{"id":{"repo_id":"usf","oai_identifier":"oai:digitalcommons.usf.edu:etd-12496"},"canonical_url":"https://search.ndltd.org/etd/usf/oai:digitalcommons.usf.edu:etd-12496","repository":{"repo_id":"usf","name":"University of South Florida","base_url":"https://digitalcommons.usf.edu/do/oai/"},"display":{"title":"AI-enabled Methodological Frameworks to Enhance the Robustness of Machine Learning Models in Various Domains and Dynamic Environments","abstract":"This dissertation explores novel approaches to address complex challenges in various domains, focusing on ensuring that machine learning (ML) models can reliably detect both the data they are trained on and unseen data. Reliable detection of unfamiliar inputs is critical for deploying models in real-world applications, where performance can degrade when faced with novel or unexpected information. Each domain addressed in this work introduces a set of contributions aimed at enhancing model performance, broadening generalization through diverse training data, and improving robustness against adversarial attacks. Three sets of contributions are presented, each encapsulated in a chapter. One of the contributions of this dissertation addresses foreign object debris (FOD) detection in airports. Existing FOD detection methods struggle to detect diverse types of debris in a timely manner. To address this, a novel object detection framework was developed using unmanned aerial systems (UAS), data augmentation techniques, and a state-of-the-art computer vision (CV) model (YOLOv4). By broadening the training data with diverse debris types and environmental conditions, this framework improves accuracy and inference speed, ensuring the timely removal of debris from runways and enhancing airport safety. Another contribution focuses on developing a mobile camera-based FOD detection system, designed to generalize detection across highly variable airport environments. This system leverages an open-world model with a YOLOv7 object detector, capable of handling both in-distribution (ID) and out-of-distribution (OOD) samples. By integrating a deep convolutional generative adversarial networks (GANs) to generate diverse synthetic data, the system adapts to changing conditions and achieves high precision while meeting Federal Aviation Administration (FAA) standards. Finally, this dissertation addresses the robustness of network intrusion detection system (NIDS) against adversarial attacks. A novel autoencoder-based generative framework is introduced to craft adversarial network packets that evade detection with minimal perturbation. This contribution highlights how adversaries can exploit model vulnerabilities by altering only small portions of data, underscoring the need for improved defenses in cybersecurity applications. In summary, this dissertation presents key contributions across different domains, demonstrating that reliable ID and OOD detection is essential for deploying models in dynamic environments. The proposed approaches—from airport safety to cybersecurity—enhance model performance, broaden generalization through diverse data sets, and protect against adversarial threats, ensuring robust and reliable performance in real-world applications.","abstract_html":"This dissertation explores novel approaches to address complex challenges in various domains, focusing on ensuring that machine learning (ML) models can reliably detect both the data they are trained on and unseen data. Reliable detection of unfamiliar inputs is critical for deploying models in real-world applications, where performance can degrade when faced with novel or unexpected information. Each domain addressed in this work introduces a set of contributions aimed at enhancing model performance, broadening generalization through diverse training data, and improving robustness against adversarial attacks. Three sets of contributions are presented, each encapsulated in a chapter. One of the contributions of this dissertation addresses foreign object debris (FOD) detection in airports. Existing FOD detection methods struggle to detect diverse types of debris in a timely manner. To address this, a novel object detection framework was developed using unmanned aerial systems (UAS), data augmentation techniques, and a state-of-the-art computer vision (CV) model (YOLOv4). By broadening the training data with diverse debris types and environmental conditions, this framework improves accuracy and inference speed, ensuring the timely removal of debris from runways and enhancing airport safety. Another contribution focuses on developing a mobile camera-based FOD detection system, designed to generalize detection across highly variable airport environments. This system leverages an open-world model with a YOLOv7 object detector, capable of handling both in-distribution (ID) and out-of-distribution (OOD) samples. By integrating a deep convolutional generative adversarial networks (GANs) to generate diverse synthetic data, the system adapts to changing conditions and achieves high precision while meeting Federal Aviation Administration (FAA) standards. Finally, this dissertation addresses the robustness of network intrusion detection system (NIDS) against adversarial attacks. A novel autoencoder-based generative framework is introduced to craft adversarial network packets that evade detection with minimal perturbation. This contribution highlights how adversaries can exploit model vulnerabilities by altering only small portions of data, underscoring the need for improved defenses in cybersecurity applications. In summary, this dissertation presents key contributions across different domains, demonstrating that reliable ID and OOD detection is essential for deploying models in dynamic environments. The proposed approaches—from airport safety to cybersecurity—enhance model performance, broaden generalization through diverse data sets, and protect against adversarial threats, ensuring robust and reliable performance in real-world applications.","abstract_has_math":false,"creators":["Noroozi, Mohammad"],"institution":"University of South Florida","degree_name":"Doctor of Philosophy (Ph.D.)","degree_level":"Dissertation","degree_discipline":"Industrial and Management Systems Engineering","degree_department":null,"school":null,"contributors":["Ankit Shah, Ph.D.","Tapas K. Das, Ph.D.","Susana K. Lai-Yuen, Ph.D.","Nasir Ghani, Ph.D.","Balaji Padmanabhan, Ph.D."],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2024,"date_issued":"2024-11-06T08:00:00Z","date_published":"2024-11-06T08:00:00Z","updated_at":"2026-11-19T08:51:46Z","subjects":["Adversarial Learning","Computer Vision","Deep Learning","Model Reliability","Open World Detection","Artificial Intelligence and Robotics","Engineering"],"languages":[],"rights":[],"rights_urls":[],"identifier_entries":[],"person_identifier_entries":[]},"links":{"outbound_url":"https://digitalcommons.usf.edu/etd/11193","outbound_label":"Repository record","outbound_source":"dc:identifier","record_url":"https://digitalcommons.usf.edu/etd/11193","record_label":"Repository record","record_kind":"record","record_source":"dc:identifier","download_url":null,"download_label":null,"download_source":null,"access_status":null,"embargo_until":null,"access_note":null},"source_record":{"url":"https://digitalcommons.usf.edu/do/oai/?verb=GetRecord&metadataPrefix=oai_etdms&identifier=oai%3Adigitalcommons.usf.edu%3Aetd-12496","prefix":"oai_etdms"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Ankit Shah, Ph.D.","Tapas K. Das, Ph.D.","Susana K. Lai-Yuen, Ph.D.","Nasir Ghani, Ph.D.","Balaji Padmanabhan, Ph.D."]},{"key":"dc:creator","label":"Author","values":["Noroozi, Mohammad"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2024-11-06T08:00:00Z"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["Industrial and Management Systems Engineering"]},{"key":"thesis:degree_level","label":"Degree Level","values":["Dissertation","Ph.D."]},{"key":"thesis:degree_name","label":"Degree Name","values":["Doctor of Philosophy (Ph.D.)"]},{"key":"thesis:institution_name","label":"Thesis Institution Name","values":["University of South Florida"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Adversarial Learning","Computer Vision","Deep Learning","Model Reliability","Open World Detection","Artificial Intelligence and Robotics","Engineering"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["https://digitalcommons.usf.edu/etd/11193"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["This dissertation explores novel approaches to address complex challenges in various domains, focusing on ensuring that machine learning (ML) models can reliably detect both the data they are trained on and unseen data. Reliable detection of unfamiliar inputs is critical for deploying models in real-world applications, where performance can degrade when faced with novel or unexpected information. Each domain addressed in this work introduces a set of contributions aimed at enhancing model performance, broadening generalization through diverse training data, and improving robustness against adversarial attacks. Three sets of contributions are presented, each encapsulated in a chapter. One of the contributions of this dissertation addresses foreign object debris (FOD) detection in airports. Existing FOD detection methods struggle to detect diverse types of debris in a timely manner. To address this, a novel object detection framework was developed using unmanned aerial systems (UAS), data augmentation techniques, and a state-of-the-art computer vision (CV) model (YOLOv4). By broadening the training data with diverse debris types and environmental conditions, this framework improves accuracy and inference speed, ensuring the timely removal of debris from runways and enhancing airport safety. Another contribution focuses on developing a mobile camera-based FOD detection system, designed to generalize detection across highly variable airport environments. This system leverages an open-world model with a YOLOv7 object detector, capable of handling both in-distribution (ID) and out-of-distribution (OOD) samples. By integrating a deep convolutional generative adversarial networks (GANs) to generate diverse synthetic data, the system adapts to changing conditions and achieves high precision while meeting Federal Aviation Administration (FAA) standards. Finally, this dissertation addresses the robustness of network intrusion detection system (NIDS) against adversarial attacks. A novel autoencoder-based generative framework is introduced to craft adversarial network packets that evade detection with minimal perturbation. This contribution highlights how adversaries can exploit model vulnerabilities by altering only small portions of data, underscoring the need for improved defenses in cybersecurity applications. In summary, this dissertation presents key contributions across different domains, demonstrating that reliable ID and OOD detection is essential for deploying models in dynamic environments. The proposed approaches—from airport safety to cybersecurity—enhance model performance, broaden generalization through diverse data sets, and protect against adversarial threats, ensuring robust and reliable performance in real-world applications."]},{"key":"dc:title","label":"Title","values":["AI-enabled Methodological Frameworks to Enhance the Robustness of Machine Learning Models in Various Domains and Dynamic Environments"]}]}],"canonical_facts":{"dc:contributor":["Ankit Shah, Ph.D.","Tapas K. Das, Ph.D.","Susana K. Lai-Yuen, Ph.D.","Nasir Ghani, Ph.D.","Balaji Padmanabhan, Ph.D."],"dc:creator":["Noroozi, Mohammad"],"dc:date":["2024-11-06T08:00:00Z"],"dc:description":["This dissertation explores novel approaches to address complex challenges in various domains, focusing on ensuring that machine learning (ML) models can reliably detect both the data they are trained on and unseen data. Reliable detection of unfamiliar inputs is critical for deploying models in real-world applications, where performance can degrade when faced with novel or unexpected information. Each domain addressed in this work introduces a set of contributions aimed at enhancing model performance, broadening generalization through diverse training data, and improving robustness against adversarial attacks. Three sets of contributions are presented, each encapsulated in a chapter. One of the contributions of this dissertation addresses foreign object debris (FOD) detection in airports. Existing FOD detection methods struggle to detect diverse types of debris in a timely manner. To address this, a novel object detection framework was developed using unmanned aerial systems (UAS), data augmentation techniques, and a state-of-the-art computer vision (CV) model (YOLOv4). By broadening the training data with diverse debris types and environmental conditions, this framework improves accuracy and inference speed, ensuring the timely removal of debris from runways and enhancing airport safety. Another contribution focuses on developing a mobile camera-based FOD detection system, designed to generalize detection across highly variable airport environments. This system leverages an open-world model with a YOLOv7 object detector, capable of handling both in-distribution (ID) and out-of-distribution (OOD) samples. By integrating a deep convolutional generative adversarial networks (GANs) to generate diverse synthetic data, the system adapts to changing conditions and achieves high precision while meeting Federal Aviation Administration (FAA) standards. Finally, this dissertation addresses the robustness of network intrusion detection system (NIDS) against adversarial attacks. A novel autoencoder-based generative framework is introduced to craft adversarial network packets that evade detection with minimal perturbation. This contribution highlights how adversaries can exploit model vulnerabilities by altering only small portions of data, underscoring the need for improved defenses in cybersecurity applications. In summary, this dissertation presents key contributions across different domains, demonstrating that reliable ID and OOD detection is essential for deploying models in dynamic environments. The proposed approaches—from airport safety to cybersecurity—enhance model performance, broaden generalization through diverse data sets, and protect against adversarial threats, ensuring robust and reliable performance in real-world applications."],"dc:identifier":["https://digitalcommons.usf.edu/etd/11193"],"dc:subject":["Adversarial Learning","Computer Vision","Deep Learning","Model Reliability","Open World Detection","Artificial Intelligence and Robotics","Engineering"],"dc:title":["AI-enabled Methodological Frameworks to Enhance the Robustness of Machine Learning Models in Various Domains and Dynamic Environments"],"thesis:degree_discipline":["Industrial and Management Systems Engineering"],"thesis:degree_level":["Dissertation","Ph.D."],"thesis:degree_name":["Doctor of Philosophy (Ph.D.)"],"thesis:institution_name":["University of South Florida"]},"updated_at":"2026-11-19T08:51:46Z"}