Software-defined networks (SDN) have demonstrated considerable benefits in various practical domains by decoupling the control plane from the data plane, thus facilitating programmable network management. This paper presents a two-stage approach for solving the problem of controller placement called DEA-GAO. In the first stage, this strategy assumes the SDN network as a graph and using Data Envelopment Analysis (DEA) and relying on graph centrality metrics such as closeness centrality, betweenness centrality, and eigenvector centrality, calculates the efficiency of nodes to determine the optimal locations for deploying controllers. In the second stage, to allocate switches to controllers, the proposed strategy employs the Green Anaconda Optimization algorithm (GAO) to achieve an optimal allocation while considering network parameters such as average delay, load balancing, and reliability. Finally, to assess the efficacy of the proposed methodology, it is juxtaposed with three extant methods utilizing diverse datasets from the Internet Topology Zoo. The experimental findings indicate that the proposed approach significantly surpasses the existing methods, specifically the hybrid RDMCP-PSO algorithm, heuristic CPP algorithm and PSO algorithm in terms of both average delay (8.8%, 28.8% and 22.2% respectively) and controller utilization (1.5%, 7.3% and 32% respectively).

Software-Defined Networking (SDN) architectures, while inherently centralized, are prone to scalability issues, security breaches, and performance bottlenecks. The use of several controllers and the optimal distribution of traffic load across them is an effective countermeasure. This manuscript presents the AP-DQN framework, a methodology for identifying the optimal placement of controllers in SDN environments by combining adaptive clustering with deep reinforcement learning. Initially, the approach uses an improved version of the Affinity Propagation algorithm (AP), which consists of SDN peculiar criteria, i.e., end-to-end latency, traffic load, line running costs, and normalized security level, to cluster network switches, and thus identify possible zones for controller installation. Subsequently, a Deep Q-network (DQN) agent formalises the controller placement problem as a Markov decision process (MDP) and incrementally learns an optimal deployment policy that simultaneously optimises relevant performance metrics. The technique was tested against the Internet Topology Zoo dataset and benchmarked against known techniques, such as MODCEP, Multi-GA, and Random-CP. Simulation results show that AP-DQN can obtain 24 % improvement in load balancing, 25 % reduction in latency, 28 % reduction in link operational expenses, and 12 % improvement in normalized security level compared to comparative techniques.

Deep learning now underpins modern computer vision across various applications, including classification, detection, segmentation, retrieval, and generation. This survey first revisits foundational architectures—CNNs, early RNN-based vision pipelines, GANs, and the first Vision Transformers—then traces the shift to today’s practice (2023–2025): vision-language models (e.g., CLIP), large-scale self-supervised pretraining (e.g., DINOv2), promptable foundation segmentation (SAM), and diffusion-based generators (DDPM, latent diffusion). We distill the core design principles and training patterns behind these systems, analyze their strengths and limitations relative to classical approaches, and identify deployment-relevant concerns: compute efficiency, data governance, safety, and evaluation rigor. Looking ahead, we highlight parameter-efficient adaptation of vision foundation models (prompting/adapters), reliable few-shot protocols, and hardware-aware neural architecture search that meets latency and energy budgets. These priorities define concrete knowledge gaps that delineate the current research frontier. In this 2025 update, we introduce a concise survey of quantum–classical hybrids for vision, covering QCNNs, QViTs, and QGANs, and highlight their promise in small-data or tight-parameter regimes.

Software-defined networks (SDN), owing to their centralized control architecture, provide high flexibility in network management, configuration, and monitoring; however, this architecture also introduces critical challenges related to scalability, performance bottlenecks, and quality of service (QoS) degradation under heavy and dynamic traffic conditions, particularly in large-scale and beyond 5G (B5G) networks with stringent real-time latency requirements. In such environments, the controller placement problem (CPP) becomes an inherently NP-hard multi-objective optimization task, where conventional sequential and heuristic methods struggle to explore the massive solution space within practical time constraints, thereby motivating the need for computationally scalable frameworks that can exploit parallel processing and high-performance computing (HPC) capabilities. To address these challenges, this paper proposes DeepWK-MSTC, an advanced multi-objective controller placement framework that integrates weighted Kmeans-based clustering with a deep learning-driven optimization mechanism. The proposed method leverages the inherent parallelism of Deep Monte Carlo Tree Search (Deep-MCTS) to enable concurrent rollouts and accelerated decision-making, while jointly optimizing three key objectives: minimizing average delay ratio (ADR), improving energy efficiency (EE), and balancing controller load under dynamic traffic patterns. By incorporating network topology characteristics and real-time traffic dynamics, DeepWK-MSTC efficiently avoids local optima and ensures stable optimization behavior. The effectiveness of the proposed framework is evaluated on six real-world network topologies from the Internet Topology Zoo, namely Aarnet, Chinanet, Deutsche Telekom, Colt, Cogent, and Tata, and compared against state-of-the-art baselines including ALO and ELA-RCP. Experimental results demonstrate that DeepWK-MSTC achieves an average reduction of 50.2% in ADR, an average energy saving of 26.45%, and a 24% decrease in maximum controller load, with an additional 11.5% relative ADR reduction compared specifically to ELA-RCP. Overall, by explicitly exploiting parallel optimization and HPC-oriented design principles, DeepWK-MSTC enhances resource utilization and ensures scalable, stable, and real-time-capable controller placement for large-scale SDN environments.

In this study, a novel hybrid algorithm named the Reinforced Zebra Optimization Algorithm (RZOA) is proposed to solve the Traveling Salesman Problem (TSP). Initially, a discrete version of the Zebra Optimization Algorithm, referred to as DZOA, was developed, in which the continuous relationships of the original ZOA were transformed into a set of discrete operators to effectively update the routes. Subsequently, to enhance the intelligence of the operator selection process and to maintain a dynamic balance between exploration and exploitation, a mechanism based on Deep Reinforcement Learning was designed. Within the framework of the proposed RZOA, each zebra agent adaptively and self-learnedly makes decisions using a Deep Q-Network (DQN). Each agent observes both its own state and the overall population state, employs the DQN to select the optimal operator, and improves its decision-making policy dynamically through an experience replay memory. The state-action-reward structure was carefully designed to ensure an intelligent and balanced learning behavior between local and global search processes. The results of numerical experiments conducted on 42 standard benchmark datasets from the TSPLIB repository demonstrate that the proposed RZOA exhibits significant superiority over the compared algorithms in terms of solution quality, convergence speed, and performance stability. In particular, the proposed method achieves near-optimal performance with an average Percentage Deviation of the Best solution (PDB) below 1% and an average Percentage Deviation of the Average solution (PDA) typically below 0.5% for small and medium-scale instances, while maintaining PDA values within 4-5% for large-scale problems. Moreover, RZOA outperforms competing algorithms in the majority of benchmark cases, demonstrating superior robustness and consistency. Furthermore, the Friedman and Wilcoxon statistical analyses confirm this superiority at a 95% confidence level. Overall, by integrating the global search capability of the Zebra Optimization Algorithm with the adaptive decision-making power of the DQN, the proposed RZOA provides a novel, intelligent, and efficient approach for solving complex combinatorial optimization problems.

Text classification (TC) is one of core element of natural language processing (NLP), becoming more important as volume of text data increases. It is widely used in applications such as sentiment analysis, and topic categorization. Large language models (LLMs) have recently shown strong performance in these tasks, even without task-specific training. In this study, we evaluate Llama 3 8B model for text classification using zero-shot, and few-shot methods across many datasets. This study evaluates Llama 3 8B results on different NLP downstream tasks. Our findings show that Llama 3 8B can perform competitively across different settings, especially when guided by well-designed prompts. This study highlights how decoder LLMs can be effectively applied to classification tasks and provides acceptable performance.

The Internet of Things (IoT) has created new security challenges by connecting billions of smart devices to each other. One of these challenges is detecting attacks in IoT networks. Traditional attack detection methods are usually not suitable for large and complex networks such as IoT networks. In this research, a new model for detecting intrusion in IoT networks using Software-Defined Networking (SDN) is introduced. The main goal of the current research was to improve the stability of IoT networks against various attacks using an optimized machine learning model in a distributed manner. The presented approach uses the advantages of SDN, such as flexibility and centralized control, to improve intrusion detection performance. The proposed method includes two phases: first, the topology of the network is divided into a set of subdomains, and a controller node is assigned to each subdomain. Then, in the second phase, an ensemble classification model based on a random forest is utilized for detecting intrusion in each subdomain. This learning model is a forest of classification and regression trees (CARTs), each component of which is optimized by genetic algorithm (GA). Controller nodes can use this classification model to identify intrusion independently or cooperatively. The main novelty of the current work lies in optimizing multiple learning models and cooperatively utilizing them for intrusion detection goals. In an experimental environment based on MATLAB software, the effectiveness of this model for detecting intrusions on two databases, NSW-NB15 and NSLKDD, was evaluated. The findings of the experiments showed that this model can identify the attacks in these two databases with 98.06 % and 99.67 % accuracy respectively, which is significantly higher than the compared models.

A wide range of network technologies and equipment used in network infrastructure are vulnerable to Denial of Service (DoS) attacks. Therefore, the identification of these attacks is of particular importance in security systems. Most of the previously presented solutions use a single machine learning model to detect DoS attacks; but it seems that improving the detection accuracy and reliability in the intrusion detection system will be possible by using the combination of learning models. This research, is an effort to improve the accuracy of DoS attacks detection, compared to previous methods. Also, overcoming the challenge of large number of classes in intrusion detection task using ECOC based hybrid classifiers is one of the main objectives of the research. In this paper, a novel method to detect DoS attacks in computer networks is proposed. The proposed method performs the intrusion detection process in three phases named as pre-processing, feature extraction and classification. Principal Component Analysis (PCA) is used for extracting features, while a combination of Error Correcting Output Codes (ECOC) and Adaptive Neuro-Fuzzy Inference System (ANFIS) is used for classification. In this classification model, Particle Swarm Optimization (PSO) algorithm has been used to optimize the structure of ANFIS. The performance of the proposed method has been evaluated using the NSLKDD database. Using a 10-fold cross validation experiment, the proposed IDS showed a sensitivity of 99.82%. The results also show that the proposed method can detect the types of DoS attacks with an average accuracy of 98.9%, which shows a significant improvement compared to the previous methods.

The main factors that affect the performance of residue number system (RNS) are speed and hardware complexity of the residue-to-binary converter. In this paper, two efficient reverse converters are proposed for the moduli sets [Formula: see text] and [Formula: see text] based on two-level approach. In the first level, a reverse converter is implemented based on new Chinese remainder theorem-I for the moduli set [Formula: see text]. The fourth modulus is joined in the second level for each converter and only simple binary adders are used in this level. The architecture of novel circuits mainly consists of simple adders thus leading to implementing efficient converters. To have a fair comparison, both unit gate model and simulation are used. The proposed converters and the recently represented converters have been implemented on Xilinx ISE 13.1 field-programmable gate array (FPGA) simulator to derive area and delay that are measured for the various ranges up to 256 bits. The experimental results show that the proposed converters have lower area–time complexity comparing to the state-of-the-art converter for similar moduli sets.

The implementation of reverse converter is one of the most important elements influencing the performance of a residue number system. In this research, we propose an efficient high-speed and low-area reverse converter for moduli set { 2 4 n , 2 2 n + 1 , 2 n + 1 , 2 n - 1 } . Due to the form of the moduli set, we have implemented the proposed reverse converter using Chinese remainder theorem and dynamic range division technique, which have resulted in a simpler design, lower area consumption and higher conversion speed compared to previous studies. To have a fair comparison between the proposed reverse converter and the recently presented converters with similar moduli sets, we used Xilinx ISE 13.1 FPGA simulator to derive area and delay, which are measured to cover the various dynamic ranges up to 256 bit. The experimental results show that the proposed reverse converter achieves an average 20% area time saving compared to the recently presented converters for similar moduli set.

The implementation of reverse converter is one of the most important elements influencing the performance of a residue number system. In this research, we propose an efficient high-speed and low-area reverse converter for moduli set { 2^4n ,2^2n + 1,2^n + 1,2^n - 1} . Due to the form of the moduli set, we have implemented the proposed reverse converter using Chinese remainder theorem and dynamic range division technique, which have resulted in a simpler design, lower area consumption and higher conversion speed compared to previous studies. To have a fair comparison between the proposed reverse converter and the recently presented converters with similar moduli sets, we used Xilinx ISE 13.1 FPGA simulator to derive area and delay, which are measured to cover the various dynamic ranges up to 256 bit. The experimental results show that the proposed reverse converter achieves an average 20

The main requirement of smart cities is intelligent and effective parking system, which currently most parking systems exist basedon various technologies such as Internet of Things, so intelligent parking management is gradually becoming a necessity. Hence inthis article a new manner is introduced in the form of prioritization, which alters the type of prioritization than the previous methods, sothat the different parameters such as the distance between parking lots, distance between vehicle and a parking lot, number of freeparking spaces in the destination parking lot, number of unsuccessful parks, and the number of parking lot visits are considered, and ingeneral a system is recommended that helps drivers to find a parking lot with the minimum cost based on the aforesaid parameters. Thesimulation of proposed approach performed in Arena application and the results of the simulation and its comparison with theprevious works suggest that the proposed method results in the better decrease of the waiting time than the existing waiting timein the previous works and contributes to minimize drivers waiting time.

Residue Number System is a kind of numerical systems that uses the remainder of division in several different moduli. Conversion of a number to smaller ones and carrying out parallel calculations on these numbers will increase the speed of the arithmetic operations in this system. However, the main factor that affects performance of system is hardware complexity of reverse converter. Reverse converters convert the resulted remainders to the conventional number system. In this paper an area efficient reverse converter is proposed for moduli set {2^n+1,2^n-1,2^n} based on two-part RNS and mixed radix conversion algorithm. Selecting appropriate order of modulus and using well-known lemmas, leads to reduce the complexity of the proposed converter comparing to previous designs. To have an accurate comparison, both unit gate model and simulation in Xilinx 13.1 FPGA are used in this paper. The results of comparison indicate that the novel proposed reverse converter has improved the time complexity and area, while having almost same delay.

Residue number system is a non-weighted system in which some arithmetic operations are performed parallelly. A factor that affects the system’s performance is the complexity of converters, as the complexity of reverse converter should not discomfit the earned speed of parallelly performing arithmetic unit. In this paper, two efficient reverse converters are proposed for moduli set {2^2n+1-1,2^2n,2^n-1} by using two-level method and mixed-radix conversion algorithm. Both unit gate model and simulation are used to have an efficient comparison. The novel converters and the recently presented reverse converters with similar dynamic range were implemented in Hardware Description Language on Xilinx 13.1 FPGA simulator. Area and delay of each converter were measured for various dynamic range up to 256 bits. As the results indicated, the novel proposed area-efficient converter and delay-efficient converter indicate an improvement about 8 and 14

Residue number system is a non-weighted system in which some arithmetic operations are performed parallelly. A factor that affects the system’s performance is the complexity of converters, as the complexity of reverse converter should not discomfit the earned speed of parallelly performing arithmetic unit. In this paper, two efficient reverse converters are proposed for moduli set \(\{2^{2n+1}-1,2^{2n},2^{n}-1\}\) by using two-level method and mixed-radix conversion algorithm. Both unit gate model and simulation are used to have an efficient comparison. The novel converters and the recently presented reverse converters with similar dynamic range were implemented in Hardware Description Language on Xilinx 13.1 FPGA simulator. Area and delay of each converter were measured for various dynamic range up to 256 bits. As the results indicated, the novel proposed area-efficient converter and delay-efficient converter indicate an improvement about 8 and 14%, respectively, in terms of time complexity comparing to the recently presented design.