Decoding Neural Networks: An Information-Theoretic Guide To Interpretability, Error Analysis And Efficiency,
2024
Florida Institute of Technology
Decoding Neural Networks: An Information-Theoretic Guide To Interpretability, Error Analysis And Efficiency, Mackenzie J. Meni
Theses and Dissertations
This dissertation addresses critical challenges in neural network design by leveraging entropy-based techniques to improve model efficiency, interpretability, and bias reduction. Focusing on the unique demands of computer vision applications, particularly object detection and classification for real-time systems, this work introduces a series of innovative methods centered on information theory. At the core of these methods is the Probabilistic Explanations of Entropic Knowledge (PEEK) framework, a tool developed to analyze and visualize entropy distributions across feature maps. PEEK offers insights into information flow within neural networks, making it possible to pinpoint layers that contribute meaningfully to decision-making or identify those …
Debtor Eligibility Prediction Using Deep Learning With Chatbot-Based Testing,
2024
Universitas Teknologi Yogyakarta, Indonesia
Debtor Eligibility Prediction Using Deep Learning With Chatbot-Based Testing, Reski Noviania, Enny Itje Sela, Luther Alexander Latumakulita, Steven R. Sentinuwo
Knowledge Engineering and Data Science
Predicting debtor eligibility is essential for effective risk management and minimizing lousy credit risks. However, financial institutions face challenges such as imbalanced data, inefficient feature selection, and limited user accessibility. This study combines Recursive Feature Elimination (RFE) and Deep Learning (DL) to improve prediction accuracy. It integrates a chatbot interface for user-friendly testing. RFE effectively identifies critical features, while the DL model achieves a validation accuracy of 97.62%, surpassing previous studies with less comprehensive methodologies. The chatbot's novel design not only ensures accessibility but also enhances user engagement through flexible input options, such as approximate values, enabling non experts to …
Manifold Learning And Undersampling Approaches For Imbalanced Class Sentiment Classification,
2024
IPB University, Indonesia
Manifold Learning And Undersampling Approaches For Imbalanced Class Sentiment Classification, L.M. Risman Dwi Jumansyah, Agus Mohamad Soleh, Utami Dyah Syafitri
Knowledge Engineering and Data Science
Movie reviews are crucial in determining a film's success by influencing audience decisions. Automating sentiment classification is essential for efficient public opinion analysis. However, it faces challenges such as high-dimensional data and imbalanced class distributions. This study addresses these issues by applying manifold learning techniques, Principal Component Analysis (PCA) and Laplacian Eigenmaps (LE) to reduce data complexity and undersampling strategies (Random Undersampling (RUS) and EasyEnsemble) to balance data and improve predictions for both sentiment classes. On reviews of The Raid 2: Berandal, EasyEnsemble achieved the highest average G-Mean of 0.694 using Term Frequency-Inverse Document Frequency (TF IDF) features with a …
Deep Learning Approach For Dental Anomalies X-Ray Imaging Using Yolov8,
2024
University Malaysia, Malaysia
Deep Learning Approach For Dental Anomalies X-Ray Imaging Using Yolov8, Amelia Ritahani Ismail, Md Salim Sadman Taseen
Knowledge Engineering and Data Science
Dental X-ray imaging is a critical diagnostic tool for identifying various dental anomalies. However, manual interpretation is time-consuming, prone to human error, and requires specialized expertise. Deep learning models, particularly object detection frameworks like YOLO, have demonstrated promising results in automating medical image analysis. This study aims to develop and evaluate a YOLOv8-based deep learning model for automated detection and classification of 14 dental anomaly categories, including Caries, Crowns, Fillings, Implants, and Periapical lesions. The proposed approach addresses limitations in previous YOLO versions by leveraging anchor-free detection and enhanced feature extraction for improved accuracy. The model was trained on a …
Optimal Strategy For Handling Unbalanced Medical Datasets: Performance Evaluation Of K-Nn Algorithm Using Sampling Techniques,
2024
Universitas Muslim Indonesia, Indonesia
Optimal Strategy For Handling Unbalanced Medical Datasets: Performance Evaluation Of K-Nn Algorithm Using Sampling Techniques, Yulita Salim, Aulia Putri Utami, Abdul Rachman Manga, Huzain Azis, Fadhila Tangguh Admojo
Knowledge Engineering and Data Science
This study addresses the critical role of medical image classification in enhancing healthcare effectiveness and tackling the challenges of imbalanced medical datasets. It focuses on optimizing classification performance by integrating Canny edge detection for segmentation and Hu-moment feature extraction and applying oversampling and undersampling techniques. Five diverse medical datasets were utilized, covering Alzheimer’s and Parkinson’s diseases, COVID-19, brain tumours, and lung cancer. The K-Nearest Neighbors (K-NN) algorithm was implemented to enhance classification accuracy, aiming to develop a more robust framework for medical image analysis. The evaluation, conducted using cross-validation, demonstrated notable improvements in key metrics. Specifically, oversampling significantly enhanced lung …
A Hierarchical Density-Based Spatial Clustering Of Applications With Noise (Hdbscan) Approach For Identifying Potential Villages In Buleleng Regency,
2024
Universitas Islam Indonesia, Indnonesia
A Hierarchical Density-Based Spatial Clustering Of Applications With Noise (Hdbscan) Approach For Identifying Potential Villages In Buleleng Regency, Dina Nur Amalina, Achmad Fauzan
Knowledge Engineering and Data Science
Buleleng Regency, located in Bali Province, possesses diverse village potential, including agricultural production and tourist attractions. However, this potential has not been fully optimized. Therefore, it is important to enhance village potential by clustering villages based on their specific characteristics to identify and prioritize those requiring special attention. This approach aims to promote equitable village development and reduce poverty levels. This study clusters villages in Buleleng Regency based on their potential using the Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN) method. The data utilized in this study comprises village potential data obtained from the Buleleng Regency Statistics Office …
Classification Of Anxiety Levels Entering The World Of Work In Final Year Students Using The Neighbor Weighted K-Nearest Neighbor Method,
2024
Universitas Pembangunan Nasional Veteran Yogyakarta, Indonesia
Classification Of Anxiety Levels Entering The World Of Work In Final Year Students Using The Neighbor Weighted K-Nearest Neighbor Method, Awang Hendrianto Pratomo, Muhammad Fahmi Adam, Dessyanto Boedi Prasetyo
Knowledge Engineering and Data Science
This study evaluates the accuracy of the Neighbor Weighted K-Nearest Neighbor (NWKNN) method in classifying the anxiety levels of final-year students as they prepare to enter the workforce, particularly in cases of unbalanced data distribution. The system was developed using the prototype method, and NWKNN was applied to classify anxiety levels into low, medium, and high categories. Testing using the Confusion Matrix demonstrated strong performance, achieving an accuracy of 94% based on a dataset of 1009 students, with a 90:10 ratio of training to test data. The results indicate that NWKNN effectively provides classification input values, making it a reliable …
Comparative Analysis Of Bpnn And Lvq For Sundanese Character Recognition,
2024
Universitas Mulawarman, Indonesia
Comparative Analysis Of Bpnn And Lvq For Sundanese Character Recognition, Haviluddin Haviluddin, Herman Santoso Pakpahan, Dinda Izmya Nurpadillah, Hario Jati Setyadi, Medi Taruk, Rayner Alfred
Knowledge Engineering and Data Science
The Sundanese script (Aksara Sunda), an essential part of Sundanese cultural heritage, has been used since the 14th century AD. However, recognizing handwritten Sundanese characters remains challenging due to variations in individual writing styles. This study compares the performance of Backpropagation Neural Network (BPNN) and Learning Vector Quantization (LVQ) for recognizing handwritten Sundanese vowel (Swara) characters. A dataset was collected from 15 individuals, each writing seven Sundanese vowel characters, which were then used for training and testing the recognition models. Experimental results show that BPNN outperforms LVQ, achieving a higher classification accuracy (95.23%), lower Mean Squared Error (MSE), and faster …
A Machine Learning Approach For Survival Analysis Of Transplanted Kidneys Based On Donors’ And Recipients’ Factors.,
2024
Florida Institute of Technology
A Machine Learning Approach For Survival Analysis Of Transplanted Kidneys Based On Donors’ And Recipients’ Factors., Alain Edward Despeignes
Theses and Dissertations
Over seven thousand people on average die each year in the United States waiting for an organ transplant due to the shortage of donated organs. With this alarming concern, efforts from the health organizations like the United Network Organ Sharing (UNOS) and government officials have considered avenues to remedy this distress, one of which is to investigate the characteristics among donors and recipients that affects the longevity of donated organs. The goal of this project is to investigate the survival time of transplanted kidneys from 1987 to 2018 with regards to the donors’ and the recipients’ characteristics including gender, ethnicity, …
Anomaly Detection Using Unsupervised Machine Learning Algorithms: A Simulation Study,
2024
The University of Texas Rio Grande Valley
Anomaly Detection Using Unsupervised Machine Learning Algorithms: A Simulation Study, Edmund F. Agyemang
School of Mathematical & Statistical Sciences Faculty Publications
This study presents a comprehensive evaluation of five prominent unsupervised machine learning anomaly detection algorithms: One-Class Support Vector Machine (One-Class SVM), One-Class SVM with Stochastic Gradient Descent (SGD), Isolation Forest (iForest), Local Outlier Factor (LOF), and Robust Covariance (Elliptic Envelope). Through systematic analysis on a synthetically simulated dataset, the study assessed each algorithm’s predictive performance using accuracy, precision, recall, and F1 score specifically for outlier detection. The evaluation reveals that One-Class SVM, Isolation Forest, and Robust Covariance are more effective in identifying outliers in the synthetic simulated dataset, with Isolation Forest slightly outperforming the other algorithms in terms of balancing …
De-Identification Is Not Enough: A Comparison Between De-Identified And Synthetic Clinical Notes,
2024
The Texas Medical Center Library
De-Identification Is Not Enough: A Comparison Between De-Identified And Synthetic Clinical Notes
Faculty, Staff and Student Publications
For sharing privacy-sensitive data, de-identification is commonly regarded as adequate for safeguarding privacy. Synthetic data is also being considered as a privacy-preserving alternative. Recent successes with numerical and tabular data generative models and the breakthroughs in large generative language models raise the question of whether synthetically generated clinical notes could be a viable alternative to real notes for research purposes. In this work, we demonstrated that (i) de-identification of real clinical notes does not protect records against a membership inference attack, (ii) proposed a novel approach to generate synthetic clinical notes using the current state-of-the-art large language models, (iii) evaluated …
Digital Humanities: Using Computational Methods On Literature To Understand Human-Water Relations,
2024
William & Mary
Digital Humanities: Using Computational Methods On Literature To Understand Human-Water Relations, Ariel Yang
Cybersecurity Undergraduate Research Showcase
By using computational techniques to analyze literature, deeper insights can be gained into human-water relationships across different historical and cultural contexts. Natural Language Processing (NLP) and other data science methods can explore applications of traditional ecological knowledge (TEK) and underlying emotions or beliefs in literature to help understand sustainability. Protecting this sensitive cultural data through ethical applications can further secure future implementations of policies, urban planning, and environmental relationships.
Holistic Correlation Measure For Enhanced Encapsulation Of Trait Heterogeneity And Discovery Of Co-Expression,
2024
University of Missouri-St. Louis
Holistic Correlation Measure For Enhanced Encapsulation Of Trait Heterogeneity And Discovery Of Co-Expression, Zachary Valleroy
Theses
Large-scale, high-dimensional data analyses can be computationally prohibitive due to combinatorial explosion of the search space for finding complex patterns; a viable alternative is network modeling for abstraction and quantifying intrinsic data associations. Prominent network analysis methods furnish frameworks for model synthesis and validation but rely on standard correlation measures impaired by semi-supervised biases, latent heterogeneity, and uneven discretization techniques. Here we investigate a holistic measure for encapsulating data heterogeneity for enhanced efficacy of revealing complex patterns through network analysis. Our unique correlation metric, K-medoids Utility for Duo Original Similarities (Kudos), exhaustively factors real-valued analyte data to compute …
Benefits And Challenges Of Constructing Low-Altitude Air Route Network Infrastructure For Developing Low-Altitude Economy,
2024
State Key Laboratory of Resources and Environmental Information System, Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing 100101, China; Key Laboratory of Civil Aviation Low-Altitude Geographic Information and Route, Civil Aviation Administration of China, Beijing 100101, China
Benefits And Challenges Of Constructing Low-Altitude Air Route Network Infrastructure For Developing Low-Altitude Economy, Xiaohan Liao, Chenchen Xu, Huping Ye
Bulletin of Chinese Academy of Sciences (Chinese Version)
Low-altitude airspace is a resource that needs to be fully explored, and the low-altitude economy is a new type of economic activity resulted from low-altitude airspace exploration and utilization. As the primary players of low-altitude flight activities, the commercialization and wide applications of unmanned aerial vehicles (UAVs) are promoting the prosperity of the low-altitude economy. A low-altitude air route network is an effective means to ensure the safe and efficient operations of a large number of UAVs, and it is also a new key infrastructure hosting low-altitude traffic. The history shows that the investment on public transport infrastructure as a …
Promote Deep Integration Of Real Economy And Digital Economy,
2024
National E-Governance Expert Committee, Beijing 100035, China
Promote Deep Integration Of Real Economy And Digital Economy, Qinmin Wang
Bulletin of Chinese Academy of Sciences (Chinese Version)
Deep integration of digital technology, industrial development, and data resources empowers development of the digital economy, constantly opens up new tracks, creates new momentum, and establishes new advantages, becoming a new powerful engine for global economic and social development. China attaches great importance to the high-quality development of the digital economy. This study discusses the practical experience of vigorously promoting digital industrialization and industrial digitization, promoting the deep integration of digital technology and the real economy, and accelerating the construction of a network power and digital China. It also proposes countermeasures and suggestions to further promote the efficient empowerment of …
Reflections On Strengthening Digital Public Goods Institutions To Promote Construction Of Open Source Innovation Ecosystem,
2024
Institute of Strategic Studies, Shanghai Jiao Tong University, Shanghai 200052, China
Reflections On Strengthening Digital Public Goods Institutions To Promote Construction Of Open Source Innovation Ecosystem, Chao Zhang, Ze Feng, Kaihua Chen, Qigang Zhu
Bulletin of Chinese Academy of Sciences (Chinese Version)
The digital public goods institutions are important foundations for promoting the development of an open source innovation ecosystem, which helps to break the key limitations faced by China’s open source innovation ecosystem, such as the lack of unity in technology routes and concentration of R&D power. Based on discussing the significance of improving digital public goods institutions, this study reviews the experience of major countries in the world in building digital public goods institutions and analyzes the prominent problems faced by the construction of China’s digital public goods institutions. This study points out that China’s open source innovation ecosystem faces …
Prospect And Problem Analysis Of Industry Data Application In Livestock And Poultry Breeding,
2024
CAS Key Laboratory of Agro-ecological Processes in Subtropical Region, Institute of Subtropical Agriculture, Chinese Academy of Sciences, Changsha 410125, China; College of Mechanical and Electronic Engineering, Shandong Agricultural University, Tai’an 271018, China; Zhongke Jieyun (Beijing) Information Technology Co. Ltd., Beijing 101499, China
Prospect And Problem Analysis Of Industry Data Application In Livestock And Poultry Breeding, Yiran Chen, Zhuqing Xiong, Jiaogen Zhou, Quan Wang, Jiancheng Shu, Yinfa Yan, Lanlin Yang, Zemeng Feng, Benhai Xiong, Yulong Yin
Bulletin of Chinese Academy of Sciences (Chinese Version)
Livestock and poultry breeding is a pillar industry in China. The massive data in livestock and poultry breeding is a valuable resource. The market-oriented utilization of livestock and poultry breeding data plays an important role in improving industry standards, increasing industry profits, and driving the development of the entire industry chain. Currently, based on the demand for marketization of livestock and poultry breeding data, the application of new generation information technologies such as artificial intelligence and the Internet of Things in the process of livestock and poultry breeding to collect breeding process data, after de-sensitization and de-classification, through cloud computing, …
Status Quo Of Large-Scale Models, Risks And Challenges, And Recommended Countermeasures,
2024
Guanghua Law School, Zhejiang University, Hangzhou 310008, China
Status Quo Of Large-Scale Models, Risks And Challenges, And Recommended Countermeasures, Le Cheng, Yang Xiao
Bulletin of Chinese Academy of Sciences (Chinese Version)
Large-scale models (large models) are not only central to technological innovation, but also deeply entwined with national security, economic transformation, and social governance. This study examines the status quo of large-model development, identifies the key risks and challenges, and proposes response strategies, aiming to provide theoretical and policy insights for China’s navigations in global artificial intelligence (AI) competition and advances technological innovation. The research indicates that competition in the large-model market is fierce, while the industry is gradually consolidating. Competition in large models between China and the United States has escalated into a form of geopolitical contest. From a technical …
Construction Of Data Factor Circulation Law System In The United States And Its Reflection For China,
2024
Shanghai Data Exchange, Shanghai 201203,China
Construction Of Data Factor Circulation Law System In The United States And Its Reflection For China, Zihan Lin, Youmei Ma, Feng Guo
Bulletin of Chinese Academy of Sciences (Chinese Version)
As a pioneer country in the global digital economy, the United States has established a relatively complete system for the circulation of data elements. The focus is on government data openness, with the United States forming a legislative system for data openness at both federal and state levels to ensure data freedom and openness. In the field of personal data trading, the United States has established a relatively relaxed legal environment and data brokers to promote data circulation and trading. At present, China is actively promoting the market-oriented allocation reform of data elements, building a multi-level data element market, and …
Dynamic Knowledge Elicitation: Leveraging Student Feedback For Improved Language Model Distillation,
2024
Kennesaw State University
Dynamic Knowledge Elicitation: Leveraging Student Feedback For Improved Language Model Distillation, Reuven Muller
Master's Theses
Large Language Models (LLMs) have significantly advanced the field of natural language processing but remain resource-intensive and impractical for many organizations. Specialist models offer a viable alternative, often developed through Knowledge Distillation (KD) techniques. However, traditional KD methods rely on predefined static datasets to elicit knowledge from the teacher model, failing to dynamically address the weaknesses of the student model during training. This research introduces two novel methods for adaptive knowledge elicitation: Feedback-Driven Question Generation and Agent-Based Targeted Question Generation. These methods iteratively expand the training dataset based on the student model’s performance, leveraging a teacher model to generate targeted …
