Artificial Intelligence
Amir Mahdi Sedghi; Shahla Nemati
Abstract
Background and Objectives: Brain tumor detection in MRI images is critical for early diagnosis and effective treatment, yet manual interpretation is time-consuming and prone to variability. Deep learning models such as YOLO have advanced real-time object detection, but their speed–accuracy tradeoff ...
Read More
Background and Objectives: Brain tumor detection in MRI images is critical for early diagnosis and effective treatment, yet manual interpretation is time-consuming and prone to variability. Deep learning models such as YOLO have advanced real-time object detection, but their speed–accuracy tradeoff remains a challenge for medical tasks involving small or low-contrast lesions. The potential of transformer-based detectors like RT-DETR to simultaneously improve accuracy and maintain real-time speed in clinical settings is not well established.Methods: This study performed a controlled head-to-head comparison between the proposed model (RT-DETR-L-based model) and the YOLOv8s models using a curated, single-class brain tumor MRI dataset of 300 images. Both models were trained and evaluated under identical conditions with comprehensive data augmentation strategies, and their performance was assessed using standard object detection metrics including precision, recall, specificity, and mean Average Precision (mAP) across multiple IoU thresholds.Results: The proposed model achieved higher localization fidelity and overall accuracy compared to YOLOv8s, with mAP@0.5:0.95 of 0.493 versus 0.421 and mAP@0.5 of 0.963 versus 0.941. Precision and specificity for the proposed model reached 1.000, eliminating false positives, while recall was slightly lower than YOLOv8s (0.925 vs. 0.932), indicating a marginal increase in missed detections. Qualitative analysis confirmed robust detection across various tumor sizes and intensities, though some small or low-contrast lesions were missed.Conclusion: Proposed model surpasses YOLOv8s in accuracy and specificity for real-time brain tumor detection in MRI images, offering a promising balance between speed and precision. However, its slightly lower recall underscores the need for further refinement to minimize false negatives. The findings suggest transformer-based detectors can narrow the speed–accuracy gap in medical imaging, but broader validation and optimization for resource-constrained environments are required for clinical deployment. Future work should focus on enhancing sensitivity and generalizability through advanced augmentation, larger datasets, and ensemble approaches.
Artificial Intelligence
Ali Bazghandi
Abstract
Background and Objectives: Random number generation is essential in simulation, cryptography, and statistical modeling. Classical PRNGs such as the Linear Congruential Generator and Mersenne Twister are efficient but exhibit predictability and correlation. Newer families like PCG and BRG improve statistical ...
Read More
Background and Objectives: Random number generation is essential in simulation, cryptography, and statistical modeling. Classical PRNGs such as the Linear Congruential Generator and Mersenne Twister are efficient but exhibit predictability and correlation. Newer families like PCG and BRG improve statistical balance yet remain static after initialization, while chaotic and neural methods face reproducibility and stability issues. To overcome these limits, we propose the Adaptive Multi-Layer Random Generator (AMLRG), designed to deliver self-regulating pseudorandomness through adaptive feedback and hybrid entropy sources.Methods: AMLRG combines three layers: (i) an index generator based on linear, logistic, or chaotic processes, (ii) a uniform distribution module, and (iii) an adaptive feedback system that tunes parameters in real time. Online diagnostics—Kolmogorov–Smirnov tests, autocorrelation analysis, and Shannon entropy—direct dynamic adjustment. Implemented in Python, the system produces binary streams and diagnostic plots. Evaluation involved ablation studies (removing feedback, switching, or stratification), comparison with LCG, PCG, BRG, and logistic-only baselines, and validation using Dieharder, TestU01, and NIST SP 800-22.Results: AMLRG produced lower KS distances, near-zero autocorrelation, and entropy close to theoretical maxima, outperforming all baselines. Ablation confirmed the contribution of each layer to statistical quality. Results show stable behavior across 200,000 values, with speed comparable to PCG but greater adaptability.Conclusion: AMLRG introduces dynamic correction that improves independence and uniformity in pseudorandom sequences. Its layered architecture suits engineering simulations, adaptive systems, and security-sensitive statistical preprocessing. Future work will target non-uniform distributions, GPU acceleration, and hardware implementation.
Artificial Intelligence
Mujtaba Sultani; Negin Daneshpour
Abstract
Background and Objectives: Discussion forums in Massive Open Online Courses (MOOCs) enable students to interact with instructors and share educational concerns. However, identifying urgent posts within the vast volume of discussions poses significant challenges. High dropout rates and the need for timely ...
Read More
Background and Objectives: Discussion forums in Massive Open Online Courses (MOOCs) enable students to interact with instructors and share educational concerns. However, identifying urgent posts within the vast volume of discussions poses significant challenges. High dropout rates and the need for timely responses to urgent queries highlight the importance of efficient classification systems. This study addresses the binary classification of student posts in the Stanford MOOC Posts dataset into urgent and non-urgent categories, and aims to improve performance in the presence of class imbalance.Methods: We propose a hybrid deep learning model that integrates BERT-based contextual embeddings with a Convolutional Neural Networks (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) architecture to capture both local textual features and long-term dependencies. To mitigate the class imbalance issue, BERT-based data augmentation technique was employed which enriches minority class samples, and enhance model generalization and urgent post detection. The model’s performance was compared against baseline methods, including CNN, LSTM, BiLSTM, and other state-of-the-art models. Evaluation metrics such as F1-weighted score and class-specific F1 scores were used to assess effectiveness.Results: The model achieved a 93.3% F1-weighted score and an 84.1% F1 score for the urgent class which surpasses the best-performing state-of-the-art model by 0.6% and 0.8%, respectively. The results show the effectiveness of augmentation and hybrid architecture while identifying urgent posts, particularly within imbalanced datasets.Conclusion: This research introduces a scalable and effective framework for urgent post detection in MOOCs. By leveraging BERT-based augmentation and a CNN–BiLSTM hybrid architecture that integrates contextual and sequential learning, the study provides automated forum analysis, offer timely insights for instructors and course designers to enhance students support, engagement, and retention.
Artificial Intelligence
Motahareh Akbari Poodineh; Fatemeh Zare Mehrjardi; Mohsen Sardari Zarchi
Abstract
Background and Objectives: Cardiovascular diseases, particularly cardiac arrhythmias, are among the leading causes of mortality worldwide. Early and accurate diagnosis is essential for improving patient outcomes. Although electrocardiogram (ECG) signals are widely used for arrhythmia detection, manual ...
Read More
Background and Objectives: Cardiovascular diseases, particularly cardiac arrhythmias, are among the leading causes of mortality worldwide. Early and accurate diagnosis is essential for improving patient outcomes. Although electrocardiogram (ECG) signals are widely used for arrhythmia detection, manual interpretation remains time consuming and error prone. Therefore, this study proposes an innovative, optimized two stage deep learning framework for the reliable diagnosis of cardiac arrhythmias from ECG signals, aiming to enhance both accuracy and robustness.Methods: The key innovation lies in the first stage, where the autoencoder’s reconstruction error threshold is optimized using a Genetic Algorithm (GA) to maximize the separation between normal and abnormal signals. Only signals identified as abnormal proceed to the second stage, a Convolutional Neural Network (CNN) that classifies them into four arrhythmia types (Supraventricular, Ventricular, Fusion, and Unknown beats). All experiments were conducted on the MIT BIH Arrhythmia Database using a stratified split, with SMOTE applied exclusively to the CNN training set to address class imbalance. Performance was evaluated through 5 fold cross validation.Results: The proposed AE GA CNN+SMOTE framework achieved an average accuracy of 97.89 ± 0.25%, precision of 97.90 ± 0.24%, recall of 97.68 ± 0.29%, and an F1 score of 97.69 ± 0.28%. It outperformed the single stage CNN+SMOTE baseline by +6.28% in accuracy (p < 0.001) and showed statistically significant improvements over all other two stage variants (p < 0.05).Conclusion: The two stage architecture, enhanced by GA driven threshold optimization and SMOTE balancing, demonstrates high accuracy and robustness for automated arrhythmia screening. The statistically validated performance gains support its potential as a decision support tool for clinical and real time ECG analysis.
Artificial Intelligence
Atefeh Mohammadi; Mohammad Reza Pajoohan; Ali Mohammad Zareh Bidoki
Abstract
Background and Objectives: In Natural Language Processing (NLP), sentiment analysis is crucial for understanding and extracting aspects and opinions expressed in textual data. Recent methods have emphasized determining polarity in multi-domain sentiment analysis while giving less attention to aspect ...
Read More
Background and Objectives: In Natural Language Processing (NLP), sentiment analysis is crucial for understanding and extracting aspects and opinions expressed in textual data. Recent methods have emphasized determining polarity in multi-domain sentiment analysis while giving less attention to aspect and opinion extraction. Furthermore, the terms that convey aspects and opinions may have different importance in different domains, and this difference should be considered to enhance the extraction of aspect-opinion pairs. Methods: To address these challenges, we propose a Weighted Words Multi-Domain (WWMD) model for aspect-opinion pairs extraction, consisting of a self-attention mechanism and a dense network. The self-attention mechanism extracts each word's importance according to the sentence's overall meaning. The dense network is used for domain prediction. It assigns greater weight to words relevant to each domain, which leads to considering the different significance of terms across various contexts. Adding an attention mechanism to the domain module allows for a clearer understanding of different aspects and opinions across various domains. We utilize a two-channel approach, one channel extracts aspects and opinions, while the other extracts the relationships between them. The weighted words extracted by our model are simultaneously considered as the input for both channels.Results: Using weighted words specific to each domain, improves the model output.Conclusion: Evaluation results on benchmark datasets demonstrate the superiority of the proposed model compared to state-of-the-art techniques.
Artificial Intelligence
Mahdi Shahbazi Khojasteh; Armin Salimi Badr
Abstract
Background and Objectives: Unmanned Aerial Vehicles (UAVs) face significant challenges in navigating narrow passages within GPS-denied environments due to sensor and computational limitations. While deep reinforcement learning (DRL) has improved navigation, many methods rely on costly sensors like depth ...
Read More
Background and Objectives: Unmanned Aerial Vehicles (UAVs) face significant challenges in navigating narrow passages within GPS-denied environments due to sensor and computational limitations. While deep reinforcement learning (DRL) has improved navigation, many methods rely on costly sensors like depth cameras or LiDAR. This study addresses these issues using a vision-based DRL framework with a monocular camera for autonomous UAV navigation.Methods: We propose a DRL-based navigation system utilizing Proximal Policy Optimization (PPO). The system processes a stack of grayscale monocular images to capture short-term temporal dependencies, approximating the partially observable environment. A custom reward function encourages trajectory optimization by assigning higher rewards for staying near the passage center while penalizing further distances. The navigation system is evaluated in a 3D simulation environment under a GPS-denied scenario.Results: The proposed method achieves a high success rate, surpassing 97% in challenging narrow passages. The system demonstrates superior learning efficiency and robust generalization to new configurations compared to baseline methods. Notably, using stacked frames mitigates computational overhead while maintaining policy effectiveness.Conclusion: Our vision-based DRL approach enables autonomous UAV navigation in GPS-denied environments with reduced sensor requirements, offering a cost-effective and efficient solution. The findings highlight the potential of monocular cameras paired with DRL for real-world UAV applications such as search and rescue and infrastructure inspection. Future work will extend the framework to obstacle avoidance and general trajectory planning in dynamic environments.
Artificial Intelligence
Oveis Dehghantanha; Nasser Mehrshad; Roksana Bakhshali; Ahmad Reza Sebzari
Abstract
Background and Objectives: Understanding the heterogeneity of breast cancer is crucial for improving treatment strategies. This study investigates the application of K-Means and Hierarchical Clustering to a local dataset of breast cancer patients from Iranmehr Hospital, Birjand, Iran, with the primary ...
Read More
Background and Objectives: Understanding the heterogeneity of breast cancer is crucial for improving treatment strategies. This study investigates the application of K-Means and Hierarchical Clustering to a local dataset of breast cancer patients from Iranmehr Hospital, Birjand, Iran, with the primary goal of identifying potential patient subgroups based on their clinical and treatment characteristics for knowledge discovery. The potential of these subgroups to inform future research on personalized treatment approaches is explored.Methods: A retrospective dataset comprising pathological and clinical information was analyzed using K-Means and Agglomerative Hierarchical Clustering to identify patient subgroups. The optimal number of clusters was consistently determined to be two (k=2) for both methods based on rigorous internal validation metrics (Elbow Method, Silhouette Analysis, Calinski-Harabasz Index, and Largest Jump Analysis for Hierarchical Clustering). Statistical tests (ANOVA and Chi-squared) were employed to assess significant differences in features across the identified clusters from both K-Means and Hierarchical analyses, providing insights into the key factors differentiating these groups. Internal cluster validity was assessed using Silhouette Score and Calinski-Harabasz Index.Results: The K-Means analysis identified two clusters exhibiting significant differences in characteristics such as age, chemotherapy session intensity, menopausal status, nodal involvement, and biomarker expression (ER, PR, HER2, Ki67). The Hierarchical Clustering also yielded two clusters with varying characteristics, and a comparison between the two methods highlighted both similarities and differences in the identified patient stratifications. The overall agreement between K-Means and Hierarchical Clustering was quantified by an Adjusted Rand Index (ARI) of 0.4697.Conclusion: Both K-Means and Hierarchical Clustering effectively revealed potential patient subgroups within the studied dataset, highlighting the heterogeneity of breast cancer presentation and treatment at a local level These clusters exhibited statistically significant differences across key clinical and treatment features. Future research is needed to validate these findings in larger, multi-center studies, explore the clinical significance of these subgroups in terms of treatment outcomes, and compare the effectiveness of different clustering methodologies for this purpose.
Artificial Intelligence
S. H. Zahiri; R. Iranpoor; N. Mehrshad
Abstract
Background and Objectives: Person re-identification is an important application in computer vision, enabling the recognition of individuals across non-overlapping camera views. However, the large number of pedestrians with varying appearances, poses, and environmental conditions makes this task particularly ...
Read More
Background and Objectives: Person re-identification is an important application in computer vision, enabling the recognition of individuals across non-overlapping camera views. However, the large number of pedestrians with varying appearances, poses, and environmental conditions makes this task particularly challenging. To address these challenges, various learning approaches have been employed. Achieving a balance between speed and accuracy is a key focus of this research. Recently introduced transformer-based models have made significant strides in machine vision, though they have limitations in terms of time and input data. This research aims to balance these models by reducing the input information, focusing attention solely on features extracted from a convolutional neural network model. Methods: This research integrates convolutional neural network (CNN) and Transformer architectures. A CNN extracts important features of a person in an image, and these features are then processed by the attention mechanism in a Transformer model. The primary objective of this work is to enhance computational speed and accuracy in Transformer architectures. Results: The results obtained demonstrate an improvement in the performance of the architectures under consistent conditions. In summary, for the Market-1501 dataset, the mAP metric increased from approximately 30% in the downsized Transformer model to around 74% after applying the desired modifications. Similarly, the Rank-1 metric improved from 48% to approximately 89%.Conclusion: Indeed, although it still has limitations compared to larger Transformer models, the downsized Transformer architecture has proven to be much more computationally efficient. Applying similar modifications to larger models could also yield positive effects. Balancing computational costs while improving detection accuracy remains a relative goal, dependent on specific domains and priorities. Choosing the appropriate method may emphasize one aspect over another.
Artificial Intelligence
S. Kabiri Rad; V. Afshin; S. H. Zahiri
Abstract
Background and Objectives: When dealing with high-volume and high-dimensional datasets, the distribution of samples becomes sparse, and issues such as feature redundancy or irrelevance arise. Dimensionality reduction techniques aim to incorporate correlation between features and map the original features ...
Read More
Background and Objectives: When dealing with high-volume and high-dimensional datasets, the distribution of samples becomes sparse, and issues such as feature redundancy or irrelevance arise. Dimensionality reduction techniques aim to incorporate correlation between features and map the original features into a lower dimensional space. This usually reduces the computational burden and increases performance. In this paper, we study the problem of predicting heart disease in a situation where the dataset is large and (or) the proportion of instances belonging to one class compared to others is significantly low.Methods: We investigated three of the prominent dimensionality reduction techniques, including Principal Component Analysis (PCA), Information Bottleneck (IB) and Variational Autoencoder (VAE) on popular classification algorithms. To have adequate samples in all classes to properly feed the classifier, an efficient data balancing technique is used to compensate for fewer positives than negatives. Among all data balancing methods, a SMOTE-based method is selected, which generates new samples at the boundary of the samples distribution and avoids the synthesis of noise and redundant data. Results: The experimental results show that VAE-based method outperforms other dimensionality reduction algorithms in the performance measures. The proposed hybrid method improves accuracy to 97.1% and sensitivity to 99.2%.Conclusion: Finally, it can be concluded that the combination of VAE with oversampling algorithms can significantly enhance system performance as well as computational time.
Artificial Intelligence
L. Hafezi; S. Zarifzadeh; M. R. Pajoohan
Abstract
Background and Objectives: Detecting multiple entities within financial texts and accurately analyzing the sentiment associated with each is a challenging yet critical task. Traditional models often struggle to capture the nuanced relationships between multiple entities, especially when sentiments are ...
Read More
Background and Objectives: Detecting multiple entities within financial texts and accurately analyzing the sentiment associated with each is a challenging yet critical task. Traditional models often struggle to capture the nuanced relationships between multiple entities, especially when sentiments are context-dependent and spread across different levels of a document. Addressing these complexities requires advanced models that can not only identify multiple entities but also distinguish their individual sentiments within a broader context. This study aims to introduce and evaluate two novel methods, ENT-HAN and SNT-HAN, built upon the Hierarchical Attention Networks, specifically designed to enhance the accuracy of both entity extraction and sentiment analysis in complex financial documents.Methods: In this study, we design ENT-HAN and SNT-HAN methods to address the tasks of multi-entity detection and sentiment analysis within financial texts. The first method focuses on entity extraction, where capture hierarchical relationships between words and sentences. By utilizing word-level attention, the model identifies the most relevant tokens for recognizing entities, while sentence-level attention helps refine the context in which these entities appear, allowing the model to detect multiple entities with precision. The second method is applied for sentiment analysis, aiming to classify sentiments into positive, negative, or neutral categories. The sentiment analysis model employs hierarchical attention to identify the most important words and sentences that convey sentiment about each entity. This approach ensures that the model not only focuses on the overall sentiment of the text but also accounts for context-specific variations in sentiment across different entities. Both methods were evaluated on FinEntity dataset, and the results demonstrate their effectiveness, with significantly improving the accuracy of both entity extraction and sentiment classification tasks.Results: The ENT-HAN and SNT-HAN demonstrated strong performance in both entity extraction and sentiment analysis, outperforming the methods they were compared against. For entity extraction, ENT-HAN was evaluated against RNN and BERT models, showing superior accuracy in identifying multiple entities within complex texts. In sentiment analysis, SNT-HAN was compared to the best-performing method previously applied to FinEntity dataset. Despite the good performance of the existing methods, SNT-HAN demonstrated superior results, achieving a better accuracy.Conclusion: The outcome of this research highlights the potential of the ENT-HAN and SNT-HAN for improving entity extraction and sentiment analysis accuracy in financial documents. Their ability to model attention at multiple levels allows for a more nuanced understanding of text, establishing them as a valuable resource for complex tasks in financial text analysis.
Artificial Intelligence
S. S. Musavian; A. Taghizade; F. Z. Ahmadi; S. Norouzi
Abstract
Background and Objectives: The purpose of this study is to propose a solution for using large fuzzy sets in assessment tasks with a significant number of items, focusing on the assessment of media and educational tools. Ensuring fairness is crucial in evaluation tasks, especially when different evaluators ...
Read More
Background and Objectives: The purpose of this study is to propose a solution for using large fuzzy sets in assessment tasks with a significant number of items, focusing on the assessment of media and educational tools. Ensuring fairness is crucial in evaluation tasks, especially when different evaluators assign different ratings to the same process or their ratings may even vary in different situations. Also, previous non-fuzzy assessment methods show that the mean value of assessors scores is not a good representation when the variance of scores is significant. Fuzzy evaluation methods can solve this problem by addressing the uncertainty in evaluation tasks. Although some studies have been conducted on fuzzy assessment, but their main focus is fuzzy calculations and no solution has been proposed for the problem arising when fuzzy rule set is considerably huge. Methods: Fuzzy rules are the main key for fuzzy inference. This part of a fuzzy system often is generated by experts. In this study,15 experts were asked to create the set of fuzzy rules. Fuzzy rules relate inputs to outputs by descriptive linguistic expressions. Making these expressions is so more convenient than if we determine an exact relationship between inputs and outputs. The number of fussy rules has an exponential relationship with the number of inputs. Therefore, for a task with more than say 6 inputs, we should deal with a huge set of fuzzy rules. This paper presents a solution that enables the use of large fuzzy sets in fuzzy systems using a multi-stage hierarchical approach.Results: Justice is always the most important issue in an assessment process. Due to its nature, a fuzzy calculation-based assessment provides an assessment in a just manner. Since many assessment tasks are often involved more than 10 items to be assessed, generating a fuzzy rule set is impossible. Results show the final score is very sensitive to slight differences in score of an item given by assessors. Besides that, assessors often are not able to consider all items simultaneously to assign a coefficient for the effect of each item on final score. This will be seriously a problem when the final score depends on many input items. In this study, we proposed a fuzzy analysis method to ensure equitable evaluation of educational media and instructional tools within the teaching process. Results of none-fuzzy scoring system show that final score has intense variations when assessment is down in different times and by different assessors. It is because of the manner that importance coefficients are calculated for each item of assessment. In fuzzy assessment no importance coefficient is used for each item.Conclusion: In this study, a novel method was proposed to determine the score of an activity, a task, or a tool that is designed for learning purposes based on Fuzzy sets and their respective calculations. Because of the nature of fuzzy systems, approximate descriptive expressions are used to relate input items to final score instead of an exact function that is impossible to be estimated. Fuzzy method is a robust system that ensure us a fair assessment.
Artificial Intelligence
B. Mahdipour; S. H. Zahiri; I. Behravan
Abstract
Background and Objectives: Path planning is one of the most important topics related to the navigation of all kinds of moving vehicles such as airplanes, surface and subsurface vessels, cars, etc. Undoubtedly, in the process of making these tools more intelligent, detecting and crossing obstacles without ...
Read More
Background and Objectives: Path planning is one of the most important topics related to the navigation of all kinds of moving vehicles such as airplanes, surface and subsurface vessels, cars, etc. Undoubtedly, in the process of making these tools more intelligent, detecting and crossing obstacles without encountering them by taking the shortest path is one of the most important goals of researchers. Significant success in this field can lead to significant progress in the use of these tools in a variety of applications such as industrial, military, transportation, commercial, etc. In this paper, a metaheuristic-based approach with the introduction of new fitness functions is presented for the problem of path planning for various types of surface and subsurface moving vehicles.Methods: The proposed approach for path planning in this research is based on the metaheuristic methods, which makes use of a novel fitness function. Particle Swarm Optimization (PSO) is the metaheuristic method leveraged in this research but other types of metaheuristic methods can also be used in the proposed architecture for path planning.Results: The efficiency of the proposed method, is tested on two synthetic environments for finding the best path between the predefined origin and destination for both surface and subsurface unmanned intelligent vessels. In both cases, the proposed method was able to find the best path or the closest answer to it.Conclusion: In this paper, an efficient method for the path planning problem is presented. The proposed method is designed using Particle Swarm Optimization (PSO). In the proposed method, several effective fitness function have been defined so that the best path or one of the closest answers can be obtained by utilized metaheuristic algorithm. The results of implementing the proposed method on real and simulated geographic data show its good performance. Also, the obtained quantitative results (time elapsed, success rate, path cost, standard deviation) have been compared with other similar methods. In all of these measurements, the proposed algorithm outperforms other methods or is comparable to them.
Artificial Intelligence
K. Moeenfar; V. Kiani; A. Soltani; R. Ravanifard
Abstract
Background and Objectives: In this paper, a novel and efficient unsupervised machine learning algorithm named EiForestASD is proposed for distinguishing anomalies from normal data in data streams. The proposed algorithm leverages a forest of isolation trees to detect anomaly data instances. Methods: ...
Read More
Background and Objectives: In this paper, a novel and efficient unsupervised machine learning algorithm named EiForestASD is proposed for distinguishing anomalies from normal data in data streams. The proposed algorithm leverages a forest of isolation trees to detect anomaly data instances. Methods: The proposed method EiForestASD incorporates an isolation forest as an adaptable detector model that adjusts to new data over time. To handle concept drifts in the data stream, a window-based concept drift detection is employed that discards only those isolation trees that are incompatible with the new concept. The proposed method is implemented using the Python programming language and the Scikit-Multiflow library.Results: Experimental evaluations were conducted on six real-world and two synthetic data streams. Results reveal that the proposed method EiForestASD reduces computation time by 19% and enhances anomaly detection rate by 9% compared to the baseline method iForestASD. These results highlight the efficacy and efficiency of the EiForestASD in the context of anomaly detection in data streams.Conclusion: The EiForestASD method handles concept change using an intelligent strategy where only those trees from the detector model incompatible with the new concept are removed and reconstructed. This modification of the concept drift handling mechanism in the EiForestASD significantly reduces computation time and improves anomaly detection accuracy.
Artificial Intelligence
M. Soluki; Z. Askarinejadamiri; N. Zanjani
Abstract
Background and Objectives: This article explores a method for generating Persian texts using the GPT-2 language model and the Hazm library. Researchers and writers often require tools that can assist them in the writing process and even think on their behalf in various domains. By leveraging the GPT-2 ...
Read More
Background and Objectives: This article explores a method for generating Persian texts using the GPT-2 language model and the Hazm library. Researchers and writers often require tools that can assist them in the writing process and even think on their behalf in various domains. By leveraging the GPT-2 model, it becomes possible to generate acceptable and creative texts, which increases writing speed and efficiency, thus mitigating the high costs associated with article writing.Methods: In this research, the GPT-2 model is employed to generate and predict Persian texts. The Hazm library is utilized for natural language processing and automated text generation. The results of this study are evaluated using different datasets and output representations, demonstrating that employing the Hazm library with input data exceeding 1000 yields superior outcomes compared to other text generation methodsResults: Through extensive experimentation and analysis, the study demonstrates the effectiveness of this combination in generating coherent and contextually appropriate text in the Persian language. The results highlight the potential of leveraging advanced language models and linguistic processing tools for enhancing natural language generation tasks in Persian. The findings of this research contribute to the growing field of Persian language processing and provide valuable insights for researchers and practitioners working on text generation applications in similar languages.Conclusion: Overall, this study showcases the promising capabilities of the GPT-2 model and Hazm library in Persian text generation, underscoring their potential for future advancements in the field This research serves as a valuable guide and tool for generating Persian texts in the field of research and scientific writing, contributing to cost and time reduction in article writing
Artificial Intelligence
M. Amoozegar; S. Golestani
Abstract
Background and Objectives: In recent years, various metaheuristic algorithms have become increasingly popular due to their effectiveness in solving complex optimization problems across diverse domains. These algorithms are now being utilized for an ever-expanding number of real-world applications across ...
Read More
Background and Objectives: In recent years, various metaheuristic algorithms have become increasingly popular due to their effectiveness in solving complex optimization problems across diverse domains. These algorithms are now being utilized for an ever-expanding number of real-world applications across many fields. However, there are two critical factors that can significantly impact the performance and optimization capability of metaheuristic algorithms. First, comprehensively understanding the intrinsic behavior of the algorithms can provide key insights to improve their efficiency. Second, proper calibration and tuning of an algorithm's parameters can dramatically enhance its optimization effectiveness. Methods: In this study, we propose a novel response surface methodology-based approach to thoroughly analyze and elucidate the behavioral dynamics of optimization algorithms. This technique constructs an informative empirical model to determine the relative importance and interaction effects of an algorithm's parameters. Although applied to investigate the Gravitational Search Algorithm, this systematic methodology can serve as a generally applicable strategy to gain quantitative and visual insights into the functionality of any metaheuristic algorithm.Results: Extensive evaluation using 23 complex benchmark test functions exhibited that the proposed technique can successfully identify ideal parameter values and their comparative significance and interdependencies, enabling superior comprehension of an algorithm's mechanics.Conclusion: The presented modeling and analysis framework leverages multifaceted statistical and visualization tools to uncover the inner workings of algorithm behavior for more targeted calibration, thereby enhancing the optimization performance. It provides an impactful approach to elucidate how parameter settings shape algorithm searche so they can be calibrated for optimal efficiency.
Artificial Intelligence
S. Nemati
Abstract
Background and Objectives: Community question-answering (CQA) websites have become increasingly popular as platforms for individuals to seek and share knowledge. Identifying users with a special shape of expertise on CQA websites is a beneficial task for both companies and individuals. Specifically, ...
Read More
Background and Objectives: Community question-answering (CQA) websites have become increasingly popular as platforms for individuals to seek and share knowledge. Identifying users with a special shape of expertise on CQA websites is a beneficial task for both companies and individuals. Specifically, finding those who have a general understanding of certain areas but lack expertise in other fields is crucial for companies who are planning internship programs. These users, called dash-shaped users, are willing to work for low wages and have the potential to quickly develop into skilled professionals, thus minimizing the risk of unsuccessful recruitment. Due to the vast number of users on CQA websites, they provide valuable resources for finding individuals with various levels of expertise. This study is the first of its kind to directly classify CQA users based solely on the textual content of their posts. Methods: To achieve this objective, we propose an ensemble of advanced deep learning algorithms and traditional machine learning methods for the binary classification of CQA users into two categories: those with dash-shaped expertise and those without. In the proposed method, we used the stack generalization to fuse the results of the dep and machine learning methods. To evaluate the effectiveness of our approach, we conducted an extensive experiment on three large datasets focused on Android, C#, and Java topics extracted from the Stack Overflow website. Results: The results on four datasets of the Stack Overflow, demonstrate that our ensemble method not only outperforms baseline methods including seven traditional machine learning and six deep models, but it achieves higher performance than state-of-the-art deep models by an average of 10% accuracy and F1-measure. Conclusion: The proposed model showed promising results in confirming that by using only their textual content of questions, we can classify the users in CQA websites. Specifically, the results showed that using the contextual content of the questions, the proposed model can be used for detecting the dash-shaped users precisely. Moreover, the proposed model is not limited to detecting dash-shaped users. It can also classify other shapes of expertise, such as T- and C-shaped users, which are valuable for forming agile software teams. Additionally, our model can be used as a filter method for downstream applications, like intern recommendations.
Artificial Intelligence
N. Ghanbari; S. H. Zahiri; H. Shahraki
Abstract
Background and Objectives: In this paper, a new version of the particle swarm optimization (PSO) algorithm using a linear ranking function is proposed for clustering uncertain data. In the proposed Uncertain Particle Swarm Clustering method, called UPSC method, triangular fuzzy numbers (TFNs) are used ...
Read More
Background and Objectives: In this paper, a new version of the particle swarm optimization (PSO) algorithm using a linear ranking function is proposed for clustering uncertain data. In the proposed Uncertain Particle Swarm Clustering method, called UPSC method, triangular fuzzy numbers (TFNs) are used to represent uncertain data. Triangular fuzzy numbers are a good type of fuzzy numbers and have many applications in the real world.Methods: In the UPSC method input data are fuzzy numbers. Therefore, to upgrade the standard version of PSO, calculating the distance between the fuzzy numbers is necessary. For this purpose, a linear ranking function is applied in the fitness function of the PSO algorithm to describe the distance between fuzzy vectors. Results: The performance of the UPSC is tested on six artificial and nine benchmark datasets. The features of these datasets are represented by TFNs.Conclusion: The experimental results on fuzzy artificial datasets show that the proposed clustering method (UPSC) can cluster fuzzy datasets like or superior to other standard uncertain data clustering methods such as Uncertain K-Means Clustering (UK-means) and Uncertain K-Medoids Clustering (UK-medoids) algorithms. Also, the experimental results on fuzzy benchmark datasets demonstrate that in all datasets except Libras, the UPSC method provides better results in accuracy when compared to other methods. For example, in iris data, the clustering accuracy has increased by 2.67% compared to the UK-means method. In the case of wine data, the accuracy increased with the UPSC method is 1.69%. As another example, it can be said that the increase in accuracy for abalone data was 4%. Comparing the results with the rand index (RI) also shows the superiority of the proposed clustering method.
Artificial Intelligence
S. Nemati
Abstract
Background and Objectives: Twitter is a microblogging platform for expressing assessments, opinions, and sentiments on different topics and events. While there have been several studies around sentiment analysis of tweets and their popularity in the form of the number of retweets, predicting the sentiment ...
Read More
Background and Objectives: Twitter is a microblogging platform for expressing assessments, opinions, and sentiments on different topics and events. While there have been several studies around sentiment analysis of tweets and their popularity in the form of the number of retweets, predicting the sentiment of first-order replies remained a neglected challenge. Predicting the sentiment of tweet replies is helpful for both users and enterprises. In this study, we define a novel problem; given just a tweet's text, the goal is to predict the overall sentiment polarity of its upcoming replies.Methods: To address this problem, we proposed a graph convolutional neural network model that exploits the text's dependencies. The proposed model contains two parallel branches. The first branch extracts the contextual representation of the input tweets. The second branch extracts the structural and semantic information from tweets. Specifically, a Bi-LSTM network and a self-attention layer are used in the first layer for extracting syntactical relations, and an affective knowledge-enhanced dependency tree is used in the second branch for extracting semantic relations. Moreover, a graph convolutional network is used on the top of these branches to learn the joint feature representation. Finally, a retrieval-based attention mechanism is used on the output of the graph convolutional network for learning essential features from the final affective picture of tweets.Results: In the experiments, we only used the original tweets of the RETWEET dataset for training the models and ignored the replies of the tweets in the training process. The results on three versions of the RETWEET dataset showed that the proposed model outperforms the LSTM-based models and similar state-of-the-art graph convolutional network models. Conclusion: The proposed model showed promising results in confirming that by using only the content of a tweet, we can predict the overall sentiment of its replies. Moreover, the results showed that the proposed model achieves similar or comparable results with simpler deep models when trained on a public tweet dataset such as ACL 2014 dataset while outperforming both simple deep models and state-of-the-art graph convolutional deep models when trained on the RETWEET dataset. This shows the proposed model's effectiveness in extracting structural and semantic relations in the tweets.
Artificial Intelligence
M. Yousefzadeh; A. Golmakani; G. Sarbishaei
Abstract
Background and Objectives: To design an efficient tracker in a crowded environment based on artificial intelligence and image processing, there are several challenges such as the occlusion, fast motion, in-plane rotation, variations in target illumination and Other challenges of online tracking are the ...
Read More
Background and Objectives: To design an efficient tracker in a crowded environment based on artificial intelligence and image processing, there are several challenges such as the occlusion, fast motion, in-plane rotation, variations in target illumination and Other challenges of online tracking are the time complexity of the algorithm, increasing memory space, and tracker dependence on the target model. In this paper, for the first time, sketch matrix theory in ridge regression for video sequences has been proposed.Methods: A new tracking object method based on the element-wise matrix with an online training method is proposed including the kernel correlation Filter (KCF), circular, and sketch matrix. The proposed algorithm is not only the free model but also increases the robustness of the tracker related to the scale variation, occlusion, fast motion, and reduces KCF drift.Results: The simulation results demonstrate that the proposed sketch kernel correlation filter (SHKCF) can increase the computational speed of the algorithm and reduces both the time complexity and the memory space. Finally, the proposed tracker is implemented and experimentally evaluated based on video sequences of OTB50, OTB100 and VOT2016 benchmarks.Conclusion: The experimental results show that the SHKCF method obtains not only OPE partial evaluation of Out of view, Occlusion and Motion Blur in object accuracy but also achieved the partial evaluation of Illumination Variation, Out of Plane Rotation, Scale Variation, Out of View, Occlusion, In of Plane Rotation, Background Clutter, Fast Motion and Deformation in object overlap which are the first rank compared to the state-the-art works. The result of accuracy, robustness and time complexity are obtained 0.929, 0.93 and 35.4, respectively.
Artificial Intelligence
Z. Ghasemi-Naraghi; A. Nickabadi; R. Safabakhsh
Abstract
Background and Obejctives: Multi-task learning is a widespread mechanism to improve the learning of multiple objectives with a shared representation in one deep neural network. In multi-task learning, it is critical to determine how to combine the tasks loss functions. The straightforward way is to optimize ...
Read More
Background and Obejctives: Multi-task learning is a widespread mechanism to improve the learning of multiple objectives with a shared representation in one deep neural network. In multi-task learning, it is critical to determine how to combine the tasks loss functions. The straightforward way is to optimize the weighted linear sum of multiple objectives with equal weights. Despite some studies that have attempted to solve the realtime multi-person pose estimation problem from a 2D image, major challenges still remain unresolved. Methods: The prevailing solutions are two-stream, learning two tasks simultaneously. They intrinsically use a multi-task learning approach for predicting the confidence maps of body parts and the part affinity fields to associate the parts to each other. They optimize the average of the two tasks loss functions, while the two tasks have different levels of difficulty and uncertainty. In this work, we overcome this problem by applying a multi-task objective that captures task-based uncertainties without any additional parameters. Since the estimated poses can be more certain, the proposed method is called “CertainPose”. Results: Experiments are carried out on the COCO keypoints data sets. The results show that capturing the task-dependent uncertainty makes the training procedure faster and causes some improvements in human pose estimation. Conclusion: The highlight advantage of our method is improving the realtime multi-person pose estimation without increasing computational complexity.
Artificial Intelligence
H. Karim Tabbahfar; F. Tabib Mahmoudi
Abstract
Background and Objectives: Considering the drought and global warming, it is very important to monitor changes in water bodies for surface water management and preserve water resources in the natural ecosystem. For this purpose, using the appropriate spectral indices has high capabilities to distinguish ...
Read More
Background and Objectives: Considering the drought and global warming, it is very important to monitor changes in water bodies for surface water management and preserve water resources in the natural ecosystem. For this purpose, using the appropriate spectral indices has high capabilities to distinguish surface water bodies from other land covers. This research has a special consideration to the effect of different types of land covers around water bodies. For this reason, two different water bodies, lake and wetland, have been used to evaluate the implementation results.Methods: The main objective of this research is to evaluate the capabilities of the genetic algorithm in optimum selection of the spectral indices extracted from Sentinel-2 satellite image due to distinguish surface water bodies in two case studies: 1) the pure water behind the Karkheh dam and 2) the Shadegan wetland having water mixed with vegetation. In this regard, the set of optimal indices is obtained with the genetic algorithm followed by the support vector machine (SVM) classifier. Results: The evaluation of the classification results based on the optimum selected spectral indices showed that the overall accuracy and Kappa coefficient of the recognized surface water bodies are 98.18 and 0.9827 in the Karkheh dam and 98.04 and 0.93 in Shadegan wetland, respectively. Evaluation of each of the spectral indices measured in both study areas was carried out using quantitative decision tree (DT) classifier. The best obtained DT classification results show the improvements in overall accuracy by 1.42% in the Karkheh Dam area and 1.56% in the Shadegan Wetland area based on the optimum selected indices by genetic algorithm followed by SVM classifier. Moreover, the obtained classification results are superior compared with Random Forest classifier using the optimized set of spectral features.Conclusion: Applying the genetic algorithm on the spectral indices was able to obtain two optimal sets of effective indices that have the highest amount of accuracy in classifying water bodies from other land cover objects in the study areas. Considering the collective performance, genetic algorithm selects an optimal set of indices that can detect water bodies more accurately than any single index.
Artificial Intelligence
S. Kalantary; J. Akbari Torkestani; A. Shahidinejad
Abstract
Background and Objectives: With the great growth of applications sensitive to latency, and efforts to reduce latency and cost and to improve the quality of service on the Internet of Things ecosystem, cloud computing and communication between things and the cloud are costly and inefficient; Therefore, ...
Read More
Background and Objectives: With the great growth of applications sensitive to latency, and efforts to reduce latency and cost and to improve the quality of service on the Internet of Things ecosystem, cloud computing and communication between things and the cloud are costly and inefficient; Therefore, fog computing has been proposed to prevent sending large volumes of data generated by things to cloud centers and, if possible, to process some requests. Today's advances in 5G networks and the Internet of Things show the benefits of fog computing more than ever before, so that services can be delivered with very little delay as resources and features of fog nodes approach the end user.Methods: Since the cloud-fog paradigm is a layered architecture, to reduce the overall delay, the fog layer is divided into two sub-layers in this paper, including super nodes and ordinary nodes in order to use the coverage of super peer networks to use the connections between fog nodes in addition to taking advantage of the features of that network and improving the performance of large-scale systems. It causes fog nodes to interact with each other in processing requests and fewer data will be sent to the cloud, resulting in a reduction in overall latency. To reduce the cost of bandwidth used among fog nodes, we have organized a sub-layer of super nodes in the form of a Perfect Difference Graph (PDG). The new platform proposed for aggregation of fog computing and Internet of Things (FOT) is called the P2P-based Fog supported Platform (PFP).Results: We evaluate the utility of our proposed method by applying ifogsim simulator and the results achieved are as follows: (1) power consumption parameter in our proposed method 24% and 38% have improved compared to the structure three-layer fog computing architecture and without fog layer respectively; (2) network usage parameter in our proposed method 26% and 32% have improved compared to the structure three-layer fog computing architecture and without fog layer respectively; (3) average response time parameter in our proposed method 17% and 58% have improved compared to the structure three-layer fog computing architecture and without fog layer respectively; and (4) delay parameter in our proposed method 1% and 0.4% have improved compared to the structure three-layer fog computing architecture and without fog layer respectively.Conclusion: Numerical results obtained from the simulation show that the delay and cost parameters are significantly improved compared to the structure without fog layer and three-layer fog computing architecture. Also, the results show that increasing number of things has the same effect in all cases.
Artificial Intelligence
M. Khazaei
Abstract
Background and Objectives: IP multimedia subsystems (IMS) have been introduced as the Next Generation Network (NGN) platform while considering Session Initiation Protocol (SIP) as the signaling protocol. SIP lacks a proper overload mechanism. Hence, this challenge causes decline in the multimedia QoS. ...
Read More
Background and Objectives: IP multimedia subsystems (IMS) have been introduced as the Next Generation Network (NGN) platform while considering Session Initiation Protocol (SIP) as the signaling protocol. SIP lacks a proper overload mechanism. Hence, this challenge causes decline in the multimedia QoS. The main propose of overload control mechanism is to keep the network throughput at the same network capacity with overload.Methods: NGN distributed with IMS is a complex innovative network consisting of interacting subsystems. Hence, multi-agent systems (MAS) receiving further attention for solving complex problems can solve the problem of overload in these networks. To this end, each IMS server is considered as an intelligent agent that can learn and negotiate with other agents while maintaining autonomy, thus eliminating the overload by communication and knowledge transfer between the agents. In the present research, using MAS and their properties, the intelligent hop by hop method is provided based on Q-learning and negotiation capability for the first time.Results: In the proposed method, parameters of overload controller are obtained by reinforcement learning. In order to check the validity of controller performance, a comparison is made with the similar method in which the optimal parameters are achieved based on trial and error. The result of the comparison confirms the validity of the proposed method. In order to evaluate the efficiency of the learner method, it is compared with similar and standard methods, for which the results are compared to show performance. The results show, the proposed method has approximately improved the throughput by 13%, the delay by 49% and the number of rejected sessions by 17% compare with methods, passing control messages through the network such as CPU occupancy methods. While compare with external controller methods like holonic, throughput is improved by 1% and the number of rejected requests is decreased by 10%, but delay is increased by 6% due to the convergence time of the learning and negotiation process.Conclusion: To overcome overload, complex IMS servers are considered as learner and negotiator agents. This is a new method to achieve the required parameters without relying on expert knowledge or person as well as, heterogeneous IMS entities can be inserted into the problem to complete study in future.
Artificial Intelligence
S.M. Notghimoghadam; H. Farsi; S. Mohamadzadeh
Abstract
Background and Objectives: Object detection has been a fundamental issue in computer vision. Research findings indicate that object detection aided by convolutional neural networks (CNNs) is still in its infancy despite -having outpaced other methods. Methods: This study proposes a straightforward, ...
Read More
Background and Objectives: Object detection has been a fundamental issue in computer vision. Research findings indicate that object detection aided by convolutional neural networks (CNNs) is still in its infancy despite -having outpaced other methods. Methods: This study proposes a straightforward, easily implementable, and high-precision object detection method that can detect objects with minimum least error. Object detectors generally fall into one-stage and two-stage detectors. Unlike one-stage detectors, two-stage detectors are often more precise, despite performing at a lower speed. In this study, a one-stage detector is proposed, and the results indicated its sufficient precision. The proposed method uses a feature pyramid network (FPN) to detect objects on multiple scales. This network is combined with the ResNet 50 deep neural network. Results: The proposed method is trained and tested on Pascal VOC 2007 and COCO datasets. It yields a mean average precision (mAP) of 41.91 in Pascal Voc2007 and 60.07% in MS COCO. The proposed method is tested under additive noise. The test images of the datasets are combined with the salt and pepper noise to obtain the value of mAP for different noise levels up to 50% for Pascal VOC and MS COCO datasets. The investigations show that the proposed method provides acceptable results. Conclusion: It can be concluded that using deep learning algorithms and CNNs and combining them with a feature network can significantly enhance object detection precision.
Artificial Intelligence
S.V. Moravvej; M.J. Maleki Kahaki; M. Salimi Sartakhti; M. Joodaki
Abstract
Background and Objectives: Text summarization plays an essential role in reducing time and cost in many domains such as medicine, engineering, etc. On the other hand, manual summarization requires much time. So, we need an automated system for summarizing. How to select sentences is critical in summarizing. ...
Read More
Background and Objectives: Text summarization plays an essential role in reducing time and cost in many domains such as medicine, engineering, etc. On the other hand, manual summarization requires much time. So, we need an automated system for summarizing. How to select sentences is critical in summarizing. Summarization techniques that have been introduced in recent years are usually greedy in the choice of sentences, which leads to a decrease in the quality of the summary. In this paper, a non-greedily method for selecting essential sentences from a text is presented.Methods: The present paper presents a method based on a generative adversarial network and attention mechanism called GAN-AM for extractive summarization. Generative adversarial networks have two generator and discriminator networks whose parameters are independent of each other. First, the features of the sentences are extracted by two traditional and embedded methods. We extract 12 traditional features. Some of these features are extracted from sentence words and others from the sentence. In addition, we use the well-known Skip-Gram model for embedding. Then, the features are entered into the generator as a condition, and the generator calculates the probability of each sentence in summary. A discriminator is used to check the generated summary of the generator and to strengthen its performance. We introduce a new loss function for discriminator training that includes generator output, real and fake summaries of each document. During training and testing, each document enters the generator with different noises. It allows the generator to see many combinations of sentences that are suitable for quality summaries.Results: We evaluate our results on CNN/Daily Mail and Medical datasets. Summaries produced by the generator show that our model performs better than other methods compared based on the ROUGE metric. We apply different sizes of noise to the generator to check the effect of noise on our model. The results indicate that the noise-free model has poor results.Conclusion: Unlike recent works, in our method, the generator selects sentences non-greedily. Experimental results show that the generator with noise can produce summaries that are related to the main subject.
