Abstract:Cross-domain aspect-based sentiment analysis (ABSA) uses annotated samples from the source domain to help train ABSA tasks on the target domain. However, not all samples from the source domain are suitable for transfer training, and some samples may have negative transfer effects on the training of the transfer model, which requires sample screening. Existing cross-domain instance transfer methods consider a one-sided transfer basis, ignoring the synergistic effect between samples and affecting cross-domain generalisation performance. In order to solve the problems of insufficient domain-specific training samples and sample screening in cross-domain transfer for ABSA tasks, this study proposes a cross-domain ABSA method based on high-quality sample selection by combining high-reliability machine learning theories and domain adaptation methods in modelling with the open environment of multi-domain sentiment analysis. First, inter-domain and intra-domain high-quality sample selection metrics are designed to filter the source domain data at the domain level and the sample level in turn, which takes into account the advantages of the two sample selection granularities. Second, similarity metrics between source and target domains are comprehensively designed and efficiently calculated through a graph neural network. Finally, the scenarios of multi-source domain transfer are included in the discussion of cross-domain ABSA, and inter-domain joint adaptability scores are designed to select the multi-source domain combinations with high inter-domain synergies by balancing the overlap and difference of domain features. A cross-domain transfer task is designed on a benchmark dataset covering six domains, and experiments are conducted on three sub-tasks of ABSA to validate the effectiveness of the proposed method.

Abstract:Adversarial training is regarded as a core defense mechanism for enhancing the robustness of deep models, yet its inherent limitations significantly constrain its effectiveness in practical applications. Traditional adversarial training methods rely on fixed attack patterns to generate adversarial examples (AEs), leading to insufficient sample diversity, limited generalization capabilities, and difficulties in achieving an effective balance between robustness and clean accuracy. More crucially, existing adversarial training frameworks lack adaptive control over the training process, resulting in the robust overfitting phenomenon. To address these challenges, an evolutionary optimization-based adaptive adversarial training framework is proposed, named trade-off robustness and generalization via adaptive strategy optimization (TRG-ASO). It innovatively integrates a genetic algorithm into adversarial training and achieves progressive complexity escalation in AE generation through dynamic adjustment of attack strategies across different training phases. This mechanism not only enhances sample diversity but also effectively suppresses overfitting risks through early stopping enabled by strategy optimization records. Experiments on CIFAR series datasets demonstrate that, compared with traditional adversarial training methods, the proposed TRG-ASO framework maintains baseline classification performance while improving robustness against multiple attack paradigms and accelerating training convergence. This study provides new insights into the robustness-generalization trade-off in adversarial training, offering significant practical value for building trustworthy deep learning systems.

Abstract:Most graph anomaly detection methods leverage graph neural network (GNN) to learn from relatively high-quality graph data. Unfortunately, such ideal scenarios are rare in real-world applications, where most data suffer from issues such as missing labels, dynamic changes, and structural incompleteness, collectively referred to as dynamic incomplete graph (DIG). To address the challenge of performance degradation of GNN under extreme conditions, this study proposes an explainable dynamic incomplete graph anomaly detection (EXDIG) method. The core is a graph masked autoencoder framework optimized with strong-weak mutual information. This framework simulates real-world DIG scenarios by masking graph structures (nodes/edges) and node features. In addition, through the strong-weak mutual information (SWMI) loss, it captures the relationship between structure and features while maintaining structural integrity, reducing overfitting, and improving generalization. Furthermore, EXDIG enhances the interpretability of anomaly detection in DIGs by incorporating masked perturbations on nodes, edges, and features, enabling the identification of key components and providing transparent, trustworthy explanations for anomaly detection results. This study evaluates EXDIG on nine real-world graph datasets, and the results demonstrate its superiority over state-of-the-art methods across different levels of DIG scenarios and across various downstream tasks and representation learning evaluations, both supervised and unsupervised. Specifically, on the Amazon anomaly detection dataset, EXDIG achieves improvements over 13% and 15% in NMI and ARI, respectively. It maintains F1-score fluctuations within 5% across dynamic incompleteness ratios from 25% to 99%. Notably, EXDIG enables node-level interpretability in dynamic incomplete graphs.

Abstract:With the rapid development of information technology, fraudulent behaviors in multiple fields such as financial transactions, social networks, and review systems show an increasingly complex and diversified trend, which poses a serious challenge to traditional fraud detection techniques. Although current mainstream graph neural network-based methods perform well in single-agency data environments, cross-agency data sharing and collaboration are difficult due to the involvement of sensitive user information, which in turn limits the training effectiveness and generalization performance of the model. Federated learning, as an emerging privacy-preserving distributed learning paradigm, provides a feasible way for cross-agency collaborative training, but existing graph federated learning methods are mostly designed for general graph tasks, making them difficult to adapt to the class imbalance and data heterogeneity problems prevalent in fraud detection, resulting in poor performance in fraud sample identification. To address the above challenges, this study proposes a risk perception dynamic aggregation graph federated learning method (FedRPDA) for fraud detection, aiming to effectively deal with complex fraud risk event recognition across organizations. FedRPDA includes two key strategies: the typical risk dynamic aggregation strategy measures the structural risk intensity of fraudulent nodes in the client graph and combines it with a dynamic weight mapping mechanism with temporal decay characteristics to adaptively adjust the aggregation weights of clients, thus enhancing the global model’s ability to discriminate between normal samples and typical fraud samples under heterogeneous data conditions; the diversified risk average aggregation strategy integrates a variance perturbation-based feature enhancement mechanism for fraud samples with a global prototype-guided contrastive learning mechanism, which effectively improves the model’s ability to represent structurally diverse and scarce a typical fraud samples, promotes their convergence toward common anomalies in the feature space, and further enhances the model’s robustness in recognizing complex fraud risk scenarios. Experimental results on several real-world fraud detection datasets show that FedRPDA significantly outperforms existing graph federated learning baseline methods in terms of detection performance and training convergence efficiency, and demonstrates good generalization ability and practical application potential.

Abstract:Open-world continual learning (OWCL) aims to simulate real-world scenarios where tasks evolve continuously, categories change dynamically, and unseen samples are encountered. A well-designed OWCL model is expected not only to retain knowledge of learned tasks while acquiring new tasks but also to recognize unknown categories, thus achieving continuous and robust knowledge accumulation and generalization. However, most existing continual learning methods are built upon the closed-world assumption and cannot effectively cope with the category uncertainty and inter-task interference introduced by open categories. In particular, they show clear limitations in balancing knowledge stability and plasticity. Therefore, based on the formal definition of the OWCL problem, this study proposes a task-aware prompt-driven mixture-of-experts model (TP-MoE), which realizes dynamic modeling of task semantics and efficient scheduling of expert modules, thus supporting knowledge transfer and knowledge update. Specifically, TP-MoE introduces a plug-and-play task prompt aggregation mechanism and improves the gating strategy for expert routing, enabling the continual integration of historical and current task knowledge during task increments. At the same time, an adaptive open-boundary thresholding strategy is incorporated, which dynamically adjusts the decision boundaries of open categories according to the transfer between new and old knowledge, thus enhancing both open-category detection capability and known-category classification accuracy. Experimental results demonstrate that TP-MoE achieves state-of-the-art performance across various metrics on the Split-CIFAR100 and Open-CORe25 benchmarks, exhibiting strong robustness and generalization. This study provides a scalable and transferable framework for knowledge modeling and task scheduling in open-world continual learning.

Abstract:In regression tasks, numerical label noise can distort the true distribution of data and weaken the generalization ability of models. Data filtering is a commonly used approach that can reduce the impact of noise to some extent. However, it is prone to the issue of over-filtering, leading to the loss of effective samples and the shift of data distribution. This study presents a progressive interval correction (PIC) algorithm for regression label noise. The aim is to tackle the problem of sample loss caused by data filtering and effectively reduce the label noise level. First, based on the posterior distribution of the true labels, the validity conditions for label correction are established to ensure a reduction in the label noise level. Then, the labels that meet the validity conditions are corrected using the maximum a posteriori method. Finally, the labels are progressively corrected and optimized by gradually narrowing the range of the credible interval. Experimental results on both benchmark and real-world datasets demonstrate that the PIC algorithm can significantly reduce the noise level of data and effectively enhance the performance of models.

Abstract:Graph anomaly detection, as a critical task in graph data mining, aims to identify anomalous nodes that significantly differ from the majority of nodes in a network. Existing methods for graph anomaly detection typically adopt dataset-specific training paradigms, i.e., training a separate model for each dataset. However, such methods lack generalization capability across datasets and incur high training costs. To overcome these limitations, recent studies have begun to focus on the generalization potential of residual features. Such features are obtained by computing the difference between a node’s own representation and the representation after neighborhood propagation, which can largely offset dataset-specific semantic information and thus retain general information closely related to anomalous patterns. Despite initial progress in this direction, the modeling of residual features still faces the following key challenges: First, when computing the difference between a node’s representations before and after neighborhood propagation, the sparsity of neighbors and potential structural noise affect the reliability of the results to some extent. Second, the computation of representations relies on graph neural network (GNN) to learn local relationships, which makes it difficult to model global relationships that are also beneficial for anomaly detection, thus limiting the expressive power of residual features. To address these issues, this study proposes GRAD, a generalizable graph anomaly detection method via joint perception of global and local residual information. Specifically, based on GNN for modeling local node relationships, GRAD introduces a linear Transformer module to model global structural correlations among nodes in the feature space without relying on the original graph structure, thus obtaining node representations with global awareness. Then, GRAD transforms the representations into residuals between each node and its neighbors from both global and local perspectives, and integrates them to form dataset-independent general node representations. Extensive experiments on multiple public graph datasets from different domains verify the effectiveness of GRAD.

Abstract:Adversarial training, a key strategy for enhancing the adversarial robustness of deep neural network (DNNs), has been widely studied in image classification but lacks sufficient research in object detection. Traditional adversarial training often relies on projected gradient descent (PGD) for robust optimization of models. However, the iterative process of generating adversarial examples greatly prolongs model training, becoming a major bottleneck for deploying adversarial training in computationally intensive tasks like object detection. To address this, this study proposes an adversarial training method based on Nesterov’s accelerated gradient (NAG). By introducing the NAG momentum mechanism, algorithm convergence is accelerated. This method maintains detection accuracy comparable to PGD-trained models while significantly improving adversarial training efficiency. In addition, the main difference between object detection and image classification lies in object bounding box localization. However, it is observed that existing methods still focus on learning adversarial examples generated from classification loss, while neglecting the particularity of localization in object detection. To address this, an adaptive loss re-weighting strategy is designed to balance the number of adversarial examples derived from different tasks during training, thus enabling the model to focus on localization to enhance robustness. Experiments on the PASCAL VOC and MS-COCO datasets demonstrate the effectiveness of the proposed method compared with existing advanced adversarial training approaches for object detection.

Abstract:In recent years, deep neural networks have achieved significant progress across various domains. However, as typical black-box models, their internal mechanisms remain difficult for humans to understand, posing serious challenges in high-stakes applications such as medical diagnosis, financial risk management, and autonomous driving. Enhancing model interpretability has become one of the core issues in building highly trustworthy machine learning systems. Existing interpretability methods can be broadly classified into two categories: information-flow-based explanations, which focus on analyzing the importance of neurons or features, such as locating the pixel regions in an image that play a decisive role in classification results. Although these methods can reveal what the model has “attended to”, they often fail to provide cognitively meaningful, human-understandable semantics. In contrast, concept-based explanations construct semantic spaces to map internal model representations to interpretable concept structures, thus answering what the model has understood. These methods offer greater semantic depth and cognitive alignment, making them especially effective in improving semantic transparency and user trust. The fundamental lack of interpretability in deep learning stems from its deficiency in semantic representation. Therefore, constructing concept spaces and representation mechanisms aligned with human cognition has become a key breakthrough point in the development of interpretable models. This study presents a comprehensive survey of concept-based modeling methods in interpretable deep learning. Based on the stage at which interpretability is introduced, existing approaches are categorized into two major paradigms: post-hoc explanations, which extract semantic representations from trained models through techniques such as neuron dissection and semantic clustering; and intrinsic explanations, which incorporate structured priors or semantic constraints during training to endow models with built-in interpretability. Within this classification framework, this study systematically reviews representative modeling strategies and key methods, compares their performance in terms of semantic transparency and practical applicability, and summarizes current challenges and future research directions. The goal is to provide a structured reference and methodological guidance for understanding and building semantically interpretable deep learning models.

Abstract:The critical behavior of Gibbs sampling is a central issue in the theory of computational phase transitions. This study takes the hard-core model, a classical model, as an example to study the rapid convergence of Gibbs sampling before the critical point. In this model, given an n-vertex graph G with a maximum degree of Δ≥3 and a parameter λ≥0, each independent set S in graph G is sampled with a probability proportional to λ^|S|. This study investigates the canonical Gibbs sampling algorithm, Glauber dynamics, which implements this sampling. Under the critical condition λ<(Δ–1)^Δ–1/(Δ–2)^Δ, it is proven that the Glauber dynamics has an asymptotically optimal spectral gap of Ω(1/n), thus establishing that this classical sampling algorithm mixes rapidly up to the critical point. The rapid convergence of the Gibbs sampling process before the critical point is an important issue in Markov chain Monte Carlo (MCMC) theory. For this problem on the hard-core model, several proofs relying on advanced mathematical tools have been previously provided. This study offers a simplified combinatorial proof for this significant problem, introducing the idea of reductions from computational complexity to analyze the convergence rate of the sampling process.

Abstract:The CDCL algorithm for SAT solving is widely applied in the field of hardware and software verification, with restart being one of its core components. Currently, mainstream CDCL solvers often employ the “warm restart” technique, which retains key search information such as variable order, assignment preferences, and learnt clauses, and has a very high restart frequency. The warm restart technique tends to make CDCL solvers more inclined to visit the search space that is explored before restarts, which may lead to being trapped in an unfavorable local search space for a long time, lacking exploration of other regions. This study first tests the existing CDCL algorithms and confirms that under different initial search settings, the runtime for mainstream CDCL solvers exhibits significant fluctuations. To leverage this observation, the proposed “cold restart” technique forgets search information, specifically by periodically forgetting variable order, assignment preferences, and learnt clauses. Experimental results demonstrate that this technique can effectively improve mainstream CDCL algorithms. In addition, this study further extends its parallel version, where each thread explores different search spaces, enhancing the performance of the parallel algorithm. Moreover, the cold restart technique primarily improves the performance of sequential and parallel solvers on satisfiable instances, providing new insights for designing satisfiable-oriented solvers. Specifically, the proposed parallel cold restart technique improves the PAR2 score of PaKis on satisfiable instances by 41.81% on average. The parallel SAT solver named ParKissat-RS, which integrates the proposed ideas, wins the parallel track of the SAT competition with a significant margin of 24% over the runner-up.

Abstract:Since UML activity diagrams are essential tools for software requirements analysis, automating the process of generating UML activity diagrams helps reduce development time and labor costs. Existing approaches directly extract the elements and their relations of a UML activity diagram from unstructured natural language requirements either by manually constructing extraction rules or adopting data-driven approaches. However, these approaches typically consider only the syntactic features of software requirements statements while neglecting the semantic features. This leads to potential errors or redundancies in the automatically generated UML activity diagrams. Therefore, this study proposes an automated approach for constructing activity diagrams that combines the semantic features of software requirements statements in Chinese. This approach integrates the relevance between software requirements statements and UML activity diagrams, as well as the temporal properties of the software requirements statements, to extract UML activity diagram elements and their relations. It compensates for the shortcomings of existing approaches, which are easily disrupted by irrelevant information in the requirements and struggle to correctly represent various business activities. Experimental results on 100 industrial cases validate the superiority of the proposed approach over state-of-the-art approaches in terms of the completeness, correctness, and redundancy of automatically generated UML activity diagrams.

Abstract:The safety of autonomous driving systems (ADSs) is crucial for the implementation of autonomous vehicles (AVs). Therefore, ADSs must undergo thorough evaluation before being released and deployed publicly. Generating diverse, safety-critical test scenarios is a key task for ADS testing. Existing methods for generating ADS test scenarios include reproducing real-world traffic accidents or using search-based techniques. However, the accident-based scenario often fails to uncover safety violations in ADSs due to the gap between human driving and ADSs. The search-based approach tends to produce scenarios with high similarity because of the limitations of the search algorithm. To address these issues, this study proposes LEADE, a road network modeling-based safety-critical scenario generation and adaptive evolution method for ADSs. Specifically, it constructs abstract scenarios from user test requirements and generates concrete scenarios through road network modeling. LEADE then employs an improved adaptive evolutionary search to generate diverse safety-critical scenarios for testing the ADS. LEADE is implemented and evaluated on an industrial-grade full-stack ADS platform, Baidu Apollo. Experimental results demonstrate that LEADE can effectively and efficiently generate safety-critical scenarios and expose 10 diverse safety violations of Apollo. LEADE outperforms two state-of-the-art search-based ADS testing techniques by identifying 4 new types of safety-critical scenarios on the same roads.

Abstract:GitHub is one of the most popular open-source project management platforms. Due to the need for team collaboration, GitHub introduced an issue tracking function to facilitate project users in submitting and tracking problems or new feature requests. When resolving issues, contributors of open-source projects typically need to execute failure reproducing test cases to reproduce the problems mentioned in the issue and verify whether the issue has been resolved. However, empirical research conducted on the SWE-bench Lite dataset reveals that nearly 90% of issues are submitted without failure reproducing test cases, leading contributors to write additional failure reproducing test cases when resolving the issues, bringing additional work burden. Existing failure reproducing test case generation methods usually rely on stack trace information, but GitHub issues do not explicitly require such information. Therefore, this study proposes a failure reproducing test case generation method based on a large language model, aimed at automatically generating failure reproducing test cases for GitHub issues, assisting issue contributors in reproducing, understanding, and verifying issues, and improving the efficiency of issue resolution. This method first retrieves diverse code context information related to the issue, including error root functions, import statements, and test case examples, then constructs precise prompts to guide the large language model in generating effective failure reproducing test cases. This study conducts comparative and ablation experiments to verify the effectiveness of this method in generating failure reproducing test cases for GitHub issues.

Abstract:With the booming development and wide application of artificial intelligence, more and more deep learning-based Android malware detection methods and tools have been developed to detect newly emerged Android malware. However, deep learning models have been extensively proven to be vulnerable to adversarial attacks. Meanwhile, attacker shave started to propose adversarial attacks against Android malware detection methods to generate adversarial Android malware that can bypass detection. This study argues that the main reason current Android malware detection methods are vulnerable to such adversarial attacks is that these detectors are mostly built on single-modal features, which can be easily manipulated by attackers. Therefore, to improve the robustness of Android malware detection against adversarial attacks, the study proposes a robust Android malware detection method based on multi-modal fusion learning, namely RMDroid. RMDroid improves robustness in identifying adversarial malware without sacrificing accuracy in general Android malware detection. Specifically, RMDroid first extracts feature information from different modalities of Android APPs and then uses the corresponding deep learning models to sufficiently learn feature vectors that characterize the deep semantics of each modality. Finally, an odd-one-out network is employed to reduce or even eliminate the influence of interfered modal features on the final malware prediction, thus improving robustness against adversarial attacks. The experimental results show that RMDroid achieves higher performance across five effectiveness metrics and one robustness metric compared to all baseline detection methods. In particular, given the same FPR, the TPR value of RMDroid is more than 10% higher than that of the best baseline detection method. In the case of the state-of-the-art adversarial attack of HRAT, RMDroid achieves over 96% in robustness, which is significantly higher than the robustness of both MaMaDroid and MalScan.

Abstract:The government Q&A system can handle user queries in real-time, improving the efficiency of businesses and the public, while reducing the pressure of manual consultation. However, the service scenarios of the government Q&A system are diverse and require accurate and standardized expression of answers. Existing methods, which either utilize preset knowledge bases to generate answers or language models with limited scale, are unable to effectively understand consultations and generate trustworthy answers that are accurate and interpretable across multiple service scenarios. Therefore, this study proposes a government Q&A system based on a large language model to provide trustworthy government responses. The method employs a large language model specific to government service as the core module for content understanding and answer generation, assisted by an analysis guidance module and a domain knowledge base module. When generating answers, the large language model references the consulting analysis results provided by the analysis guidance module and the domain knowledge offered by the domain knowledge base module to produce answers that are accurate and consistent with the facts. The reference information during answer generation serves as a foundation to enhance the interpretability of the answers. A comprehensive dataset, containing multi-level and multi-granularity government public information, is collected and organized to construct the modules involved in the method and to test their effectiveness. This dataset includes 1901 documents and 10503 question-answer pairs. Finally, experiments verify that the prototype system, implemented based on the proposed method, can generate accurate and interpretable answers for user inquiries in multiple service scenarios, proving the effectiveness of each module in the system.

Abstract:This study focuses on the problem of procedure planning in instructional videos. Given the start and end observations, the task is to plan an action sequence that transforms the start state into the end state within the action space provided by the instructional videos. Instructional videos record and demonstrate the operational processes of various events. Each event includes a specific set of actions, forming the action space for that event. Therefore, the action space is composed of various subspaces corresponding to different events in instructional videos. Previous methods fail to effectively utilize the semantic information of events and overly rely on techniques such as reinforcement learning, resulting in complex training schemes and poorly explainable approaches. In contrast, this study considers the characteristics of instructional videos and proposes the classification-retrieval-based procedure planner (CPP), a pipeline that addresses procedure planning from coarse to fine. Specifically, the planner first identifies the event category based on the given observations, narrowing the action space to a smaller subspace. Then, action planning is performed within the selected subspace, which is significantly easier than planning in the entire action space. Moreover, this study introduces a hybrid planning method that combines retrieval and prediction approaches to generate the action sequence. The proposed method achieves competitive results on three popular procedure planning datasets of varying scales, establishing itself as a simple yet robust baseline for procedure planning.

Abstract:Blockchain, also known as a distributed ledger, is a prominent example of next-generation information technology. It has been widely applied in various fields, including finance, healthcare, energy, and government affairs. Privacy protection technologies within the blockchain that can be regulated not only safeguard users’ privacy and enhance trust but also prevent misuse of blockchain for illegal activities, ensuring compliance with regulations. Current privacy protection schemes for regulatable blockchains are typically based on bilinear pairing, which exhibit relatively low computational efficiency and fail to meet the demands of high-concurrency scenarios. To address these issues, this study proposes an efficient regulatable identity privacy protection scheme in blockchain. By designing a zero-knowledge proof to verify the consistency of the receiver’s identity without bilinear pairing, along with a traceable ring signature scheme, this approach effectively protects the identity privacy of both parties in transactions while maintaining the effectiveness of supervision. The experimental results indicate that when the number of ring members is set to 16, as required by Monero, the execution time of all algorithms in the efficient regulatable identity privacy protection scheme in blockchain is within 5 milliseconds. Compared to similar schemes, efficiency has improved by more than 14 times, and the message length has been reduced to 50% of the original scheme, demonstrating enhanced computational efficiency and a shorter message length.

Abstract:The performance and operational characteristics of the domain name system (DNS) protocol continue to attract significant attention from both the research community and network operators. In this study, data collected from a large-scale DNS recursive service is measured and analyzed to examine user access patterns and resolution behavior from the perspective of a major DNS operator. To handle the massive volume of DNS data, this study proposes a distributed parallel measurement mechanism and a big data-based storage and monitoring solution, enabling efficient processing and analysis. The characteristics of DNS data are systematically examined across several dimensions, including user request response rates, domain name request patterns, user distribution, and resolution outcomes. Several valuable insights are presented, offering meaningful guidance for DNS operation optimization and improved understanding of DNS behavior. Finally, based on the analysis of DNS cache hit rates, this study proposes a general framework for online anomaly detection tailored to large-scale DNS operators. The correctness and feasibility of the proposed framework are preliminarily verified.

Abstract:The (t, N) threshold multi-party private set intersection (TMP-PSI) protocol allows a given party’s data element x to appear in the private sets of no fewer than t–1 other parties. The data element x is then output as the intersection result, which is widely applied in scenarios such as proposal voting, financial transaction threat identification, and security assessment. Existing threshold multi-party private set intersection protocols suffer from low efficiency, high communication rounds, and a limitation that only a specific participant can obtain the intersection. To address these issues, this study proposes a threshold testing method based on robust secret sharing (RSS) and a TMP-PSI scheme combined with oblivious key-value store (OKVS), which effectively reduces both computational overhead and the number of communication rounds. To meet the demand for multiple participants to access the intersection information from their private sets, this study also proposes a second extended threshold multi-party private set intersection (ETMP-PSI) protocol, which modifies the share distribution method. Compared to the first scheme, the secret distributor and secret reconstructor do not incur additional communication rounds or computational complexity, allowing multiple participants to obtain the intersection elements from their private sets. The proposed protocol runs in 6.4 seconds (TMP-PSI) and 8.7 seconds (ETMP-PSI) in a three-party scenario with a dataset size of n=2¹⁶. Compared to existing threshold multi-party private set intersection protocols, the communication complexity between the reconstructor and distributor is reduced from O(nNtlognλ) to O(bNλ).

Abstract:In recent years, many studies have proposed using consensus mechanisms to enhance network layer security. However, existing consensus mechanisms have limitations, such as heavy key maintenance, inflexible trust expansion, and high authentication overhead. To address these issues, this study proposes a lightweight consensus framework based on source address validation. The framework optimizes consensus efficiency at multiple levels: First, among consensus nodes within the same domain, the framework uses authentic IP addresses as identities and achieves key aggregation by sharing the same key among nodes within the domain, thus efficiently reducing the number of keys that need to be maintained. Second, at the domain level, the framework constructs a trusted network alliance based on trust derived from authentic IP addresses and aggregates trusted domains through a prefix tree, thus further reducing the number of keys to maintain while enabling flexible trust expansion. Finally, at the node level, to address the issue of high authentication overhead, the framework designs a two-step authentication mechanism based on authentic IP addresses and symmetric keys, effectively reducing overhead and enabling a lightweight consensus process. Simulation experiment results show that the proposed framework can improve consensus throughput by 70% and reduce consensus latency by 40% on average, compared to the consensus mechanism based on ECDSA authentication, significantly improving consensus efficiency.

Abstract:With the rapid development of deep learning technologies and diffusion models, image and video generation models have displayed powerful capabilities to produce high-quality and diverse results. How to leverage these models for efficient and precise personalized generation has become a current research hotspot. Personalized image generation methods can combine text descriptions with specific concepts or subjects provided by users to enable the creation of customized images and meet the diverse needs of users for personalized visual content. This study reviews personalized image generation methods based on diffusion models, categorizing existing methods from the perspective of the generation target into single-subject driven generation and multi-concept combination generation. The former focuses on generating customized images according to individual subjects, emphasizing the accurate capture and reconstruction of the subjects’ visual features. The latter focuses on merging multiple concepts or subjects into a single image, addressing challenges like semantic alignment across concepts and visual consistency. This study provides a detailed analysis of representative work in personalized generation by combining specific tasks and application scenarios. Additionally, this study compares and summarizes common datasets, evaluation methods of generation models, and performance comparisons between different personalized generation methods. It further discusses the challenges that personalized generation methods face in practical applications and the future development directions, and offers a prospect for the research trends. This study aims to provide comprehensive references for researchers in relevant fields, fostering the development and innovation of personalized generation methods.