Mahdi Kherad

Purpose: This study aims to address the challenges of analyzing high-dimensional big data by enhancing the Extreme Learning Machine (ELM) with a scalable and efficient feature extraction mechanism. Specifically, we integrate a Scalable Kernel Principal Component Analysis (S-KPCA) into the hidden layer of ELM to improve generalization and computational efficiency. Design/methodology/approach: We propose a hybrid framework, termed Fast and Scalable Kernel Principal Component Analysis-Hidden-nodes-based Extreme Learning Machine (FS-KPCA-H-ELM), which embeds KPCA-derived hidden nodes computed via a divide-and-conquer approximation strategy. This approach drastically reduces time complexity from O(n3) to approximately O(n log n), enabling applicability to large-scale datasets. Findings: Experiments on 10 big data benchmarks demonstrate that FS-KPCA-H-ELM achieves competitive or superior classification accuracy and substantially reduces training time compared to several existing ELM-based methods. Statistical tests confirm significant improvements over baseline ELM and MapReduce variants, while the method maintains the best average rank among all evaluated approaches. Originality/value: The study presents a novel synergy between KPCA and ELM by replacing random weights with KPCA-derived features, ensuring more informative hidden representations. The scalable approximation ensures feasibility for massive datasets, making the method suitable for real-world big data analytics.

Multi-objective optimization problems (MOPs) pose significant challenges due to the presence of multiple conflicting objectives. This paper introduces MOPOA, a novel Multi-Objective Pufferfish Optimization Algorithm inspired by the defensive behaviors of pufferfish in nature. MOPOA extends the original single-objective POA by incorporating Pareto dominance, an external archive for preserving non-dominated solutions, and a crowding distance mechanism to maintain solution diversity. The algorithm balances exploration and exploitation through biologically inspired phases simulating predator-prey interactions. To evaluate MOPOA’s performance, it was benchmarked against several state-of-the-art algorithms, including NSGA-III, MOPSO, MODA, and MOFDO, on two well-known test suites: the ZDT and CEC-2019 multi-objective functions. Results indicate that MOPOA not only achieves superior convergence to the Pareto front but also maintains high diversity and robustness across diverse optimization scenarios. These findings position MOPOA as a powerful and adaptive tool for solving complex real-world multi-objective problems.

Purpose: The purpose of this study is to develop a hybrid approach for document image quality assessment (DIQA) that effectively integrates traditional handcrafted features with deep learning-based features to improve accuracy, robustness and generalization across various document types and degradations. Design/methodology/approach: The proposed method extracts handcrafted features such as Local Binary Patterns, Gray-Level Co-occurrence Matrix, Gabor filters, Scale-Invariant Feature Transform and Haar-like features, as well as deep features using pre-trained convolutional neural networks (EfficientNet, Visual Geometry Group Network (VGGNet), InceptionV3 and ResNet). These features are fused into a composite vector, followed by dimensionality reduction using Principal Component Analysis. A Random Forest regression model is then trained to predict image quality scores. The method is evaluated on the Smartdoc-QA dataset. Findings: Experimental results demonstrate that the proposed method achieves higher correlation with ground-truth scores than several state-of-the-art methods. The integration of traditional and deep features enhances robustness against various distortions such as blur, noise, skew and lighting variations, and results in improved Pearson Linear Correlation Coefficient and Spearman Rank Order Correlation Coefficient metrics. Originality/value: This paper presents a novel hybrid DIQA framework that leverages both low-level structural/textural cues and high-level semantic representations. It is among the first to systematically combine shallow and deep features with dimensionality reduction and regression to form an end-to-end DIQA system. The approach is scalable, accurate and applicable to real-world document processing scenarios.

Purpose: Wireless sensor networks (WSNs) consist of multiple sensor nodes that are often randomly distributed in the environment and operate independently without human intervention. However, the limited power supply of these nodes presents a significant challenge for WSNs, making the development of energy-saving algorithms critical to increase network lifespan. Node clustering is a commonly used solution to reduce data transfer nodes and achieve data aggregation. Furthermore, the mobility of nodes in the network is often overlooked but can impact energy consumption. Design/methodology/approach: This study proposes a Dynamic Dual Head Clustering algorithm for WSNs (DDHCWSN) to reduce energy consumption. Instead of re-clustering in each round, the DDHCWSN updates existing clusters based on energy and position, allowing normal nodes to send data to the closest cluster head (CH) node. The algorithm considers two CHs, one for data aggregation and the other for sending data to the base station (BS), which can have a better effect on network power consumption. Findings: To evaluate the proposed method and compare it with state-of-the-art methods in the field of WSN clustering, three different criteria (scalability, network energy consumption and network lifetime) in six different scenarios are considered. According to the obtained results, it can be said that the proposed algorithm shows acceptable behavior in networks with low density, unlike other algorithms. In large networks, DDHCWSN has a longer lifetime than others. Although in small networks, CRDP and DMH-Leach algorithm performed better, but DDHCWSN also performed well. Originality/value: The main topics of the proposed algorithm are: (1) Dividing the network into different areas and then generating clusters based on them. (2) Using two separate CHs, one to communicate with members of cluster (data aggregation) and the other to communicate with other CHs and BSs. (3) Updating clusters instead of re-clustering nodes. (4) Select single or multi-hop mode.

In cyber-physical systems governed by nonlinear partial differential equations (PDEs), real-time control is often limited by sparse sensor data and high-dimensional system dynamics. Deep reinforcement learning (DRL) has shown promise for controlling such systems, but training DRL agents directly on full-order simulations is computationally intensive. This paper presents a sensor-driven, non-intrusive reduced-order modeling (NIROM) framework called FAE-CAE-LSTM, which combines convolutional and fully connected autoencoders with a long short-term memory (LSTM) network. The model compresses high-dimensional states into a latent space and captures their temporal evolution. A DRL agent is trained entirely in this reduced space, interacting with the surrogate built from sensor-like spatiotemporal measurements, such as pressure and velocity fields. A CNN-MLP reward estimator provides data-driven feedback without requiring access to governing equations. The method is tested on benchmark systems including Burgers’ equation, the Kuramoto–Sivashinsky equation, and flow past a circular cylinder; accuracy is further validated on flow past a square cylinder. Experimental results show that the proposed approach achieves accurate reconstruction, robust control, and significant computational speedup over traditional simulation-based training. These findings confirm the effectiveness of the FAE-CAE-LSTM surrogate in enabling real-time, sensor-informed, scalable DRL-based control of nonlinear dynamical systems.

In recent years, social networks have become an integral part of people's lives and play a significant role in the real world. The primary aim of influence maximization problem is finding a set of nodes in the network which can maximize the influence if the diffusion process starts from them. Therefore, the ...

Due to saving time and manpower, automatic and semi‐automatic methods can be used to identify and analyse ancient artefacts. Such methods are usually among the studies of neural networks and machine learning systems, which are carried out using remote sensing data and are completely based on spatial information. In the present research, the aim is to detect archaeological phenomena in the landscape of the historical city of Zuzan using convolutional neural network and object detection using the YOLO v8 algorithm, which uses aerial images from the 1960s and 1990s as input data. The most important steps of this method are: training and learning model, image pre‐processing, feature extraction and feature labelling are implemented to provide an automatic pattern recognition system for recognizing archaeological phenomena in an urban landscape. The training data set consists of old aerial images in which features such as the city wall (fence), Citadel and Aqueduct (Qanat) are labelled. The results of CNN training with aerial images of the 60s and 90s and Yolo modelling show the detection of feature such as the aqueduct with 69% accuracy, the city wall with 91% accuracy and the citadel with 100% accuracy.