The federated learning technology provides a new method for data integration, which realizes sharing of a global model and prevent the leakage of user’s original data information. In order to resist data poisoning attack from some participants, ensure reliability and accuracy of the global model, and ensure fairness of the aggregation process in federated learning, we propose a blockchain-based fairness enhanced federated learning scheme. The accuracy of global model and fairness of the aggregation process is guaranteed by an adaptive aggregation algorithm which can defense data poisoning attack. The reliability of federated learning process is ensured by recording the entire process of the model training on the blockchain and using digital signatures. The privacy of each participant of federated learning is protected by public key encryption combined with the use of random numbers. Theoretical analysis and experiments show that the scheme can protect privacy of each participant, mitigate data poisoning attack and ensure the reliability and fairness of the entire federated learning process.
With the development of the big data era, the amount of data has entered an explosive growth phase. Limited by the constraints of cost, efficiency, and security of self-built storage systems, enterprises are forced to outsource files to cloud storage systems. However, the lack of file security and auditability in cloud storage systems continues to threaten the security of outsourced files. This paper designs and implements BFSOut, which is a secure file outsourcing method based on consortium blockchain. It uses Hyperledger Fabric and Interplanetary File System (IPFS) as the underlying storage engine, which solves the problem of cloud storage security issues. In BFSOut, in order to ensure the security of outsource files, the client-side offline block encryption is used. Furthermore, a dynamic hybrid encryption scheme is adopted, making the overall encryption effect more Efficient. Experimental performance analysis show that the system has good performance.
Blockchain is experiencing the transition from the first generation to the second generation, and smart contract is the symbol of the second generation blockchain. Under the background of the explosive growth of the second-generation blockchain platform and applications represented by smart contracts, frequent smart contract vulnerability events seriously threaten the ecological security of the blockchain, reflecting the importance and urgency of smart contract vulnerability detection. In this paper, we proposed a smart contract vulnerability detection method based on a Siamese network. We combined the Siamese network with Long Short-Term Memory (LSTM) Network neural network to complete the task of smart contract vulnerability detection. The Siamese network used in this paper consists of two subnetworks that share the same parameters onto a low dimension and easily separable feature space. Siamese network is now widely used in the field of image similarity and target tracking. In this paper, we improve the Siamese network so that it can be used for smart contract vulnerability detection. By comparing with previous research results, the model has better vulnerability detection performance and a lower false-positive rate.
Based on the influence of dark obstacles caused by insufficient light in an underground mine on the driving safety of an electric locomotive. This paper proposes an improved YOLOX target detection algorithm to effectively identify and classify the track obstacles of the unmanned electric mine locomotive. On the basis of the YOLOX target detection network, the CBAM attention module is added to the CSPDarket and the FPN part of the feature pyramid, and the loss function of YOLO head part is replaced by SIOU. The collected image data of track obstacles of electric locomotive under different lighting conditions are used as the training set. The Pytorch deep learning framework is used to construct an object detection model for training and verification. Experiments show that the average accuracy and recall rate of the improved YOLOX underground electric locomotive track obstacle detection model can reach 93.05% and 88.29%, and the speed is improved to 45.3 fps. Compared with other target detection models, this model can better realize the accuracy and real-time detection of underground electric locomotive track obstacles. It provides the basis for the intelligence of underground mine transportation equipment.
Over the past several years, due to the progression toward data-driven scientific disciplines, the field of Big Data has gained significant importance. These developments pose certain challenges in the area of efficient, effective, and secure management and transmission of digital information. This paper presents and evaluates a novel Distributed Ledger Technology (DLT) system, Fibereum, in a variety of use-cases, including a DLT-based system for Big Data exchange, as well as the fungible and non-fungible exchange of artwork, goods, commodities, and digital currency. Fibereum’s innovations include the application of non-linear data structures and a new concept of Lazy Verification. We demonstrate the benefits of these novel features for DLT system applications’ cost performance and their added resilience towards cyber-attacks via the consideration of several use cases.
Machine learning-based models are one of the main methods for detecting reentrant vulnerabilities. However, these models extract smart contract features only from a single form, resulting in incompleteness and inaccuracy of features. To address this problem, we propose a novel machine learning-based model for reentrant vulnerabilities detection. We extract and fuse features from abstract syntax trees, opcodes, control flow graph basic blocks, and combine machine learning algorithms for reentrant vulnerabilities detection. Additionally, to address the time-consuming problem of manual labeling, we also propose an approach for automatically adding dataset labels. We perform experiments on Smartbugs and SolidiFi-benchmark datasets and results show that our model outperforms existing models.
DNA is one of the most important information in every living thing. The DNA matching experiment is helpful for the study of paternity testing, species identification, gene mutation, suspect determination, and so on. How to study the DNA matching in the case of privacy protection has become the inevitable problems in the research of information security. The Hamming distance can reflect the similarity degree of two DNA sequences. The smaller the Hamming distance is, the more similar the two DNA sequences are. In this paper, the DNA sequence with $$l$$ length is encoded with a 0–1 string with $$3l$$ length, and the protocol of confidentially computing Hamming distance is designed, which calculated the matching degree of two DNA under the premise of protecting DNA privacy. In addition, in view of the criminal suspect DNA matching problem, we design a secure computation protocol against malicious adversaries using the zero-knowledge proof and the cut-choose method to prevent or find malicious behaviors, which can resist malicious attacks.
Mediation analysis was concerned with the decomposition of the total effect of exposure on the outcome into the indirect effects and the remaining indirect effects, through a given mediation. However, when longitudinal data including time varying exposure and mediator variables, the estimated causal effects are affected by time varying confounders. Standard generalized linear equations did not give unbiased estimates. In this paper, we introduced inverse probability weighting technique to adjust such time varying confounders. Considering that the amount of data may be small and the distribution is not uniform, we decide to use Bayesian Inference to estimate the Structural Equation Model (SEM) parameters, and finally estimates the causal effect through counterfactual thought. This paper summarized the relevant theoretical knowledge of this method, verified the feasibility of this method by using the simulated data, and compared the performance of different methods.
Cryptocurrency applications with blockchain technology as the underlying architecture have gradually developed into a new means of payment, and are expanding to all walks of life with the support of cryptography and consensus algorithms. Due to the disadvantages of low throughput and high latency, the blockchain has seriously hindered the widespread use of upper-layer applications and cannot meet the growing demand for users and transaction volumes. Drawing on the sharding idea of traditional databases, blockchain sharding, as a representative of on-chain scaling solutions, greatly improves the throughput of the blockchain system. At present, most of the network sharding schemes in the sharded blockchain adopt a strategy based on random sharding. This strategy does not take into account the performance of the node itself, resulting in large performance differences between different shards, further reducing the throughput of the entire system. In addition, the aggregation behavior of malicious nodes may also occur, reducing the security of the system. Aiming at the performance of each node, this paper proposes a sharding strategy based on the approximate ideal solution model (TOPSIS). Through the TOPSIS model, the nodes are scored according to the hardware performance of the node, the response time to the transaction and the results, etc., and the nodes are allocated to the corresponding shards according to the scoring results. The sharding strategy based on this model balances the performance differences among shards and improves the throughput of the entire system.
This paper proposes a blockchain-based digital-currency model based on certificates, and uses stablecoins as an example to illustrate the model. A stablecoin is anchored on a fiat currency, and often backed by 100% fiat such as USD. Unlike traditional cryptocurrency models, the blockchain system using this model does not store the assets directly, but instead store the certificates of these assets. People can trade digital certificates like trading digital currencies, but the value of these certificates in stored in custodian banks. The transaction process is slightly different from cryptocurrency transactions. The advantage of trading certificates is that the loss of these digital certificate does not mean the loss of value. If a private key is lost, a client can recover her assets from the custodian bank with a replacement certificate. Due to this nature, using this model, the center of digital economy remains at banks, rather than the blockchain network, as suggested by Digital-Currency Areas (DCA).
With the arrival of big data era and the rapid development of artificial intelligence, deep learning has made breakthroughs in many fields. However, although it has been widely used in many fields, there are still many challenges in itself, such as the slow reference speed of neural networks. In stream processing scenarios that integrate deep learning algorithms, data is often massive and generated with high-speed, requiring the system to respond in seconds or even milliseconds. If the response speed is too slow, the actual application requirements may not be met, and the user experience cannot be guaranteed. How to use stream processing technology to improve the speed and throughput of such systems has become an urgent problem to be solved. This paper used the popular real-time stream processing framework Flink to implement a complete data stream processing program, and integrated three algorithms of face detection, facial key points detection and face mosaic into the processing logic. Setting the operator parallelism realized parallel processing of video data, which improved the system throughput. The user can choose which algorithm to perform on the video, and can also choose the parallelism according to the performance of the machine. The system implemented the Flink framework to process video in parallel, achieved the effect of improving processing efficiency, and completely implemented the front-end interface and back-end program.
Traditional blockchain systems the consensus protocol and transaction management is performed together. In other words, the completion of the consensus protocol means the completion of a transaction. Furthermore, the transaction is considered as settled. This scheme is not compatible with modern financial transaction rules where transaction is distinct from settlement, and thorough regulation compliance should be performed to ensure that no money laundering. This paper proposes using five Merkle trees to maintain data in a blockchain system to separate consensus protocol from transaction management.
For companies using blockchain technology, it is critical to select the most suitable blockchain platform to develop enterprise applications. However, it is still a challenge for enterprises. As an important part of modern decision science, multi-criteria decision-making can well solve the problem of blockchain platform selection. Blockchain platforms integrate various blockchain technologies, and enterprises also need to consider multiple different criteria in the decision-making process. Therefore, this paper will use a heterogeneous multi-criteria decision-making method to solve the blockchain platform selection problem. First, the blockchain platform alternatives and evaluation criteria used for decision-making are identified. Second, blockchain platform alternatives are evaluated with appropriate fuzzy numbers based on defined evaluation criteria. Then, the original evaluations are consistency and normalized to obtain a normalized evaluation. Next, the improved information content formulas of the axiomatic design is proposed to obtain the information content of each normalized evaluation. Then, the weights of all evaluated criteria are obtained using the entropy weight method. Finally, the total weighted information content of each blockchain platform alternative is obtained. With validation, the decision-making model of blockchain platform proposed in this paper has a strong reference value.
In recent years, blockchain research has set off an upsurge in academia, and it is called the next generation of value Internet. Because of its decentralization, anonymity, security, immutability, traceability and other characteristics, blockchain is gradually accepted and developed by people. With the deepening of research and the integration of technologies such as deep learning, blockchain has gradually been applied to various fields such as credit reporting, government, medical care, and industrial Internet of Things, not just the initial virtual currency field. This article mainly discusses the three important stages of blockchain public chain development, namely Bitcoin, Ethereum, and meta-verse, and introduces some basic supporting technologies of blockchain, as well as the research status and future trends of blockchain. Simple Analysis. By vertically introducing the development history of the blockchain, researchers can have a more concrete understanding of the status quo of the blockchain, and provide ideas for blockchain-related research.
This paper introduces an early content propagation popularity prediction model based on graph neural network and variational inference topic dependent dynamic variational autoencoder model (CD-VAE). CD-VAE captures the dynamics in the content propagation process, aggregates the topological information in the information diffusion process using GraphSAGE, approaches the uncertainty in terms of time and node from the perspective of probability by introducing two variational autoencoders, considers the changes in semantic characteristics in the process by integrating natural language processing methods into the model, and therefore significantly improves its prediction performance.
At present, the research on human action recognition has achieved remarkable results and is widely used in various industries. Among them, human action recognition based on deep learning has developed rapidly. With sufficient labeled data, supervised learning methods can achieve satisfactory recognition performance. However, the diversification of motion types and the complexity of the video background make the annotation of human motion videos a lot of labor costs. This severely restricts the application of supervised human action recognition methods in practical scenarios. Since the zero-shot learning method can realize the recognition of unseen action categories without relying on a large amount of labeled data. In recent years, action recognition methods based on zero-shot learning have received great attention from researchers. In this paper, we propose an attention-based zero-shot action recognition model ADZSAR. We design a novel attention-based mechanism feature extraction method that introduces the current state-of-the-art semantic embedding model (Word2Vec). Experiments show that this method performs the best among similar zero-shot action recognition methods based on spatio-temporal features.
We propose a novel smart contract re-entry vulnerability detection model based on BiGAS. The model combines a BiGRU neural network that introduces an attention mechanism with an SVM. We start from the data features of smart contracts, learn the model layer by layer to achieve feature extraction and vulnerability identification, introduce batch normalization, Dropout processing and use improved model classifiers to improve the vulnerability identification accuracy, model convergence speed and generalization capability of smart contracts. We had conducted numerous experiments, and the experimental results showed that BiGAS Detection Model has a strong vulnerability detection ability. The accuracy of vulnerability detection reached 93.24
Smart contracts are one of the most successfully applied technologies on the blockchain, which are decentralized and immutable. Smart contracts cannot be modified once deployed. Therefore, security detection of smart contracts before deployment is essential. Some smart contracts may have tx.origin dependency vulnerabilities. In this paper, we propose a critical path vulnerability detection method for detecting tx.origin dependency vulnerabilities in smart contracts. Then, in order to solve the problem that the traditional search algorithm cannot determine the critical path, we propose a path determination method based on path priority. Our method determines the critical path in the control flow graph, which enables us to detect the vulnerabilities existing in smart contracts more quickly. The experimental results show that our method is more efficient than the existing technology and the false positive rate is lower.
With the rapid development of the software industry, the number of requirements included in the software is also increasing . In particular, in a complex software system such as an avionics software system, dozens of subsystems are usually included, and each subsystem includes hundreds of requirements, so the entire software system will include thousands of requirements. Furthermore, during software development, requirements will undergo frequent changes. When the number of requirements is huge and there are many people from different backgrounds participating in requirements development, it is easy to duplicate requirements. Therefore, in order to prevent redundant development, in the requirements analysis stage, requirements analysts usually need to detect all duplicate requirements in the requirements document and eliminate them, thereby improving the quality of software requirements and saving software development costs. However, when the number of requests is large, manually detecting duplicate requests can be a time-consuming and error-prone task. Aiming at this problem, this paper studies the method of duplicate requirements detection in English requirements.
Unlike highly purposeful search, a recommender system tends to uncover the user’s potential interests and is a personalized information filtering system. Recently, the performance of hypergraph neural networks in classification tasks has attracted much attention. Compared with traditional recommender systems, hypergraph neural network-based recommender systems have better mining higher-order associations, accurate modeling of multivariate relationships, handling of multimodal and heterogeneous data, and clustering advantages. This fact drives the development of recommendation algorithms based on hypergraph neural networks. To this end, we 1) define generic links of recommender systems, and systematically analyze the challenges of hypergraph neural network-based recommender systems in different research directions. 2) present some new perspectives on existing weaknesses and future developments.
