22505 research outputs found
Sort by
Uncertainty-Aware Meta-Learning for Learning from Limited Data
Get PDFDeep Learning (DL) models have achieved great success in large data fields ranging from computer vision and natural language processing to digital arts and robotics. However, the effectiveness of the DL models is challenged by many real-world limited data problems (e.g., medicine, healthcare, and security intelligence) where data for model training is scarce. Unlike DL models, humans can use the prior knowledge stored in their brains to quickly learn new tasks with limited data. Inspired by such human learning, various meta-learning models have been developed that aim to address the challenge of learning from limited data. However, existing models are computationally expensive, lack fine-grained uncertainty-quantification capabilities, and the predictions are not always trustworthy. The dissertation focuses on different instances of the two most popular limited data problems: few-shot regression and few-shot classification. For both problems, the developed models need to be robust, and output well-calibrated trustworthy predictions while remaining computationally cheap and label-efficient to ensure real-world applicability. In this dissertation, we develop a novel uncertainty-aware meta-learning framework based on evidential deep learning that contributes towards developing a reliable model that can address the above challenges. We first introduce the evidential multidimensional belief theory for meta-learning that leads to computationally-efficient uncertainty-aware few-shot classification models. We then extend the evidential regression theory to meta-learning models that leads to computationally-efficient uncertainty-aware outlier-robust few-shot regression models. We then carry out a thorough analysis of the evidential deep learning framework to identify fundamental learning deficiency that helps explain the suboptimal performance, especially in challenging settings. We then develop theoretically justified, empirically validated solution to address the fundamental learning deficiency of the evidential models. Improving on the developed theory, we introduce the Bayesian-evidential framework for parameter-efficient-fine-tuning of vision foundation models that leads to well-calibrated uncertainty-aware few-shot learning models. We then study the adversarial robustness of the developed uncertainty-aware models. We also explore applications of the ideas developed in this dissertation to real-world problems of healthcare and high-density-energy physics. The theoretically grounded, empirically justified solutions of the uncertainty-aware meta-learning framework developed in this dissertation contribute towards development of trustworthy uncertainty-aware models that are capable of effectively learning from limited data
Sales Opportunities Lead Qualification in B2B Market
Get PDFThis thesis addresses the challenge of inefficient lead qualification in the business-to-business (B2B) market by applying machine learning techniques to predict the likelihood of winning a sales opportunity. Using a real-world dataset of over 78,000 records and 17 variables, the study aims to improve how sales teams identify and prioritize high-conversion leads. A thorough data preparation process was conducted, including handling of missing values, outlier detection, and under-sampling to resolve class imbalance between won and lost opportunities. After thorough data cleaning, preprocessing, and under-sampling to address class imbalance, five machine learning models were developed: Logistic Regression, Random Forest, Neural Network, Linear SVM, and XGBoost Tree. Each model was evaluated using metrics such as accuracy, precision, recall, F1-score, and AUC. Among them, XGBoost outperformed all others, achieving an accuracy of 95.57%, precision of 98.3%, recall of 95.9%, and an AUC score of 0.993, indicating its strong ability to differentiate between won and lost opportunities. Feature importance analysis using the F-score method revealed that Sales Velocity, Qualification Board Score, and Opportunity Size were among the top predictors of opportunity success. Statistical tests including Chi-square, Mann-Whitney U, and Kruskal-Wallis supported these findings by highlighting significant relationships between key variables and opportunity status. This study demonstrates that machine learning can enhance B2B lead qualification by providing data-driven insights into what drives successful outcomes. By moving beyond manual scoring and intuition, organizations can improve lead targeting, increase sales efficiency, and better allocate their resources. The research offers a practical framework for integrating AI into the sales decision-making process and contributes to the growing exploration of machine learning in B2B environments
Predicting the Probability of Crime Related Danger in Los Angeles
Get PDFThis thesis examines the use of advanced machine learning to predict crime danger in Los Angeles, where 2023 violent crime rates (503 per 100,000) surpass the national average (363.8 per 100,000). Rooted in theories like social disorganization and victim vulnerability, it addresses the lack of combined victim-centric modeling by focusing on three questions: (1) Do blended ensemble models outperform individual models in predicting crime danger? (2) Can unsupervised learning techniques enhance supervised models’ accuracy through label generation or augmentation? (3) How can we interpret accurate machine learning models\u27 decision-making processes? Using historical crime data from Los Angeles (2020–2025) and various environmental and demographic variables, this mixed-methods approach includes approximately 1,000,000 crime incidents. Blended models such as Random Forest, Gradient Boosting, XGBoost, and Multilayer Perceptron were tested against individual models. Clustering methods (K-means and DB Scan) identified crime patterns and improved label quality. SHAP analyses explained model decisions. Findings show blended models have slightly higher predictive accuracy of 68% compared to individual ones (67%). However, XGBoost algorithm outperformed the blended model with an accuracy of 69%. Clustering improved model performance by 3–5%. Socioeconomic factors and temporal patterns were significant influencers, offering insights for both individuals and law enforcement. In summary, blended ensemble algorithms may fail to outperform standard models, but unsupervised learning enhance crime prediction with a victim-centric approach, though ethical concerns about biases remain. Recommendations include using these models for targeted resource allocation in Los Angeles. Future research should extend to other cities, use real-time data, and incorporate fairness-aware algorithms to address ethical issues
Methods of Quantum Circuit Simulation: A Comprehensive Comparative Analysis
Get PDFThe advent of quantum computing has ushered in a new era of computational capabilities, promising to solve problems that were previously intractable for classical computers. Quantum circuit simulation is an essential aspect of harnessing the potential of quantum algorithms, and understanding quantum systems. This thesis presents an exploration and comparative analysis of various methods used in the simulation of quantum circuits, aimed at providing a comprehensive understanding of their strengths, weaknesses, and practical applications. This starts with a dive into the foundational concepts of quantum circuits, quantum gates, and the fundamental principles underlying quantum computation. Quantum circuit simulation emphasizes the growing significance of quantum technologies, as they become more accessible, and play a pivotal role in quantum algorithm development. To this end, a rigorous analysis of simulation methods, including state vector, matrix product state, density matrix is conducted. Gaps in each method are identified, with potential solutions offered. With the limitations of the full suite of simulators gleaned from these experiments, a more robust understanding of the proper usage of each can be inferred and used in future experiments and analysis
Creative Writing and the Common Core
Get PDFReview of Imaginative Teaching Through Creative Writing: A Guide for Secondary Classrooms. Edited by Amy Ash, Michael Dean Clark, and Chris Drew. Bloomsbury, 2021
Those Who Did Teach
Get PDFReview of Writers and Their Teachers. Ed. Dale Salwak. Bloomsbury, 2023. 137 pages
Exploring Rural vs. Urban Variations in EMT Scope of Practice: Advanced Airway and IV Access Policies
Get PDFThis thesis investigates how states across the United States (US) modify the Emergency Medical Technician (EMT) Scope of Practice (SoP) to permit advanced skills, with a focus on advanced airway management via supraglottic airways (SGAs) and intravenous (IV) access procedures, and whether rurality plays a role in these decisions. While the National EMS SoP Model establishes minimum competencies for each emergency medical services (EMS) provider level, individual states often alter their standards in response to local demographic, geographic, and resource-based needs. This study reviewed SoP policies from all 50 states, and 44 states were included. Logistic regression analysis was utilized to explore relationships between advanced skill allowances and predictors such as percent rural population, state EMS funding, population density, and geographic size. This analysis aimed to assess whether states with a higher percent rural population were more likely to permit EMTs to perform SGA and IV access skills. Although statistical models did not yield significant results due to sample size limitations, descriptive trends suggested that states allowing SGA placement at the EMT level tended to have higher rural populations and increased EMS funding. Conversely, IV access at the EMT level was permitted more often in less rural, lower-density states with fewer EMS agencies. This research highlights the complexities of SoP modifications along with the critical role of EMTs in resource-limited areas and the ongoing debate over expanding their clinical responsibilities. Findings underscore the need for improved standardization in reporting SoPs, greater transparency in policy rationale, and more localized or regional analyses to capture nuanced EMS operational needs
Deepfake Image Detection Using Explainable Ai And Deep Learning
Get PDFDeepfake technology has grown exponentially, and it has changed our perspective of managing digital content. The technology has many applications, but its ease of access has brought many risks. These risks include identity theft, misinformation spread, development of non-consensual content, and a decrease in trust in media. To address these challenges, we propose a hybrid deep learning model combining Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks for deepfake image detection. We developed a CNN model and four hybrid CNN-LSTM models in our study. Our approach integrates the strengths of CNNs for spatial feature extraction and LSTMs for pattern recognition in features, offering a robust and adaptable detection mechanism. The results show that the CNN-LSTM hybrid model performs very well in detecting deepfake images with high accuracy. In comparison to existing works, our model demonstrates superior accuracy and robustness. Unlike other studies, we validated the performance of all models using k-fold cross validation and incorporated SHAP analysis to provide interpretability, identifying critical image features contributing to predictions. This transparency enhances the model\u27s reliability and usability in real-world applications. Our findings highlight the flexibility and effectiveness of the hybrid CNN-LSTM model, which outperforms traditional approaches in detecting deepfakes. By addressing various deepfake techniques and ensuring adaptability to evolving manipulations, the proposed model represents a significant advancement in cybersecurity and the fight against malicious use of AI technologies
Domain-Specific Customization for Improving Speech to Text
Get PDFThe advent of transformer-based models has revolutionized natural language processing, bringing remarkable improvements in tasks like automatic speech recognition (ASR). Inspired by these advancements, this thesis explores the optimization of a transformer-based ASR model to improve transcription accuracy in educational settings, particularly for lecture content. The goal of this research is to provide real-time, high-accuracy captions that enhance accessibility for all students, while offering a cost-effective solution for educators. To assess the potential of domain-specific fine-tuning, Whisper-small underwent two phases of fine-tuning. In the first phase, it was finetuned on care- fully selected, publicly available datasets: SpeechColab’s Gigaspeech-XS [39], AMI Meeting corpus [14]. In the second phase, fine-tuned model was optimized on a self-curated dataset [16] consisting of roughly 10 hours of live lecture recordings collected and assembled by me. Finally, a real-time captioning assistant application was developed to leverage the finetuned model and transcribe speech in real time with live editing capabilities. The optimized Whisper-small model was evaluated against Whisper’s retrained small, medium and large(version 2) counterparts. The evaluation was performed on a clean unseen data [15] prepared by me. The fine-tuned model achieved lower Word Error Rates (WER) of 4.53%, compared to 5.51% and 5.78% for Whisper-Medium and Whisper-Large-V2 respectively. These results demonstrate that fine-tuning a transformer-based ASR model on domain- specific data can significantly enhance its performance in a targeted context, such as live lecture transcription. The findings of this experiment highlight the promise of transformer-based models for improving educational accessibility. From thereon, building an application tailored to live lecture settings, this research contributes to the development of adaptable, low-cost technologies that support inclusive learning environments. The success of this experiment lays the groundwork for future breakthroughs in speech recognition, aiming to make education more accessible for everyone