1. Elegante+: A Machine Learning-Based Optimization Framework for Sparse Matrix–Vector Computations on the CPU Architecture

Sparse matrix–vector multiplication (SpMV) plays a significant role in the computational costs of many scientific applications such as 2D/3D robotics, power network problems, and computer vision. Numerous implementations using different sparse matrix formats have been introduced to optimize this kernel on CPUs and GPUs. However, due to the sparsity patterns of matrices and the diverse configurations of hardware, accurately modeling the performance of SpMV remains a complex challenge. SpMV computation is often a time-consuming process because of its sparse matrix structure. To address this, we propose a machine learning-based tool, namely Elegante+, that predicts optimal scheduling policies by analyzing matrix structures. This approach eliminates the need for repetitive trial and error, minimizes errors, and finds the best solution of the SpMV kernel, which enables users to make informed decisions about scheduling policies that maximize computational efficiency. For this purpose, we collected 1000+ sparse matrices from the SuiteSparse matrix market collection and converted them into the compressed sparse row (CSR) format, and SpMV computation was performed by extracting 14 key sparse matrix features. After creating a comprehensive dataset, we trained various machine learning models to predict the optimal scheduling policy, significantly enhancing the computational efficiency and reducing the overhead in high-performance computing environments. Our proposed tool, Elegante+ (XGB with all SpMV features), achieved the highest cross-validation score of 79% and performed five times faster than the default scheduling policy during SpMV in a high-performance computing (HPC) environment.

2. Fine-Tuned RoBERTa Model for Bug Detection in Mobile Games: A Comprehensive Approach

In the current digital era, the Google Play Store and the App Store are major platforms for the distribution of mobile applications and games. Billions of users regularly download mobile games and provide reviews, which serve as a valuable resource for game vendors and developers, offering insights into bug reports, feature suggestions, and documentation of existing functionalities. This study showcases an innovative application of fine-tuned RoBERTa for detecting bugs in mobile phone games, highlighting advanced classification capabilities. This approach will increase player satisfaction, lead to higher ratings, and improve brand reputation for game developers, while also reducing development costs and saving time in creating high-quality games. To achieve this goal, a new bug detection dataset was created. Initially, data were sourced from four top-rated mobile games from multiple domains on the Google Play Store and the App Store, focusing on bugs, using the Google Play API and App Store API. Subsequently, the data were categorized into two classes: binary and multi-class. The Logistic Regression, Convolutional Neural Network (CNN), and pre-trained Robustly Optimized BERT Approach (RoBERTa) algorithms were used to compare the results. We explored the strength of pre-trained RoBERTa, which demonstrated its ability to capture both semantic nuances and contextual information within textual content. The results showed that pre-trained RoBERTa significantly outperformed the baseline models (Logistic Regression), achieving superior performance with a 5.49% improvement in binary classification and an 8.24% improvement in multi-class classification, resulting in cross-validation scores of 96% and 92%, respectively.

3. Multilingual hope speech detection from tweets using transfer learning models

Social media has become a powerful tool for public discourse, shaping opinions and the emotional landscape of communities. The extensive use of social media has led to a massive influx of online content. This content includes instances where negativity is amplified through hateful speech but also a significant number of posts that provide support and encouragement, commonly known as hope speech. In recent years, researchers have focused on the automatic detection of hope speech in languages such as Russian, English, Hindi, Spanish, and Bengali. However, to the best of our knowledge, detection of hope speech in Urdu and English, particularly using translation-based techniques, remains unexplored. To contribute to this area we have created a multilingual dataset in English and Urdu and applied a translation-based approach to handle multilingual challenges and utilized several state-of-the-art machine learning, deep learning, and transfer learning based methods to benchmark our dataset. Our observations indicate that a rigorous process for annotator selection, along with detailed annotation guidelines, significantly improved the quality of the dataset. Through extensive experimentation, our proposed methodology, based on the Bert transformer model, achieved benchmark performance, surpassing traditional machine learning models with accuracies of 87% for English and 79% for Urdu. These results show improvements of 8.75% in English and 1.87% in Urdu over baseline models (SVM 80% English and 78% in Urdu).

4. Predictive Analysis of Breast Cancer Survival Using Machine Learning, Deep Learning, and GPT-4

Breast cancer remains one of the leading causes of cancer-related deaths worldwide, making accurate survival prediction essential for improving patient care and treatment decisions. Advances in artificial intelligence, particularly deep learning (DL) models, offer promising solutions for enhancing predictive accuracy. In this study, we compare the performance of conventional machine learning (ML) models and DL models using a dataset of 4,024 patients. We evaluate Long Short-Term Memory (LSTM), Bidirectional LSTM (BiLSTM), Support Vector Machine (SVM), Logistic Regression, and K-Nearest Neighbors (KNN), alongside a fine-tuned GPT-4 model. Our results indicate that BiLSTM outperforms other models, achieving an F1 score of 0.95 and an accuracy of 91% for the “Alive” class, followed closely by LSTM with an F1 score of 0.94 and an accuracy of 90%. Traditional ML models such as SVM and Logistic Regression perform moderately well in detecting the “Dead” class, with F1 scores of 0.60 and 0.58, respectively. KNN lags behind with an F1 score of 0.47. Additionally, the fine-tuned GPT-4 model demonstrates strong predictive capability, reaching an F1 score of 0.97 for the “Alive” class and an overall accuracy of 96.67%. While deep learning models, particularly BiLSTM and GPT-4, prove effective in predicting survival, challenges remain in addressing class imbalance to improve prediction for the minority “Dead” class. This study highlights the potential of deep learning in medical prognosis and underscores the need for further optimization to ensure balanced performance across all patient groups.

5. Hate Speech Detection Using Social Media Discourse: A Multilingual Approach with Large Language Model

Online social networks (OSN) and microblogging websites are attracting Internet users and have revolutionized how we communicate with individuals, share their feelings, and exchange ideas across the world with ease. In the extensive age of social media, there is increasing online hate speech, which can provoke violence and contribute to societal division. Hate speech based on race, gender, or religion puts those affected at risk of mental health problems and exacerbates social problems. While current protocols have reduced overt hate speech, subtler forms known as implicit hate speech have emerged, making detection more challenging. This study focuses on hate speech detection using social media discourse, by creating a comprehensive multilingual dataset [25] in Urdu and English and applied multiple machine learning, deep learning, transfer learning, and Large Language model models, such as GPT-3.5 Turbo. By comparing GPT-3.5 Turbo, we identified the effectiveness of large language models in detecting both explicit and implicit forms of hate speech. Our analysis underscores the potential of automated classification systems to reduce reliance on human intervention and to promote constructive online discourse. Our proposed methodology achieved the highest accuracy of 0.91, and achieved the highest performance improvement of 5.81% over transformer models such as BERT. This research adds to the growing body of work on multilingual natural language processing (NLP) and offers insights for reducing hate speech and fostering respectful communication across diverse communities.

6. UE-NER-2025: A GPT-based Approach to Multilingual Named Entity Recognition on Urdu and English

Named Entity Recognition (NER) is a fundamental task that identifies and classifies entities into predefined categories from unstructured text. As textual data continues to grow and span diverse linguistic communities, NER is rarely studied as a multilingual task, particularly for low-resource languages. While many researchers have focused on name identification in various high-resource languages, only a few research efforts have addressed NER for the Urdu script. This is primarily due to a lack of resources and annotated datasets. Furthermore, previous research has mostly concentrated on monolingual techniques, leaving significant gaps in addressing multilingual challenges, especially for the Urdu language. To fill this gap, this study makes four key contributions. First, we created a unique multilingual dataset (UE-NER-2025) sourced from Twitter, which contains 182,411 tokens and 8 uniquely annotated entity types. Second, we applied two novel techniques that are relatively new to the UE-NER-2025 dataset: 1) a joint multilingual approach and 2) a joint translation-based approach. Third, we conducted 30 different experiments using 5-fold cross-validation, combining traditional supervised learning with token-based feature extraction, deep learning with pre-trained word embeddings such as FastText and GloVe, and advanced transfer learning models using contextual embeddings, to evaluate their effectiveness in enhancing NER performance for both English and Urdu, particularly addressing the challenges of low-resource and morphologically rich languages. Finally, we performed statistical analysis on our top-performing models to determine whether the differences in performance were statistically significant or occurred by chance. Based on the analysis of the results, our transformer-based language model (XLM-RoBERTa-base) achieved strong performance compared to traditional supervised learning models. We observed a performance improvement of 3.99% in the English translation-based approach, 3.72% in the multilingual approach, and 2.32% in the Urdu translation-based approach over traditional supervised learning (RF in Urdu = 0.927, in English = 0.9258, and multilingual = 0.9272).

7. Hope Speech Detection Using Social Media Discourse (Posi-Vox-2024): A Transfer Learning Approach

Background: The notion of hope is characterized as an optimistic expectation or anticipation of favorable outcomes. In the age of extensive social media usage, research has primarily focused on monolingual techniques, and the Urdu and Arabic languages have not been addressed. Purpose: This study addresses joint multilingual hope speech detection in the Urdu, English, and Arabic languages using a transfer learning paradigm. We developed a new multilingual dataset named Posi-Vox-2024 and employed a joint multilingual technique to design a universal classifier for multilingual dataset. We explored the fine-tuned BERT model, which demonstrated a remarkable performance in capturing semantic and contextual information. Method: The framework includes (1) preprocessing, (2) data representation using BERT, (3) fine-tuning, and (4) classification of hope speech into binary (‘hope’ and ‘not hope’) and multi-class (realistic, unrealistic, and generalized hope) categories. Results: Our proposed model (BERT) demonstrated benchmark performance to our dataset, achieving 0.78 accuracy in binary classification and 0.66 in multi-class classification, with a 0.04 and 0.08 performance improvement over the baselines (Logistic Regression, in binary class 0.75 and multi class 0.61), respectively. Conclusion: Our findings will be applied to improve automated systems for detecting and promoting supportive content in English, Arabic and Urdu on social media platforms, fostering positive online discourse. This work sets new benchmarks for multilingual hope speech detection, advancing existing knowledge and enabling future research in underrepresented languages.

8. Automated diagnosis of lung diseases using vision transformer: a comparative study on chest X-ray classification

Background: Lung disease is a significant health issue, particularly in children and elderly individuals. It often results from lung infections and is one of the leading causes of mortality in children. Globally, lung-related diseases claim many lives each year, making early and accurate diagnoses crucial. Radiographs are valuable tools for the diagnosis of such conditions. The most prevalent lung diseases, including pneumonia, asthma, allergies, chronic obstructive pulmonary disease (COPD), bronchitis, emphysema, and lung cancer, represent significant public health challenges. Early prediction of these conditions is critical, as it allows for the identification of risk factors and implementation of preventive measures to reduce the likelihood of disease onset. Methods: In this study, we utilized a dataset comprising 3,475 chest X-ray images sourced from Mendeley Data provided by Talukder, M. A. (2023) [14], categorized into three classes: normal, lung opacity, and pneumonia. We applied five pre-trained deep learning models, including CNN, ResNet50, DenseNet, CheXNet, and U-Net, as well as two transfer learning algorithms such as Vision Transformer (ViT) and Shifted Window (Swin) to classify these images. This approach aims to address diagnostic issues in lung abnormalities by reducing reliance on human intervention through automated classification systems. Our analysis was conducted in both binary and multiclass settings. Results: In the binary classification, we focused on distinguishing between normal and viral pneumonia cases, whereas in the multi-class classification, all three classes (normal, lung opacity, and viral pneumonia) were included. Our proposed methodology (ViT) achieved remarkable performance, with accuracy rates of 99% for binary classification and 95.25% for multiclass classification.