August 12-13, 2023, Virtual Conference
Umay Kulsoom1, Dr. M. Naufal B. M. Saad1, Dr. Syed Saad Azhar Ali2, 1Centre for Intelligent Signal & Imaging Research, Universiti Teknologi Petronas, Perak, Malaysia, 2Department of Aerospace Engineering, King Fahd University of Petroleum and Minerals, Dhahran, Saudi Arabia
Personality classification is critical for gaining a better understanding of human behaviour and individual variances. Understanding people’ cognitive and emotional responses to stressful situations, as well as their potential consequences for mental health, targeted care, workforce adaptability, communication, and public health message, is facilitated by personality classification. Resting state electroencephalogram (EEG) provides a realistic assessment of brain function and connectivity by capturing the intrinsic brain activity and allowing us to understand the underlying neural processes. Patterns and features identified from resting state EEG data analysis can serve as potential biomarkers for personality assessment. These biomarkers, paired with the capabilities of machine learning, can be helpful in building automated personality classifiers with higher accuracy. To investigate the feasibility of personality classification from resting state EEG, power spectral characteristics derived from EEG and self-reported evaluations (NEO-FFI scores) are fed into the Support Vector Machine (SVM) classifier. The initial results have demonstrated encouraging outcomes and supported our claim that personality traits can be predicted from resting state EEG data. According to our findings, it is possible to classify personality traits according to the Big Five model with an approximate accuracy of 70%.
Big Five Model, Resting State EEG, Machine Learning, Power Spectral Density (PSD).
Fouzi Takelait1 and Omar Mohamed Ahmed2, 1Kennedy College of Sciences, University of Massachusetts Lowell, Lowell, USA, 2Department of Computer Science, Helwan University, Cairo, Egypt
Natural Language Inference (NLI) is a crucial aspect of Natural Language Processing (NLP) that involves classifying the relationship between two sentences into categories such as entailment or contradiction. In this paper, we propose a novel approach for Arabic NLI utilizing pre-trained transformer models - Arabert-v2, Marbert , and Qarib. The existing methods mostly focus on syntactic, lexical, and semantic strategies, but these approaches lack the sophistication of understanding intricate language semantics.The novelty of our work lies in the ability of transformer models to capture both lexical and semantic features, thus potentially outperforming traditional NLI methods. We present a comprehensive study and comparison of different pre-trained models using the Arabic datasets ArNLi and ArbTEDS. Our results are compared with state-of-the-art methods to provide a new benchmark in the literature. Our work aims to contribute to the evolving landscape of Arabic NLI, providing advanced methods for further research and applications in the field. By improving the capability to infer relationships between sentences, we hope to refine machine understanding of Arabic language, paving the way for more sophisticated applications in areas like machine translation, question answering, and text summarization.
Natural Language Inferences, NLP, Transformers.
Bibi Qurat Ul Ain and Bingcai Chen, Department of Computer Science and Engineering, Dalian University of Technology, Dalian 116024, China
Pneumonia, a Significant Global Cause of Mortality, Especially in Pakistan, Poses Challenges for Accurate Diagnosis From Chest Radiographs, Even for Expert Radiologists. Consequently, the Development of an Automatic, Rapid, and Precise Pneumonia Detection System Would Greatly Benefit Both Patients and Physicians. Recent Literature Has Explored Diverse Deep Learning Algorithms for Pneumonia Detection, but Their Practicality is Limited by High Computational Demands and the Need for Fast Gpus. In This Study, a Lightweight Approach Leveraging Transfer Learning of Pre-trained Architectures (Ssd Mobilenet V2, Ssd Mobilenet V2 Fpnlite 320x320, and Ssd Mobilenet V2 Fpnlite 640x640) Was Employed, Followed by Comparative Analysis of These Pre-trained Models for Pneumonia Detection. The Proposed Models Underwent Extensive Testing in Various Scenarios, Employing Different Dataset Distributions, Hyper-parameters, Classification Loss Functions, and Image Pre-processing Techniques Using a Set of Evaluation Metrics. The Ssd Mobilenetv2, Ssd Mobilenetv2 Fpnlite 320x320, and Ssd Mobilenetv2 Fpnlite 640x640 Models Achieved Map Scores of 76%, 85%, and 80%, Respectively, Alongside Accuracies of 81.3%, 94.6%, and 92.6% on an Unseen Dataset From the Ghuangzhou Women and Children's Medical Center Pneumonia Dataset. Notably, the Ssd Mobilenetv2 Fpnlite 320x320 Model Exhibited Superior Performance Among the Three. These Findings Demonstrate Great Potential in Accurately Detecting Pneumonia Cases in Medical Images, Offering Computational Efficiency, Cost-effectiveness, and Faster Results Compared to Existing Methods in the Literature.
Deep Learning, Pneumonia Detection, Convolutional Neural Network (CNN), Transfer Learning, Single Shot Detector (SSD), MobileNetV2.