{"id":{"repo_id":"baylor","oai_identifier":"oai:baylor-ir.tdl.org:2104/13984"},"canonical_url":"https://search.dev.ndltd.org/etd/baylor/oai:baylor-ir.tdl.org:2104/13984","repository":{"repo_id":"baylor","name":"Baylor University","base_url":"https://baylor-ir.tdl.org/server/oai/request"},"display":{"title":"Deep learning of visual features with limited supervision.","abstract":"Recent advances in deep learning have led to significant improvements in various computer vision tasks. Typically, deep learning models require large-scale labeled data for training, but obtaining annotations is costly in fields like medical imaging and underwater imaging. This dissertation explores methods for learning deep visual features with limited human supervision, expanding deep learning’s applicability to diverse real-world tasks. We propose solutions in three key areas. First, we introduce a selective pretraining approach that enhances transfer learning by selecting pre-training samples more relevant to the target domain. Second, we design a self-supervised learning algorithm that can leverage large amounts of unlabeled image and video data. Third, we develop an active learning (AL) algorithm that iteratively selects the most informative samples for annotation to optimize learning with minimal supervision. On selective pre-training, we achieve 77% accuracy on imbalanced CIFAR-10 using only 500k pre-training samples—outperforming full ImageNet pre-training. On Penn Action video classification, our contrastive learning model reaches 76% accuracy, significantly surpassing ViViT model. In active learning on aquatic invasive species data, we achieve 78% balanced accuracy with just 200 labeled samples, improving 27% over random sampling. Together, selective pre-training, self-supervised learning, and active learning provide a flexible framework to approach deep visual feature learning depending on the availability of labeled data, unlabeled data and computational resources.","abstract_html":"Recent advances in deep learning have led to significant improvements in various computer vision tasks. Typically, deep learning models require large-scale labeled data for training, but obtaining annotations is costly in fields like medical imaging and underwater imaging. This dissertation explores methods for learning deep visual features with limited human supervision, expanding deep learning’s applicability to diverse real-world tasks. We propose solutions in three key areas. First, we introduce a selective pretraining approach that enhances transfer learning by selecting pre-training samples more relevant to the target domain. Second, we design a self-supervised learning algorithm that can leverage large amounts of unlabeled image and video data. Third, we develop an active learning (AL) algorithm that iteratively selects the most informative samples for annotation to optimize learning with minimal supervision. On selective pre-training, we achieve 77% accuracy on imbalanced CIFAR-10 using only 500k pre-training samples—outperforming full ImageNet pre-training. On Penn Action video classification, our contrastive learning model reaches 76% accuracy, significantly surpassing ViViT model. In active learning on aquatic invasive species data, we achieve 78% balanced accuracy with just 200 labeled samples, improving 27% over random sampling. Together, selective pre-training, self-supervised learning, and active learning provide a flexible framework to approach deep visual feature learning depending on the availability of labeled data, unlabeled data and computational resources.","abstract_has_math":false,"creators":["Chowdhury, Shaif, 1996-"],"institution":"Baylor University.","degree_name":"Ph.D.","degree_level":"Doctoral","degree_discipline":null,"degree_department":null,"school":null,"contributors":[],"advisors":["Hamerly, Gregory James, 1977-"],"committee_chairs":[],"committee_members":[],"year":2025,"date_issued":"2025-08","date_published":"2025-08","updated_at":"2026-07-24T01:08:00Z","subjects":["Deep learning.","Transfer learning.","Self-supervised learning.","Active learning.","Limited supervision."],"languages":["en"],"rights":["Baylor University works are protected by copyright. They may be viewed from this source for any purpose, but reproduction or distribution in any format is prohibited without written permission. Contact libraryquestions@baylor.edu for inquiries about permission."],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":"https://hdl.handle.net/2104/13984","outbound_label":"Handle","outbound_source":"dc:identifier.uri"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Hamerly, Gregory James, 1977-"]},{"key":"dc:creator","label":"Author","values":["Chowdhury, Shaif, 1996-"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.accessioned","label":"Dc Date Accessioned","values":["2025-09-19T20:23:30Z"]},{"key":"dc:date.issued","label":"Date","values":["2025-08"]},{"key":"dc:type","label":"Dc Type","values":["Thesis"]},{"key":"thesis:degree_level","label":"Degree Level","values":["Doctoral"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Ph.D."]},{"key":"thesis:institution_name","label":"Thesis Institution Name","values":["Baylor University."]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Deep learning.","Transfer learning.","Self-supervised learning.","Active learning.","Limited supervision."]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language.iso","label":"Language (ISO)","values":["en"]},{"key":"dc:rights","label":"Dc Rights","values":["Baylor University works are protected by copyright. They may be viewed from this source for any purpose, but reproduction or distribution in any format is prohibited without written permission. Contact libraryquestions@baylor.edu for inquiries about permission."]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://hdl.handle.net/2104/13984"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Recent advances in deep learning have led to significant improvements in various computer vision tasks. Typically, deep learning models require large-scale labeled data for training, but obtaining annotations is costly in fields like medical imaging and underwater imaging. This dissertation explores methods for learning deep visual features with limited human supervision, expanding deep learning’s applicability to diverse real-world tasks. We propose solutions in three key areas. First, we introduce a selective pretraining approach that enhances transfer learning by selecting pre-training samples more relevant to the target domain. Second, we design a self-supervised learning algorithm that can leverage large amounts of unlabeled image and video data. Third, we develop an active learning (AL) algorithm that iteratively selects the most informative samples for annotation to optimize learning with minimal supervision. On selective pre-training, we achieve 77% accuracy on imbalanced CIFAR-10 using only 500k pre-training samples—outperforming full ImageNet pre-training. On Penn Action video classification, our contrastive learning model reaches 76% accuracy, significantly surpassing ViViT model. In active learning on aquatic invasive species data, we achieve 78% balanced accuracy with just 200 labeled samples, improving 27% over random sampling. Together, selective pre-training, self-supervised learning, and active learning provide a flexible framework to approach deep visual feature learning depending on the availability of labeled data, unlabeled data and computational resources."]},{"key":"dc:format.mimetype","label":"Dc Format Mimetype","values":["application/pdf"]},{"key":"dc:title","label":"Title","values":["Deep learning of visual features with limited supervision."]}]}],"canonical_facts":{"dc:contributor.advisor":["Hamerly, Gregory James, 1977-"],"dc:creator":["Chowdhury, Shaif, 1996-"],"dc:date.accessioned":["2025-09-19T20:23:30Z"],"dc:date.issued":["2025-08"],"dc:description.abstract":["Recent advances in deep learning have led to significant improvements in various computer vision tasks. Typically, deep learning models require large-scale labeled data for training, but obtaining annotations is costly in fields like medical imaging and underwater imaging. This dissertation explores methods for learning deep visual features with limited human supervision, expanding deep learning’s applicability to diverse real-world tasks. We propose solutions in three key areas. First, we introduce a selective pretraining approach that enhances transfer learning by selecting pre-training samples more relevant to the target domain. Second, we design a self-supervised learning algorithm that can leverage large amounts of unlabeled image and video data. Third, we develop an active learning (AL) algorithm that iteratively selects the most informative samples for annotation to optimize learning with minimal supervision. On selective pre-training, we achieve 77% accuracy on imbalanced CIFAR-10 using only 500k pre-training samples—outperforming full ImageNet pre-training. On Penn Action video classification, our contrastive learning model reaches 76% accuracy, significantly surpassing ViViT model. In active learning on aquatic invasive species data, we achieve 78% balanced accuracy with just 200 labeled samples, improving 27% over random sampling. Together, selective pre-training, self-supervised learning, and active learning provide a flexible framework to approach deep visual feature learning depending on the availability of labeled data, unlabeled data and computational resources."],"dc:format.mimetype":["application/pdf"],"dc:identifier.uri":["https://hdl.handle.net/2104/13984"],"dc:language.iso":["en"],"dc:rights":["Baylor University works are protected by copyright. They may be viewed from this source for any purpose, but reproduction or distribution in any format is prohibited without written permission. Contact libraryquestions@baylor.edu for inquiries about permission."],"dc:subject":["Deep learning.","Transfer learning.","Self-supervised learning.","Active learning.","Limited supervision."],"dc:title":["Deep learning of visual features with limited supervision."],"dc:type":["Thesis"],"thesis:degree_level":["Doctoral"],"thesis:degree_name":["Ph.D."],"thesis:institution_name":["Baylor University."]},"updated_at":"2026-07-24T01:08:00Z"}