# Data Augmentation using LLMs: Data Perspectives, Learning Paradigms and Challenges

**Authors:** Bosheng Ding, Chengwei Qin, Ruochen Zhao, Tianze Luo, Xinze Li, Guizhen Chen, Wenhan Xia, Junjie Hu, Anh Tuan Luu, Shafiq Joty  
**arXiv:** [2403.02990](https://arxiv.org/abs/2403.02990)  
**Date:** 2024-03-05  
**Summary page:** https://www.summarizepaper.com/en/paper/2403.02990v1/

## Key points

- Data augmentation is a crucial technique in machine learning for improving model performance without additional data collection
- Large Language Models (LLMs) have a significant impact on data augmentation, particularly in natural language processing (NLP) and beyond
- Various strategies utilize LLMs for data augmentation, including using LLM-generated data for further training
- Key challenges in this area include controllable data augmentation and multi-modal data augmentation
- LLMs have brought about a paradigm shift in data augmentation, providing valuable guidance for researchers and practitioners
- The authors provide insightful perspectives on how LLMs are reshaping the landscape of data augmentation and influencing future advancements in ML

## Summary

Data augmentation has become a crucial technique in machine learning (ML) for improving model performance without the need for additional data collection. This survey explores the significant impact of Large Language Models (LLMs) on data augmentation, specifically in natural language processing (NLP) and beyond. Through both data and learning perspectives, the study delves into various strategies that utilize LLMs for data augmentation, including innovative approaches where LLM-generated data is used for further training. The paper also addresses key challenges in this area such as controllable data augmentation and multi-modal data augmentation. By highlighting the paradigm shift brought about by LLMs in data augmentation, this survey serves as a valuable guide for researchers and practitioners looking to effectively leverage these advanced models. The authors—Bosheng Ding, Chengwei Qin, Ruochen Zhao, Tianze Luo, Xinze Li, Guizhen Chen, Wenhan Xia, Junjie Hu, Anh Tuan Luu, and Shafiq Joty—provide insightful perspectives on how LLMs are reshaping the landscape of data augmentation and shaping future advancements in ML.