Federated Learning (FL) is a distributed learning framework that enables multiple parties to train models without disclosing potentially sensitive personal data, breaking down data silos and ensuring strong privacy protection. However, its decentralized nature means that individual participants can be compromised, potentially introducing backdoors into the model through data or gradient poisoning. Recent approaches have either focused on detecting backdoors during the aggregation process to reduce contamination or on erasing the backdoors post-training, using techniques such as model pruning. However, these methods often lead to a degradation in model performance and rely on several idealized conditions, such as having a clean dataset on the server side for detection, which is unrealistic in real-world scenarios. To address these challenges, we propose a novel method that serves as an effective defense against DBA(DISTRIBUTED BACKDOOR ATTACKS) in FL. Our method is designed to resist two types of attack settings and performs well even under client-side data heterogeneity. Specifically, our approach employs Self supervised learning to train a cleaner feature extractor and integrates a two stage malicious detection method to identify suspicious local updates, effectively mitigating backdoor attacks. We conduct comprehensive experimental evaluations across different datasets and attack scenarios, showing that our method outperforms state-of-the $\operatorname{art}($ SOTA) defenses against both single-round and continuous backdoor attacks. Notably, our approach provide a significant reduction in the backdoor effect(e.g., attack success rate is less than 14.5%) while maintaining nearly unchanged performance on the main task.

With the rapid development of information technology, data has become a key resource for driving social progress and technological innovation. However, the sharing and utilization of data face serious security challenges, especially in non-trusted environments such as third-party cloud servers. Searchable encryption allows users to search without revealing plaintext information. Boneh et al. introduced searchable encryption in the asymmetric cryptosystem, proposing Public Key Searchable Encryption (PEKS). Inner Product Function Encryption (IPFE) technology allows the decryptor to use the key to decrypt the ciphertext corresponding to the vector, thereby obtaining the inner product of the vector, without revealing any other information. Although these two technologies have performed well in some application scenarios, how to comprehensively apply these technologies in the more complex scenario of cloud data sharing remains a topic worthy of in-depth discussion. Therefore, this paper proposes a data sharing scheme IPFE-PEKS, which integrates index-based public key searchable encryption technology and inner product function encryption technology. By constructing an encryption index structure based on a complete binary tree, it realizes the search of encrypted data keywords. At the same time, it uses asymmetric inner product function encryption methods to protect the index structure and ensure secure access to the index. This paper not only elaborates on the design principles of IPFE-PEKS in detail but also analyzes and verifies its correctness and security. Performance comparison with other existing schemes shows that IPFE-PEKS ensures data security while maintaining high search efficiency.

Recommender systems are widely used artificial intelligence technologies that provide personalized recommendations to users from massive amounts of data. In the era of the Internet, recommender systems have become essential components in e-commerce, social media, news media, audio-video entertainment, and other fields. However, traditional recommender systems often rely on user's historical behavior or related attribute information for recommendations, which may lead to issues such as “over-recommendation” or “inefficient recommendation” due to their limited ability to uncover the latent connections between users and items. To address these issues, many deep learning-based recommendation models have emerged in recent years, such as DeepFM, NCF, etc,which have achieved significant improvements in recommendation performance. RippleNet and KGCN are two popular recommendation models among them. The RippleNet model employs graph neural networks to explore the interaction relationships between users and items as the basis for recommendations. On the other hand, the KGCN model utilizes knowledge graphs to better understand the semantic relationships between items.However, both models have their respective limitations. For instance, RippleNet only focuses on user representation while neglecting item representation, whereas KGCN overlooks the shortcomings in user representation. To further enhance recommendation performance, this paper proposes a new RNKN recommendation model that combines the strengths of RippleNet and KGCN, paying attention to both user and item representations to better uncover the latent connections between them. And apply the model to three datasets: MovieLens-1M (movies), Book Crossing (books) and Last.FM (music). Compared with RippleNet and KGCN, the AUC index of RNKN on the MovieLens-1M data set has increased by 0.4% and 1.4% respectively; the ACC index has increased by 0.45% and 1.2%; compared with the AUC index of RNKN on the Book-Crossing data set RippleNet and KGCN have increased by 5.9% and 1.9% respectively; the ACC index has increased by 3.4% and 3%; the AUC index of RNKN on the Last.FM dataset has increased by 1.8% and 0.1% respectively compared to RippleNet and KGCN; The ACC index increased by 3.0% and 1.4%.

Smart contracts have driven the development of blockchain technology, but their security vulnerabilities pose numerous problems for blockchain applications.The existing smart contract vulnerability detection methods based on deep learning are not comprehensive in detecting vulnerability categories, have low accuracy, and are limited to binary classification tasks. This paper proposes a smart contract vulnerability detection model called SCGRU combining CNN and bidirectional gating recurrent unit-attention mechanism (BiGRU-Attention) for Ethereum smart contracts to solve these problems. The model transforms the standardized data into word vectors representation of smart contracts with semantic information through the Word2Vec word embedding module. The feature extraction module takes the word vector as input. It uses CNN and BiGRU training to extract high-level abstract features and sequence features of the smart contract respectively. Following the BiGRU feature extraction, an attention mechanism is introduced to highlight the key features related to vulnerabilities of the smart contract code. The features extracted by CNN are concatenated with the features extracted by BiGRU-Attention as the input to the vulnerability classification module. The softmax function normalizes the vulnerability classification module to complete the detection of vulnerabilities in Ethereum smart contracts. The experiments in this paper show that SCGRU has a high accuracy rate of vulnerability detection on the current public dataset. Moreover, SCGRU can identify all categories of smart contract vulnerabilities in the dataset with an average accuracy rate of 92.64%, among which the precision of Infinite Loop vulnerability detection is 99.06%.

NFTs (Non-Fungible Tokens) are a type of digital asset based on blockchain technology that has become an attractive investment tool for many investors. This paper proposes a model based on knowledge graphs and LSTM (Long Short-Term Memory) networks to provide investors with a ranking prediction of future NFT returns, assisting traders in making more accurate investment decisions in the NFT market. To verify the usability and accuracy of the model, we collected data from 345 different types of NFTs on the OpenSea platform and conducted experimental validation, comparing our method with other benchmark methods to demonstrate the accuracy of our proposed method in practical scenarios.

The use of attack graphs in the study of network vulnerability assessment is a classical and effective way, which breaks the deficiency of traditional methods that can only do a static assessment based on the threat level of vulnerabilities. In order to assess the overall vulnerability of the network system, this paper studies different vulnerability assessment methods based on attack graphs and proposes the AGH model, which uses CVSS to statically assess the probability of vulnerability exploitation and combines the Hidden Markov Model and Viterbi decoding algorithm to calculate the maximum probability of attacker attack path. Finally, the feasibility of the method is verified by conducting experiments with a real laboratory network topology.

With the increment of cyber traffic, there is a growing demand for cyber security. How to accurately detect cyber intrusions is the hotspot of recent research. Traditional Intrusion Detection Systems (IDS), based on traditional machine learning methods, lacks reliability and accuracy. In this paper, we build an IDS model with deep learning methodology. Instead of the traditional machine learning used in previous researches, we think deep learning has the potential to perform better in extracting features of massive data considering the massive cyber traffic in real life. Therefore, we propose to train an IDS model based on Convolution Neural Networks (CNN), a typical deep learning method, using entire NSL-KDD dataset. We study the performance of the model using multi class classification to compare with the performance of traditional machine learning methods including Random Forest (RF) and Support Vector Machine (SVM), and deep learning methods including Deep Belief Network (DBN) and Long Short Term Memory (LSTM). The experimental results show that the performance of our IDS model is superior to the performance of models based on traditional machine learning methods and novel deep learning methods in multi-class classification. Our model improves the accuracy of the intrusion detection and provides a new research direction for intrusion detection.

Population topology of particle swarm optimization (PSO) has an important impact on solving performance of PSO. The more commonly used population topology is with static structure, such as fully connected structure and ring structure. In the process of evolution, the static population topology is always the same, which affects the information exchange between individuals of the population to a certain extent. In this paper, several feasible dynamic random population topologies are proposed based on the study of random population topology. In the PSO algorithm with dynamic random population topology, the neighborhood particles of a particle will evolve according to certain rules. In detail, a population topology is abstracted into an undirected connected graph which could be randomly generated according to predefined rule and degree. By tuning the rule and degree, the communication mechanisms evolve in the evolutionary process and the solving performance of PSO will be enhanced significantly. Furthermore, for the generalized portfolio selection model in the financial engineering field, the proposed several PSO algorithms are employed to solve the problems related to the generalized portfolio selection model, and the performance of them have been compared with the classic PSO variant in detail. The data of experiment is the weekly prices in a certain period which include the indices of HangSeng, DAX 100, FTSE 100, S&P 100 and Nikkei 225. The computational results demonstrate that the proposed dynamic random population topology could obviously improve the performance of PSO. It is especially worth noting that one proposed dynamic random population topology strategy shows an excellent performance on most data sets which could find good solutions to the generalized portfolio selection problems.

An important objective of wireless sensor network is to prolong the network life cycle, and topology control is of great significance for extending the network life cycle. Based on previous work, for cluster head selection in hierarchical topology control, we propose a solution based on fuzzy clustering preprocessing and particle swarm optimization. More specifically, first, fuzzy clustering algorithm is used to initial clustering for sensor nodes according to geographical locations, where a sensor node belongs to a cluster with a determined probability, and the number of initial clusters is analyzed and discussed. Furthermore, the fitness function is designed considering both the energy consumption and distance factors of wireless sensor network. Finally, the cluster head nodes in hierarchical topology are determined based on the improved particle swarm optimization. Experimental results show that, compared with traditional methods, the proposed method achieved the purpose of reducing the mortality rate of nodes and extending the network life cycle.

Dynamic deployment methods for wireless sensor network (WSN) can improve the quality of service (QoS) of the network by adjusting positions of mobile nodes. In the dynamic deployment problem model of this paper, not only the coverage rate of WSN but also the moving distance of mobile nodes is taken into consideration. This kind of model can be abstracted into multi-objective optimization problem, and particle swarm optimization (PSO) is introduced to solve this problem. In this paper, combined with previous work, an improved dynamic deployment method is proposed based on multi-swarm PSO. Specifically, we propose a discrete PSO to calculate the distance of mobile solutions, and a multi-swarm PSO is designed to optimize network performance for enhancing the QoS of deployment which includes higher coverage rate and lower energy consumption of mobile nodes. Experimental results demonstrate that the proposed method has a good performance in solving the WSN deployment problem.

The problem of portfolio selection in the field of financial engineering has received more attention in recent years. This paper presents a novel heterogeneous multiple population particle swarm optimization algorithm (HMPPSO) for solving a generalized Markowitz mean-variance portfolio selection model. The proposed HMPPSO is based on heterogeneous multiple population strategy, in which the whole population is divided into several sub-populations and all the sub-populations evolve with different PSO variants. The communication between the sub-populations is executed at regular intervals to maintain the information exchange inside the entire population and coordinate exploration and exploitation according to certain migration rules. The generalized portfolio selection model is classified as a quadratic mixed-integer programming model for which no computational efficient algorithms have been proposed. We employ the proposed HMPPSO to find the solution for the model and compare the performance of HMPPSO with several classic PSO variants. The test data set is the weekly prices from March, 1992 to September, 1997 including the following indices: Hang Seng in Hong Kong, DAX 100 in Germany, FTSE 100 in UK, S&P 100 in USA and Nikkei 225 in Japan. The computational results demonstrate that HMPPSO is much effective and robust, especially for problems with high dimensions, thus provides an effective solution for the portfolio optimization problem.

Recommendation algorithms in social networks have attracted much attention in recent years. Collaborative filtering recommendation algorithm is one of the most commonly used recommendation algorithms. Traditional user-based collaborative filtering recommendation algorithm recommends based on the user-item rating matrix, but the large amounts of data may cause low efficiency. In this paper, we propose an improved collaborative filtering recommendation algorithm based on community detection. Firstly, the user-item rating matrix is mapped into the user similarity network. Furthermore, a novel discrete particle swarm optimization algorithm is applied to find communities in the user similarity network, and finally Top-N items are recommend to the recommended user according to the communities. The experiments on a real dataset validate the effectiveness of the proposed algorithm for improving the precision, coverage and efficiency of recommendation.

The community structure is one of the most common and important attributes in complex networks. Community detection in complex networks has attracted much attention in recent years. As an effective evolutionary computation technique, particle swarm optimization (PSO) algorithm has become a candidate for many optimization applications. However, PSO algorithm was originally designed for continuous optimization. In this paper, an improved simple discrete particle swarm optimization (ISPSO) algorithm and a discrete particle swarm optimization with redefined operator (IDPSO-RO) algorithm are proposed in the discrete context of community detection problem. Furthermore, a community correcting strategy is used to optimize the results. The performance of the two algorithms is tested on three real networks with known community structures. The experiment results show that ISPSO and IDPSO-RO algorithms using community correcting strategy can detect community structures more efficiently without prior knowledge about the size of communities and the number of communities.

The problem of portfolio selection in the field of financial engineering has received more attention in recent years and many portfolio selection models has been proposed in succession. To solve a generalized Markowitz mean-variance portfolio selection model, this paper proposed four improved particle swarm optimization algorithms (RTWPSO-AD, RTWPSO-D, DRWTPSO-AD, DRWTPSO-D) based on the strategies of Random Population Topology. We abstract the topology of particle swarm optimization (PSO) into an undirected connected graph which can be generated randomly according to a predetermined degree. The topology changes during the evolution when Dynamic Population Topology strategy is adopted. By setting the degree, we can control the communication mechanisms in the evolutionary period, enhancing the solving performance of PSO algorithms. The generalized portfolio selection model is classified as a quadratic mixed-integer programming model for which no computational efficient algorithms have been proposed. We employ the proposed four algorithms to solve the model and compare the performance of them with the classic PSO variant. The computational results demonstrate that the population topologies of PSO have direct impacts on the information sharing among particles, thus improve the performance of PSO obviously. In particular, the proposed DRTWPSO-D shows an extraordinary performance in most set of test data, providing an effective solution for the portfolio optimization problem.

Recommendation technology has experienced its great popularity for resource recommendation in web-based social networks. This paper proposed an effective recommendation model based on communities and trust network (CTNRM). In the proposed model, the neighbors of the recommended user are selected from the users in the same community with the recommended user, the users in the recommended user's circle of friends and the users in the recommended user's trusted community. Experiments on the Epinions dataset demonstrate the feasibility and effectiveness of CTNRM. The experimental results validate the effectiveness of our proposed recommendation model for improving the precision and rating coverage especially for inactive users.

Because of its simplicity and effectiveness, collaborative filtering (CF) became one of the most successful recommendation algorithms. User-based CF is one classic method of CF algorithms. In order to solve the problem that common rating items are often too few to be used to effectively calculate the similarity of two users in user-based CF, we proposed an improved collaborative filtering model with item similarity called ISCF in this paper. In ISCF model, the similarity of items was considered in user-based collaborative filtering, which contributes to alleviate the problem of data sparsity and therefore calculate the similarity of user. Experimental results illustrate that our approach ISCF outperforms the average method and user-based CF. Compared with user-based CF, the average improvement in the percentage of ISCF at MAE and RMSE are 21.9% and 17.7%, respectively. In addition, our approach ISCF can predict more items than user-based CF, and the average improvement in the percentage of ISCF at prediction diversity is 33.86%.

With the increasing number and types of cloud services, it is difficult for cloud service users to evaluate these cloud services and select them to satisfy their requirements. Hence, appropriate recommendation mechanisms are needed to help them to select the optimal services in the set of cloud services with identical or similar functionality. In this paper, we propose a Trust-based Recommendation System in service-oriented Cloud computing(TRSC). We first present the architecture of TRSC which measures cloud services based on the trust of them. In TRSC, the trust value is the combination of direct trust and recommendation trust. Direct trust of a cloud service user A on a cloud service C is computed basing on the direct interaction of A with C. And the evaluation method for calculating recommendation trust is based on the interaction of cloud service users Bs with C where Bs are users who are trusted by A, correlative with A or authorities of the field. The experimental results showed that the proposed approach can get a better recommendation accuracy than traditional collaborative filtering.

Answer Set Programming (ASP) is widely used in many areas of Artificial Intelligence. A parallel answer set solving algorithm based on multi-core processor technology is proposed in this paper. The parallel algorithm is designed on the shared-memory parallel computing model which is the abstract model of multi-core processor. The algorithm can distribute the whole solving task to several threads separately run on different cores of the multi-core processor. To make this parallel algorithm more efficient, we implement load balancing among different threads by the technique of shared global queue. The experimental results show that the parallel algorithm can improve the solving efficiency by times with the growth of processor cores.

In many real-world classifications such as video surveillance, web retrieval and image segmentation, we often encounter that class information is reflected by the pairwise constraints between data pairs rather than the usual labels for each data, which indicate whether the pairs belong to the same class or not. A common solution is combining the pairs into some new samples labeled by the constraints and then designing a smoothness-driven regularized classifier based on these samples. However, it still utilizes the limited discriminative information involved in the constraints insufficiently. In this paper, we propose a novel semi-supervised discriminatively regularized classifier (SSDRC). By introducing a new discriminative regularization term into the classifier instead of the usual smoothness-driven term, SSDRC can not only use the discriminative information more fully but also explore the local geometry of the new samples further to improve the classification performance. Experiments demonstrate the superiority of our SSDRC.

Prioritized preference based decision making is pervasive in real problems solving. We propose a new paradigm of logic programming to handle prioritized preference. The paradigm is interpreted based on answer set semantics. We introduce two semantics to optimize answer sets of the programs. Then, we present the properties of those semantics by investigating their order characterization. Finally, compared with related works, it shows our new paradigm has strong expressive power of preference representation and reasoning.