Date of Publication
4-23-2024
Document Type
Master's Thesis
Degree Name
Master of Science in Electronics and Communications Engineering
Subject Categories
Electrical and Computer Engineering | Engineering
College
Gokongwei College of Engineering
Department/Unit
Electronics And Communications Engg
Thesis Advisor
John Anthony C. Jose
Defense Panel Chair
Edwin Sybingco
Defense Panel Member
Melvin Cabatuan
Richard TanAi
Abstract/Summary
In the field of computer vision, training accurate deep learning models necessitates a substantial number of manually annotated images, often posing financial and practical challenges, particularly in domains like medicine and engineering.
This research builds upon existing literature in computer vision and active learning, identifying gaps in current approaches and proposing an innovative strategy that combines diversity-based active learning sampling with feature mixing. This novel approach aims to select unlabeled samples that exhibit both diversity and informativeness, thereby enhancing model performance while managing annotation budgets effectively.
However, this thesis recognizes the inherent challenges in classifying high-dimensional data with limited samples. By bridging the gap between diversity-based active learning and feature mixing, this research contributes to advancing the state of the art in computer vision and holds the potential to impact not only the academic community but also real- world applications in medical imaging, engineering, and related fields. It addresses the pressing need for more efficient and cost-effective deep learning model training in domains where expert annotation is crucial.
Abstract Format
html
Language
English
Format
Electronic
Keywords
Neural networks (Computer science); Machine learning
Recommended Citation
Tiberio, J. L. (2024). Adversarial active learning with feature mixing. Retrieved from https://animorepository.dlsu.edu.ph/etdm_ece/33
Upload Full Text
wf_yes
Embargo Period
4-23-2026
