21 Jun 2024 | Penghao Zhao*, Hailin Zhang*, Qinhan Yu, Zhengren Wang, Yunteng Geng, Fangcheng Fu†, Ling Yang, Wentao Zhang†, Jie Jiang, Bin Cui†
This paper presents a comprehensive survey of Retrieval-Augmented Generation (RAG) for AI-Generated Content (AIGC). RAG is a paradigm that integrates information retrieval with generative models to enhance the accuracy and robustness of AI-generated content. The paper reviews existing efforts that apply RAG techniques in AIGC scenarios, classifying RAG foundations based on how the retriever augments the generator. It discusses various RAG foundations, including query-based, latent representation-based, logit-based, and speculative RAG, and explores their applications across different modalities and tasks. The paper also summarizes additional enhancement methods for RAG, facilitating effective engineering and implementation of RAG systems. It surveys practical applications of RAG across different modalities and tasks, offering valuable references for researchers and practitioners. Furthermore, the paper introduces benchmarks for RAG, discusses the limitations of current RAG systems, and suggests potential directions for future research. The paper also provides an overview of RAG foundations, including retrievers and generators, and explores various enhancement methods that further improve the effectiveness of RAG systems. The paper concludes with a discussion of the current limitations of RAG and potential future directions for research.This paper presents a comprehensive survey of Retrieval-Augmented Generation (RAG) for AI-Generated Content (AIGC). RAG is a paradigm that integrates information retrieval with generative models to enhance the accuracy and robustness of AI-generated content. The paper reviews existing efforts that apply RAG techniques in AIGC scenarios, classifying RAG foundations based on how the retriever augments the generator. It discusses various RAG foundations, including query-based, latent representation-based, logit-based, and speculative RAG, and explores their applications across different modalities and tasks. The paper also summarizes additional enhancement methods for RAG, facilitating effective engineering and implementation of RAG systems. It surveys practical applications of RAG across different modalities and tasks, offering valuable references for researchers and practitioners. Furthermore, the paper introduces benchmarks for RAG, discusses the limitations of current RAG systems, and suggests potential directions for future research. The paper also provides an overview of RAG foundations, including retrievers and generators, and explores various enhancement methods that further improve the effectiveness of RAG systems. The paper concludes with a discussion of the current limitations of RAG and potential future directions for research.