AUTOMATIC CONTROL AND COMPUTER SCIENCES最新文献

In view of improved performance of new spheres and directions of social development, Russian government pays attention to robotization based on modern Russian hardware elements for the purpose of import substitution. In this regard, association of the notions of collective of algorithms, collective of automata, collective of robots, and artificial intelligence is one of the topical problems. Of special importance are the capabilities of cybernetic investigation of multicore neural network automata for the purpose of constructing more complex automata, robots, and the behavior of a collective of robots on their basis. The current paper is aimed at demonstrating the capabilities of the set-theoretic cybernetic approach to artificial and complex natural objects and systems on their basis and creating a conceptual model of selection and joint simultaneous design of hardware and software tools of neural network automata on the basis of unified study of parallelization processes in a collective of automata in the form of explicit and implicit clusterization. The options of structures of collective of algorithms for providing cybersecurity and protection against threats in the form of hierarchy of security provision practices are analyzed, shown, and proposed. The paper proposes a method for analyzing and selecting the best architecture of a multicore neural network collective of automata and a collective of robots based on automata implemented on chip. An expert system based on VLSI 1879VM8Ya (NM6408) with a developed user interface is implemented.

The paper is dedicated to testing the performance indicators of automatic machine learning platforms when they function in standard and confidential modes using the example of a nonlinear multidimensional regression. A general protocol of distributed machine learning trusted in the sense of security is proposed. It is shown that within the framework of confidential virtualization, when optimizing the architecture of machine learning pipelines and hyperparameters, the best quality indicators of generated pipelines for multidimensional regressors and speed characteristics are demonstrated by solutions based on Auto Sklearn compared with Azure AutoML, which is explained by different learning strategies. The results of the experiments are presented.

This article discusses the ways of using distributed ledger technology to ensure the security of smart city information systems. The authors outline the limitations of the existing solutions in this area and the main directions of development of distributed ledger technology, determining its successful integration into the smart city ecosystem.

The developed neural network attack detection algorithm, whose peculiarity lies in the possibility of launching two parallel processes, is described: searching for the optimal model of an artificial neural network and normalization of the training sample data. It is shown that the artificial neural network architecture is selected taking into account the loss function for a limited set of attack classes. The application of TensorFlow and Keras Tuner libraries (frameworks) for the software implementation of an attack detection algorithm is shown. An experiment on the selection of neural network architecture and its training is described. The accuracy obtained in experiments is 94–98% for different classes of attacks.

This article proposes a model of embedding steganographic data into a message stream using noise-immune coding. The embedded data is disguised as noise in the message transmission channel. Extraction of the embedded message is based on the procedure of detecting corrupted bits using error-correcting codes. The presented model is related to key schemes. The stability of the scheme is based on the impossibility for the analyst to obtain complete information about the embedding. The model is universal and can be implemented in any network protocol using noise-immune codes.

This article proposes a new approach to the application of the well-known indicator function method, which is used to simulate the detection and neutralization of suspicious information objects in the information environment of an industrial enterprise, as well as to simulate a security system during operation. Formula dependencies are given for calculating stochastic indicators, allowing us to obtain some objective estimates of the values of time parameters and their influence on the safe functioning of information systems.

The security of the main distributed consensus algorithms used in distributed legers in the digital environment of a smart city is analyzed. A new concept for protecting a smart city’s distributed legers is presented, which consists of using a hybrid distributed consensus protocol based on the joint use of a tangle-class algorithm and a proof-of-authority-class algorithm, protected by trusted computing and remote attestation technologies. The proposed protocol compensates for the weaknesses and vulnerabilities of the existing distributed consensus algorithms that are inherent in distributed ledger technology and hinder its widespread use in large digital environments of smart cities.

A comparison of the existing software analysis methods is presented with an assessment of the possibility of identifying defects contained in several components implemented in different programming languages. A set of three universal graph representations of code is proposed for implementing security analysis methods that take into account interactions between software components using different sets of programming languages. For the proposed universal code representations, a concept of a security analysis system and a prototype of the system implementation are developed, supporting cross-language analysis of the software components implemented in the programming languages PHP and C, as well languages using the .NET platform.

It can be argued that in certain areas of information technology, classical systems of authentication of the user of a computer system based on passwords and tokens are being completely replaced by biometric technologies. However, biometric systems are vulnerable to various types of security threats. For example, unlike the same passwords and tokens, biometric-based templates cannot be replaced in the event of a compromise. To solve this problem, new protection schemes have been developed. They can be conditionally divided into two groups: biometric cryptography and cancelable biometrics. Biometric cryptography methods show average values of errors of the first and second kind and experimental work in this area is widely known. Cancellable biometrics can be highly reliable, but there are limited experimental data on them. A comparative analysis of the reliability of existing methods is presented. It has been shown that among static biometric parameters, the iris is of greatest interest, and among dynamic ones, keyboard handwriting is of greatest interest. However, the use of these methods, like others, has its own difficulties and risks.

This paper presents adversarial attacks on machine learning (ML) algorithms in intrusion detection systems (IDSs). Some examples of existing IDSs are examined. The existing approaches for detecting these attacks are considered. Requirements are developed to increase the stability of ML algorithms. Two approaches to detect adversarial attacks on ML algorithms are proposed, the first of which is based on a multiclass classifier and a Honeypot system, and the second approach uses a combination of a multiclass and binary classifier. The proposed approaches can be used in further research aimed at detecting adversarial attacks on ML algorithms.