Date of Publication
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
Robert Billones
Richard TanAi
Abstract/Summary
Artificial intelligence has grown in popularity over recent years and is used in many applications. Deep learning is a type of artificial intelligence that can learn complex representations from data, giving them the ability to obtain human-like accuracy. However, this requires deep learning models to be trained with a large amount of labeled data. A recent trend in deep learning is that models are increasing in size and complexity, which is accompanied by growth in the sizes of labeled training datasets. Acquiring labeled data, or data annotation, can already be a problem for deep learning projects. The need for larger datasets intensifies this problem.
Active learning is a machine learning training method that can reduce data annotation requirements by selecting and using only the most important training data. This method can be beneficial for reducing the data annotation requirements for deep learning projects by reducing the data needed to train a model. However, active learning aims to reduce the training data, whereas deep learning benefits from more training data. This contradiction can lead to poor model performance which can negate the benefits of active learning for deep learning. This issue of poor model performance is known as the cold-start problem, and is especially prevalent in the early stages of training. Curriculum learning is another machine learning training method that promises to improve the speed of a model's training by introducing training samples from easy-to-difficult. This study aims to address this issue by proposing a novel curriculum-based active learning framework to improve model performance in the earlier stages of training for an image classification task.
Abstract Format
html
Language
English
Format
Electronic
Keywords
Machine learning; Supervised learning (Machine learning)
Recommended Citation
Keh, J. U. (2024). Curriculum active learning for image classification. Retrieved from https://animorepository.dlsu.edu.ph/etdm_ece/34
Upload Full Text
wf_yes
Embargo Period
4-23-2026
