deep learning

Visual object tracking has recently benefited from the adoption of transformer architectures, which provide strong modeling capacity but incur high computational and memory costs, limiting real-time deployment.  Existing efficiency-focused trackers primarily address this challenge through architectural redesign, often trading accuracy for speed.

Violence detection in video surveillance systems is a critical challenge for ensuring public safety in smart cities.  This study presents a comprehensive analysis of deep learning architectures and transfer learning techniques for violence detection, evaluating their performance across benchmark datasets, including Hockey Fight, Movies, and UCF-Crime.  ResNet50, MobileNetV2, ConvLSTM, DenseNet121, and Xception models are compared in terms of accuracy, computational efficiency, and real-world applicability.  Results highlight ResNet50 as the most effective model, achievi

Traditional biomarker testing for Programmed Death-Ligand 1 (PD-L1) in Non-Small Cell Lung Cancer (NSCLC) remains invasive and costly.  This study proposes a non-invasive alternative by integrating radiomic features extracted from CT scans with advanced deep learning architectures.  We evaluated Artificial Neural Networks (ANN), Convolutional Neural Networks (CNN), and Recurrent Neural Networks (RNN), including Long Short-Term Memory (LSTM) and Gated Recurrent Units (GRU).  Our results demonstrate that Transformer-based models significantly outperform conventional appro

With the advancement of blockchain technology, Solidity-based smart contracts have become essential for automating and securing digital transactions across various sectors, from finance to supply chain management.  These contracts enable decentralized exchanges without intermediaries, enhancing transparency.  However, their immutable nature poses security challenges: any flaw in the code becomes permanent, exposing contracts to attacks and leading to financial and reputational losses.  This paper provides a comparative analysis of recent machine learning (ML) and deep l

Stock price values are known for their volatility due to multiple factors making their predictability a difficult task.  As social media posts and news can be considered as one of the major factors in price change, we aim in this paper to predict the next-day stock price of 4 different companies, using both social media and financial datasets that range from September 30, 2021, to September 30, 2022, as inputs.  The datasets go through a preprocessing pipeline that includes sentiment analysis methods, where tweets are classified by employing TextBlob and finetuned RoBER

The increasing prevalence of automated social media accounts, or Social Media Bots (SMBs), presents significant challenges in maintaining authentic online discourse and preventing disinformation campaigns on social platforms.  This research introduces a novel multiclass classification framework for detecting and categorizing SMBs, leveraging fine-tuned transformer-based models.  In this study, we conducted a comprehensive comparative analysis of various transformer variants, including BERT, DistilBERT, RoBERTa, DeBERTa, XLNet, and ALBERT, to evaluate their efficacy in r

This study developed a real-time fish classification and counting system for six types of fish using the YOLOv5 machine learning model with high accuracy.  The system achieved an F1-score of 0.87 and a precision confidence curve with an all-classes value of 1.00 at a confidence level of 0.920, demonstrating the model's reliability in object detection and classification.  Real-time testing showed that the system could operate quickly and accurately under various environmental conditions with an average inference speed of 30 FPS.  However, several challenges remain, such