Knowledge and data integration for deep learning under small data

• Main Author: Teo, Hazel Kai Xin
• Other Authors: Mao Kezhi
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Engineering; Synoynm replacement
• Online Access: https://hdl.handle.net/10356/176309
• Institution: Nanyang Technological University

This research addresses limited training data in deep learning, where data volume, quality, and diversity significantly influence model performance. The availability of diverse and abundant data is crucial for effective training models. However, in many real-world scenarios, obtaining such varied data can be challenging, potentially leading to biased models, particularly affecting minority classes. Recent literature and research by various scholars emphasize data augmentation techniques as a promising solution to mitigate data scarcity and enhance model accuracy without exhaustive labeling efforts. This study explores the potential of data augmentation, particularly text augmentation, in alleviating the dependency on extensive training data. The aim is to enhance the effectiveness and accuracy of deep learning models, especially in the context of natural language processing (NLP). We investigate the benefits of employing synonym replacement as a primary text augmentation technique, assessing its ability to generate supplementary data and improve model performance.