide a useful way for malware detection and protect the network security of smart city.

Human comments have become an integral part on evaluating the effectiveness of online courses. Most nature language processing studies consider comments as a composition of statistical texts, which distorts its essence in semantic relation and emotional expression from other disciplines' definition. In order to enlarge its denotation and semantics in cross-discipline perspectives, we firstly define online comments as a complex model that could realize feeling communication, express semantic knowledge, prompt social interaction, and fertilize time character. The social-emotional semantic model (SESM) and its complete construction methods are also introduced to extract comment's social and emotional semantic meaning. Utilizing three user-based and topic-based emotional algorithms, the presented model makes it possible to generate topic-based and learner-based time series. Also, this study evaluates the possibility to visualize SESM on 67084 Chinese MOOC comments and 278 time series. The time-varying phenomenon in double time-series may help teachers determine the reason of the emotion change and then decide to conduct course adjustment or personalized instruction. Future learning analysis on comments should consider multiple semantics and emotional time series.

Public transports, such as subway lines and buses, offer affordable ride-sharing services and reduce the road network traffic. Extracting passengers’ preferences from their public transit choices is important to city planners but technically non-trivial. When traveling by taking public transits, passengers make sequences of transit choices, and their rewards are usually influenced by other passengers’ choices. This process can be modeled as a Markov Game (MG). In this paper, we make the first effort to model travelers’ preferences of making transit choices using MGs. Based on the discovery that passengers usually do not change their policies, we propose novel algorithms to extract reward functions from the observed deterministic equilibrium joint policy of all agents in a general-sum MG to infer travelers’ preferences. First, we assume we have the access to the entire joint policy. We characterize the set of all reward functions for which the given joint policy is a Nash equilibrium policy. In order to remove the degeneracy of the solution, we then attempt to pick reward functions so as to maximize the sum of the deviation between the the observed policy and the sub-optimal policy of each agent. This results in a skillfully solvable linear programming algorithm for the multi-agent inverse reinforcement learning (MA-IRL) problem. Then, we deal with the case where we have access to the equilibrium joint policy through a set of actual trajectories. We propose an iterative algorithm inspired by single-agent apprenticeship learning algorithms and the cyclic coordinate descent approach. We evaluate the proposed algorithms on both a simple Grid Game and a unique real-world dataset (from Shenzhen, China). Results show that when we have access to the full policy, our algorithm can efficiently recover most of the reward structure, especially the interaction of agents. In the case where we only have access to a set of sampled expert trajectories, our algorithm can provide an explanation of the expert trajectories. Measured with respect to the experts’ unknown reward function, the performance of the policy output by our algorithm is close to that of the expert policy.

Machine Learning (ML) plays very significant role in the Internet of Things (IoT) cybersecurity for malicious and intrusion traffic identification. In other words, ML algorithms are widely applied for IoT traffic identification in IoT risk management . However, due to inaccurate feature selection, ML techniques misclassify a number of malicious traffic in smart IoT network for secured smart applications. To address the problem, it is very important to select features set that carry enough information for accurate smart IoT anomaly and intrusion traffic identification. In this paper, we firstly applied bijective soft set for effective feature selection to select effective features, and then we proposed a novel CorrACC feature selection metric approach. Afterward, we designed and developed a new feature selection algorithm named Corracc based on CorrACC, which is based on wrapper technique to filter the features and select effective feature for a particular ML classifier by using ACC metric. For the evaluation our proposed approaches, we used four different ML classifiers on the BoT-IoT dataset. Experimental results obtained by our algorithms are promising and can achieve more than 95% accuracy.

Crowdsensing systems collect various types of data from sensors embedded on mobile devices owned by individuals. These individuals are commonly referred to as workers that complete tasks published by crowdsensing systems. Because of the relative lack of control over worker identities, crowdsensing systems are susceptible to data poisoning attacks which interfering with data analysis results by injecting fake data conflicting with ground truth. Frameworks like TruthFinder can resolve data conflicts by evaluating the trustworthiness of the data providers. These frameworks somehow make crowdsensing systems more robust since they can limit the impact of dirty data by reducing the value of unreliable workers. However, previous work has shown that TruthFinder may also be affected by the data poisoning attack when the malicious workers have access to global information. In this article, we focus on partially observable data poisoning attacks in crowdsensing systems. We show that even if the malicious workers only have access to local information, they can find effective data poisoning attack strategies to interfere with crowdsensing systems with TruthFinder. First, we formally model the problem of partially observable data poisoning attack against crowdsensing systems. Then, we propose a data poisoning attack method based on deep reinforcement learning, which helps malicious workers jeopardize with TruthFinder while hiding themselves. Based on the method, the malicious workers can learn from their attack attempts and evolve the poisoning strategies continuously. Finally, we conduct experiments on real-life data sets to verify the effectiveness of the proposed method.

With the development of Internet-of-Things (IoT) technology, various types of information, such as social resources and physical resources, are deeply integrated for different comprehensive applications. Social networking, car networking, medical services, video surveillance, and other forms of the IoT information service model gradually change people's daily lives. Facing the vast amounts of IoT information data, the IoT search technology is used to quickly find accurate information to meet the real-time search needs of users. However, IoT search requires using a large amount of user private information, such as personal health information, location information, and social relations information, to provide personalized services. Employing private information from users will encounter security problems if an effective access control mechanism is missing during the IoT search process. An access control mechanism can effectively monitor the access activities of resources and ensure that authorized users access information resources under legitimate conditions. This survey examines the growing literature on access control for an IoT search. Problems and challenges of access control mechanisms are analyzed to facilitate the adoption of access control solutions in real-life settings. This article aims to provide theoretical, methodological, and technical guidance for IoT search access control mechanisms in large-scale dynamic heterogeneous environments. Based on a literature review, we also analyzed the future development direction of access control in the age of IoT.

Trust management of the Internet of Connected Vehicles has been a hot topic in recent years with the rapid development of UGV technologies. However, existing resolutions based on trustworthiness verification among vehicles make the traffic event transmission quite inefficient. In this article, we assume that the deployed roadside units (RSUs) can provide efficient communication between any pair of RSU and vehicle and propose vehicle cash (Vcash), a reputation framework for identifying denial of traffic service, to resolve the trustworthiness problem in the application level of the Internet of Connected Vehicles. In our reputation framework, every vehicle communicates with the RSU directly for traffic event verification, and spreads verified traffic event notification. We borrow the idea of market trading, and set up trading rules to restrict the malicious vehicle's spread of false message, and to encourage vehicles to contribute to the traffic event monitoring and verification. To evaluate the effectiveness of our reputation framework, we conduct a simulation experimental. Our experiment results indicate that our proposal manages to avoid bogus event spread, and a vehicle in our framework has to contribute to the traffic event detection to normally employ the traffic service.

With the rapid development of smart cities, various types of sensors can rapidly collect a large amount of data, and it becomes increasingly important to discover effective knowledge and process information from massive amounts of data. Currently, in the field of knowledge engineering, knowledge graphs, especially domain knowledge graphs, play important roles and become the infrastructure of Internet knowledge-driven intelligent applications. Domain concept extraction is critical to the construction of domain knowledge graphs. Although there have been some works that have extracted concepts, semantic information has not been fully used. However, the excellent concept extraction results can be obtained by making full use of semantic information. In this article, a novel concept extraction method, Semantic Graph-Based Concept Extraction (SGCCE), is proposed. First, the similarities between terms are calculated using the word co-occurrence, the LDA topic model and Word2Vec. Then, a semantic graph of terms is constructed based on the similarities between the terms. Finally, according to the semantic graph of the terms, community detection algorithms are used to divide the terms into different communities where each community acts as a concept. In the experiments, we compare the concept extraction results that are obtained by different community detection algorithms to analyze the different semantic graphs. The experimental results show the effectiveness of our proposed method. This method can effectively use semantic information, and the results of the concept extraction are better from domain big data in smart cities.

With the development of the Internet of Things and big data technology, the intelligent transportation system is becoming the main development direction of future transportation systems. The time required for a given trajectory in a transportation system can be accurately estimated using the trajectory data of the taxis in a city. This is a very challenging task. Although historical data have been used in existing research, excessive use of trajectory information in historical data or inaccurate neighbor trajectory information does not allow for a better prediction accuracy of the query trajectory. In this article, we propose a deep learning method based on neighbors for travel time estimation (TTE), called the Nei-TTE method. We divide the entire trajectory into multiple disjoint segments and use the historical trajectory data approximated at the time level. Our model captures the characteristics of each segment and utilizes the trajectory characteristics of adjacent segments as the road network topology and speed interact. We use velocity features to effectively represent adjacent segment structures. The experiments on the Porto dataset show that the experimental results of our model are significantly better than those of the existing models.

With the development of Internet of Things (IoT) and cloud technologies, numerous IoT devices and sensors transmit huge amounts of data to cloud data centers for further processing. While providing us considerable convenience, cloud-based computing and storage also bring us many security problems, such as the abuse of information collection and concentrated web servers in the cloud. Traditional intrusion detection systems and web application firewalls are becoming incompatible with the new network environment, and related systems with machine learning or deep learning are emerging. However, cloud-IoT systems increase attacks against web servers, since data centralization carries a more attractive reward. In this article, based on distributed deep learning, we propose a web attack detection system that takes advantage of analyzing URLs. The system is designed to detect web attacks and is deployed on edge devices. The cloud handles the above challenges in the paradigm of the Edge of Things. Multiple concurrent deep models are used to enhance the stability of the system and the convenience in updating. We implemented experiments on the system with two concurrent deep models and compared the system with existing systems by using several datasets. The experimental results with 99.410% in accuracy, 98.91% in true positive rate (TPR), and 99.55% in detection rate of normal requests (DRN) demonstrate the system is competitive in detecting web attacks.

Many real-world human behaviors can be modeled and characterized as sequential decision-making processes, such as a taxi driver’s choices of working regions and times. Each driver possesses unique preferences on the sequential choices over time and improves the driver’s working efficiency. Understanding the dynamics of such preferences helps accelerate the learning process of taxi drivers. Prior works on taxi operation management mostly focus on finding optimal driving strategies or routes, lacking in-depth analysis on what the drivers learned during the process and how they affect the performance of the driver. In this work, we make the first attempt to establish Dynamic Human Preference Analytics. We inversely learn the taxi drivers’ preferences from data and characterize the dynamics of such preferences over time. We extract two types of features (i.e., profile features and habit features) to model the decision space of drivers. Then through inverse reinforcement learning, we learn the preferences of drivers with respect to these features. The results illustrate that self-improving drivers tend to keep adjusting their preferences to habit features to increase their earning efficiency while keeping the preferences to profile features invariant. However, experienced drivers have stable preferences over time. The exploring drivers tend to randomly adjust the preferences over time.

Deep learning has revolutionized the field of computer vision and image processing. Its ability to extract the compact image representation has taken the person re-identification (re-id) problem to a new level. However, in most cases, researchers are focused on developing new approaches to extract more fruitful image representation and use it in the re-id task. The extra information about images is rarely taken into account because the traditional person re-id datasets usually do not have it. Nevertheless, the research in multimodal machine learning has demonstrated that the utilization of the information from different sources leads to better performance. In this work, we demonstrate how a person re-id problem can benefit from the utilization of multimodal data. We have used the UAV drone to collect and label the new person re-id dataset, which is composed of pedestrian images and its attributes. We have manually annotated this dataset with attributes, and in contrast to the recent research, we do not use the deep network to classify them. Instead, we employ the continuous bag-of-words model to extract the word embeddings from text descriptions and fuse it with features extracted from images. Then the deep neural decision forest is used for pedestrians classification. The extensive experiments on the collected dataset demonstrate the effectiveness of the proposed model.

Deep learning methods, e.g., convolutional neural networks (CNNs) and Recurrent Neural Networks (RNNs), have achieved great success in image processing and natural language processing especially in high level vision applications such as recognition and understanding. However, it is rarely used to solve information security problems such as attack detection studied in this paper. Here, we move forward a step and propose a novel multi-channel intelligent attack detection method based on long short term memory recurrent neural networks (LSTM-RNNs). To achieve high detection rate, data preprocessing, feature abstraction, and multi-channel training and detection are seamlessly integrated into an end-to-end detection framework. Data preprocessing provides high-quality data for subsequent processing, then different types of features are extracted from the processed data. Multi-channel processing is used to generate classifiers by training neural networks with different types of features, which preserve attack features of input vectors and classify the attack from normal data. With the results of the classifier's attack detection, we introduce a voting algorithm to decide whether the input data is an attack or not. Experimental results validate that the proposed attack detection method greatly outperforms several attack detection methods that use feature detection and Bayesian or SVM classifiers.

The Internet of Things (IoT) is growing day by day, and new IoT devices are introduced and interconnected. Due to this rapid growth, IoT faces several issues related to communication in the edge computing network. The critical issue in these networks is the effective edge computing IoT device selection whenever there are several edge nodes to carry information. To overcome this problem, in this paper, we proposed a new framework model named SoftSystem based on the soft set technique that recommends useful IIoT devices. Then, we proposed an algorithm named Softsystemalgo. For the proposed system, three different parameters are selected: IoT Device Security (IDSC), IoT Device Storage (IDST), and IoT Device Communication Speed (IDCS). We also find out the most significant parameters from the given set of parameters. It is evident that our proposed system is effective for the selection of edge computing devices in the IoT network.

Bike-sharing systems become more and more popular in the urban transportation system, because of their convenience in recent years. However, due to the high daily usage and lack of effective maintenance, the number of bikes in good condition decreases significantly, and vast piles of broken bikes appear in many big cities. As a result, it is more difficult for regular users to get a working bike, which causes problems both economically and environmentally. Therefore, building an effective broken bike prediction and recycling model becomes a crucial task to promote cycling behavior. In this paper, we propose a predictive model to detect the broken bikes and recommend an optimal recycling program based on the large scale real-world sharing bike data. We incorporate the realistic constraints to formulate our problem and introduce a flexible objective function to tune the trade-off between the broken probability and recycled numbers of the bikes. Finally, we provide extensive experimental results and case studies to demonstrate the effectiveness of our approach.

In recent years, mobile racks or auto robots have been widely used in e-commerce warehouses where storage location assignment is a fundamental problem in the order picking process. The present storage location assignment strategies mainly allocate stocks into various racks according to a specific objective function or the relationships between stocks. These strategies include the random storage assignment strategy (RAS) and the good- clustering storage location assignment strategy (GCAS). In this paper, we first analyze the key factors that affect the efficiency of the order picking system.The results show that the rack- moved-number (RMN) is a significant factor in the order picking process. Then, we propose a genetic- algorithm (GA) based method for the storage location assignment problem which adopts RMN as its fitness function. To find a better solution, we take the natural deduplicated stock sequence of history orders (NDSSHO) as a seed to initialize the population of chromosomes. We also define a specific cross mutation strategy to avoid checking the validity of chromosomes by exchanging selected genes and adjusting new generated chromosomes. At last, we compare the RMN of our proposed method with RAS and GCAS. The experimental results show that the RMN of our proposed method is about 50% less than RAS and GCAS.

The burgeoning technology of Mobile Edge Computing (MEC) is attracting the traditional Location-Based Service (LBS) and Location Service (LS) to deploy due to its nature characters such as low latency and location awareness. Although this transplant will avoid the location privacy threat from the central cloud provider, there still exist the privacy concerns in the LS of MEC scenario. Location privacy threat arises during the procedure of the fingerprint localization, and the previous studies on location privacy are ineffective because of the different threat model and information semantic. To address the location privacy in MEC environment, we designed LoPEC, a novel and effective scheme for protecting location privacy for the MEC devices. By the proper model of the Radio Access Network (RAN) access points, we proposed the noise-addition method for the fingerprint data, and successfully induce the attacker from recognizing the real location. Our evaluation proves that LoPEC effectively prevents the attacker from obtaining the user's location precisely in both single-point and trajectory scenarios.

In resource exchange networks, nodes or agents may cooperate or compete with each other to maximize their own profits. In competitive networks, they may determine their exchange strategy selfishly, which makes it a challenge problem to design fair and efficient cooperation strategy. A popular resource exchanging mechanism called Max-Min Fairness-Based Coalitional (MMFC) is proposed in [1] to solve the problem in competitive networks, however it is still an open problem whether the mechanism is trustful if an agent lies about its resource information. In this paper, we demonstrate the trustfulness of the MMFC mechanism; combining theoretical analyses and numerical examples, we show that an agent cannot gain more benefit by misreporting its resource information.

Tor hidden services (HSs) are used to provide anonymity services to users on the Internet without disclosing the location of the servers so as to enable freedom of speech. However, existing Tor HSs use decentralized architecture that makes it easier for an adversary to launch DHT-based attacks. In this paper, we present practical Eclipse attacks on Tor HSs that allow an adversary with an extremely low cost to block arbitrary Tor HSs. We found that the dominant cost of this attack is IP address resources, the experimental results show that we can use only three IP addresses to eclipse an arbitrary HS with 100% success probability. To understand the severity of the Eclipse attack problems on Tor HSs, and its security implications, we present the first formal analysis to evaluate the extent of threat such vulnerabilities may cause and quantify the costs of Eclipse attacks involved in our attack via probabilistic analysis. Theoretical analysis suggests that adversaries with a modest number of IP address resources can block a large number of HSs at any time. Finally, we discuss countermeasures and future works.

Edge computing provides high-class intelligent services and computing capabilities at the edge of the networks. The aim is to ease the backhaul impacts and offer an improved user experience. However, the edge artificial intelligence exacerbates the security of the cloud computing environment due to the dissociation of data, access control, and service stages. In order to prevent users from carrying out lateral movement attacks in an edge-cloud computing environment, in this paper we propose a real-time lateral movement detection method, named CloudSEC, based on an evidence reasoning network for the edge-cloud environment. First, the concept of vulnerability correlation is introduced. Based on the vulnerability knowledge and environmental information of the network system, the evidence reasoning network is constructed, and the lateral movement reasoning ability provided by the evidence reasoning network is then used. The experiment results show that CloudSEC provides a strong guarantee for the rapid and effective evidence investigation, as well as real-time attack detection.

While various ransomware defense systems have been proposed to deal with traditional randomly‐spread ransomware attacks (based on their unique high‐noisy behaviors at hosts and on networks), none of them considered ransomware attacks precisely aiming at specific hosts, e.g., using the common Remote Desktop Protocol (RDP). To address this problem, we propose a systematic method to fight such specifically targeted ransomware by trapping attackers via a network deception environment and then using traceback techniques to identify attack sources. In particular, we developed various monitors in the proposed deception environment to gather traceable clues about attackers, and we further design an analysis system that automatically extracts and analyze the collected clues. Our evaluations show that the proposed method can trap the adversary in the deception environment and significantly improve the efficiency of clue analysis. Furthermore, it also helps us trace back RDP‐based ransomware attackers and ransomware makers in the practical applications.

Early data leakage protection methods for smart mobile devices usually focus on confidential terms and their context, which truly prevent some kinds of data leakage events. However, with the high dimensionality and redundancy of text data, it is difficult to detect the documents which contain confidential contents accurately. Our approach updates cluster graph structure based on CBDLP (Data Leakage Protection Based on Context) model by computing the importance of confidential terms and the terms within the range of their context. By applying CBDLP with pruning procedure which has been validated, we further remove the redundancy terms and noise terms. Actually, not only can confidential terms be accurately detected but also the sophisticated rephrased confidential contents are detected during the experiments.

Abstract This paper presents a parallel algorithm Parallel Extended Bloom Filter (PEBF) for exact multi‐pattern matching based on Bloom filter. To improve the throughput and parallelism of the algorithm, we divided the pattern set intoNsubsets where the length of patterns is the same, and different subsets would not intersect each other. We construct an EBF for each subset and useNthreads to simultaneously process the subsets in parallel. We implement our solution on the graphics processing unit, called G‐PEBF. Experimental results demonstrate that PEBF performs better than the Wu–Manber (WM) algorithm in terms of time and space. And G‐PEBF outperforms the G‐WM (WM algorithm implemented on graphics processing unit). The speedup of G‐PEBF is up to 60 times at peak performance and almost 10 times at worst performance to the PEBF algorithm. Copyright © 2014 John Wiley & Sons, Ltd.

Network anomaly detection is a classically difficult research topic in intrusion detection. However, existing research has been solely focused on the detection algorithm. An important issue that has not been well studied so far is the selection of normal training data for network anomaly detection algorithm, which is highly related to the detection performance and computational complexity. Based on our previous proposed TCM-KNN (Transductive Confidence Machines for K-Nearest Neighbors) anomaly detection method, which can detect anomalies with high detection rate and low false positive rate, we develop an instance selection mechanism for TCM-KNN based on EFCM (Extended Fuzzy C-Means) clustering algorithm in this paper, aiming at limiting the size of training dataset, thus reducing the computational cost of TCM-KNN and boosting its detection performance. We report the experimental results over real network traffic. The results demonstrate the instance selection method presented in this paper is effective for TCM-KNN and thus optimizing it as an effectively lightweight network anomaly detection scheme.

In this poster, based on our previous work in building a lightweight DDoS (Distributed Denial-of-Services) attacks detection mechanism for web server using TCM-KNN (Transductive Confidence Machines for K-Nearest Neighbors) and genetic algorithm based instance selection methods, we further propose a more efficient and effective instance selection method, named E-FCM (Extend Fuzzy C-Means). By using this method, we can obtain much cheaper training time for TCM-KNN while ensuring high detection performance. Therefore, the optimized mechanism is more suitable for lightweight DDoS attacks detection in real network environment.

Intrusion detection can be defined as the process of identifying malicious behavior that targets a network and its resources. An important problem in the field of intrusion detection is the management of alerts. This paper describes a realtime aggregation and correlation system named Alertclu. With the aid of similarity-based alert clustering analysing technology, Alertclu can improve the aggregation of intrusion detection system outputs and allow one to seamlessly incorporate additional information. In addition, Alertclu supports the operators by classifying alerts into true positives and false positives. The results of experiment show that the proposed system is able to reduce the numerous redundant alerts and effectively reduces the analyst operators' workload.

Spam fighting is a classic puzzle in network security. In the past decades, many filtering spam solutions have been proposed. However, currently the conventional techniques still suffer from high false positives and false negatives, especially the former which usually cannot be accepted by the end users. This paper proposes a novel online URL-based spam filter (UBSF) on the basis of analyses over the conventional especially the URL-based anti-spam techniques. UBSF identifies spam by comparing the similarity of the extracted URLs (universal resource locator) from them with the URLs from our user-oriented standard URL Library (SUL). Experimental evaluations both from the contrast experiment and the prototype based on UBSF demonstrate it can significantly raise the filtering accuracy, effectively reduce false positives and can be applied to online process and heavy traffic environment by reducing the computational cost than the state-of-the-art techniques.

Flash crowd events (FCEs) and malicious traffic including DDoS and worm attacks present a real threat to the stability of Web services. In this paper, we design a practical defense system that can provide some needed relief from the two types of events and protect the availability of Web services. A novel method of dynamic bandwidth arbitration using Generalized Vickrey auction based on microeconomics is proposed. By adopting this approach, not only the availability of Web services is improved but also the total utility of users can be maximized. Initial simulations have shown that this mechanism is promising direction to control both FCEs and malicious traffic. The presentation in this paper is a first step towards a more rigorous evaluation.