نشریه مهندسی برق و مهندسی کامپیوتر ایران
سال بیستم شماره 2 (پیاپی 71، تابستان 1401)

The complexity of real-world applications, especially in the field of the Internet of Things, has brought with it a variety of security risks. IoT Botnets are known as a type of complex security attacks that can be detected using machine learning tools. Detection of these attacks, on the one hand, requires the discovery of their behavior patterns using batch processing with high accuracy, and on the other hand, must be operated in real time and adaptive like stream processing. This highlights the importance of using batch/stream hybrid processing techniques for botnet detection. Among the important challenges of these processes, we can mention the selection of appropriate features to build basic models and also the intelligent selection of basic models to combine and present the final result. In this paper, we present a solution based on a combination of stream and batch learning methods with the aim of botnet anomaly detection. This approach uses a dynamic feature selection method that is based on a genetic algorithm and is fully compatible with the nature of hybrid processing. The experimental results in a data set consisting of two known types of botnets indicate that on the one hand, the proposed approach increases the speed of hybrid processing and reduces the detection time of the botnets by reducing the number of features and removing inappropriate features, and on the other hand, increases accuracy by selecting appropriate models for combination.

In the present research, a type of permutation optimization was introduced. It is assumed that the cost function has an unknown probability distribution function. Since the solution space is inherently large, solving the problem of finding the optimal permutation is complex and this assumption increases the complexity. In the present study, an algorithm based on distributed learning automata was presented to solve the problem by searching in the permutation answer space and sampling random values. In the present research, in addition to the mathematical analysis of the behavior of the proposed new algorithm, it was shown that by choosing the appropriate values of the parameters of the learning algorithm, this new method can find the optimal solution with a probability close to 100% and by targeting the search using the distributed learning algorithms. The result of adopting this policy is to decrease the number of samplings in the new method compared to methods based on standard sampling. In the following, the problem of finding the minimum spanning tree in the stochastic graph was evaluated as a random permutation optimization problem and the proposed solution based on learning automata was used to solve it.

Today, billions of devices are connected via the Internet of Things, often through insecure communications. Therefore, security and privacy issues of these devices are a major concern. Since devices in IoT are typically resource-constrained devices, the security solutions of this environment in terms of processing and memory must be secure and lightweight. However, many existing security solutions are not particularly suitable for IoT due to high computation. So there is a need for a lightweight authentication protocol for IoT devices. In this paper, a mutual lightweight authentication protocol between nodes with limited resources and IoT servers is introduced that uses node prioritization based on traffic rates. This scheme is light due to the use of lightweight XOR and Hash operations. The proposed is resistant to cyber-attacks such as eavesdropping attack, and replay attack. The proposed is also secure using the AVISPA tool in the Dolev-Yao threat model. The security risks of this scheme are low compared to other lightweight methods. In addition, the proposal is compared with existing authentication schemes reduces the computational cost, protects privacy through anonymity of nodes, and provides forward secrecy. In our method, the execute time of authentication is reduced by 15% compared to the other methods.

SIP is considered as a signaling protocol for IP multimedia subsystem (IMS) and IMS is introduced as the next generation networking platform. Unlike positive features such as text-based, IP-based, data-independent, support mobility and end-to-end, SIP lacks a proper overload control mechanism. Hence, this challenge will cause the widespread users of next generation networks to loss quality of service. IMS is a complex network consisting of subsystems, interacting with each other. As a result, multi-agent systems can be a useful tool to solve the IMS overload. Therefore, each IMS server is considered as an intelligent agent with learning and negotiation ability with other agents while maintaining autonomy therefore, the overload is eliminated by communication and knowledge transferred between agents. In this paper, multi-agent system and their properties presents a hop-by-hop elimination-based method which simulation results show performance improvement compared to known methods.

Polynomial neural network (PNN) is a supervised learning algorithm which is one of the most popular models used in real applications. The architectural complexity of polynomial neural network in terms of both number of partial descriptions (PDs) and number of layers, leads to more computation time and more storage space requirement. In general, it can be said that the architecture of the polynomial neural networks is very complex and it requires large memory and computation time. In this research, a novel approach has been proposed to improve the classification performance of a polynomial neural network using the Whale Optimization Algorithm (PNN-WOA). In this approach, the PDs are generated at the first layer based on the combination of two features. The second layer nodes consists of PDs generated in the first layer, input variables and bias. Finally, the polynomial neural network output is obtained by sum of weighted values of the second layer outputs. Using the Whale Optimization Algorithm (WOA), the best vector of weighting coefficients will be obtained in such a way that the PNN network reach to the highest classification accuracy. Eleven different dataset from UCI database has been used as input data of proposed PNN-WOA and the results has been presented. The proposed method outperforms state-of-the-art approaches such as PNN-RCGA, PNN-MOPPSO, RCPNN-PSO and S-TWSVM in most cases. For datasets, an improvement of accuracy between 0.18% and 10.33% can be seen. Also, the results of the Friedman test indicate the statistical superiority of the proposed PNN-WOA model compared to other methods with p value of 0.039.

Iranian "Dastgah" music classification by computer is a very interesting yet complex and challenging topic for those who are interested in Iranian Dastgah music. The aforementioned problem is important, firstly, due to its many applications in different areas such as composing and teaching music, and secondly, because of the needs of ordinary people to computer to detect the Dastgah. This paper presents a method for recognition of the genre (Dastgah) and subgenre (sub-Dastgah) of Iranian music based on sequential note extraction, hierarchical classification, and the use of LSTM networks. In the proposed method, the music track is first classified into one of the three general categories. The first category includes only "Mahour" Dastgah, the second category includes "Shour" and "Nava", and the third category includes "Homayoun", "Segah" and "Chahargah". Then, for each category, depending on its type, a different number of classifiers are applied until one of the 6 Dastgah and 11 sub-Dastgah of Iranian music are recognized. This research is not limited to any particular style of playing or instruments, it is also not affected by neither the speed nor the techniques of player. The labeled tracks in the "Arg" database, which is created for this research, are solo. However, some of them are also played by percussion instruments (such as the Tombak) along with melodic instruments. The results show that recognition of 6 main Dastgah and 11 sub-Dastgah have been approved by an average accuracy of 74.5% and 66.35%, respectively, which is more promising compared to other few similar studies.

Machine learning has been widely used over the past decades due to its wide range of applications. In most machine learning applications such as clustering and classification, data dimensions are large and the use of data reduction methods is essential. Non-negative matrix factorization reduces data dimensions by extracting latent features from large dimensional data. Non-negative matrix factorization only considers how to model each feature vector in the decomposed matrices and ignores the relationships between feature vectors. The relationships between feature vectors provide better factorization for machine learning applications. In this paper, a new method based on non-negative matrix factorization is proposed to reduce the dimensions of the data, which sets constraints on each feature vector pair using distance-based criteria. The proposed method uses the Frobenius norm as a cost function to create update rules. The results of experiments on the data sets show that the proposed multiplicative update rules converge rapidly and give better results than other algorithms.

In this paper, uplink secure transmission in a non-orthogonal multiple access (NOMA) network is investigated by selection of the users for data transmission to the base station (BS) and also jammers with the capability of energy harvesting. In fact, each frame has two phases. In the first phase, jammers harvest energy from BS and in the second phase, the selected users transmit their data to BS using NOMA technique while selected jammer emits the artificial noise for confusing the eavesdropper. In fact, the problem is maximizing the secrecy throughput by selection of the users for uplink data transmission to BS in each frequency channel and suitable jammers to make the artificial noise for eavesdropper with constraints on the secrecy outage probability (SOP) and connection outage probability (COP). The problem is solved based on the convex optimization methods and Karush-Kuhn-Tucker (KKT) conditions. An algorithm is proposed for solving the problem and the system performance is evaluated. Simulation results present that the proposed algorithm has the better performance for the throughput and security of the network in comparison with the benchmark algorithms in different situations and scenarios.