RbFT: Robust Fine-tuning for Retrieval-Augmented Generation against Retrieval Defects

2025/01/30 by Yiteng Tu, Y. L. Tu, Weihang Su +8 · 1 voice · 13 citations
Computer Science · Engineering · #Advanced Data Compression Techniques #Advanced Image and Video Retrieval Techniques #Computation and Language (cs.CL) #Computer science #Engineering #FOS: Computer and information sciences #Information Retrieval (cs.IR) #Information retrieval #Process engineering #Speech and Audio Processing #cs.CL #cs.IR

paper · pdf · doi:10.48550/arxiv.2501.18365

published in arXiv (Cornell University) (Cornell University)

openalex publication_date 2025/01/30 · arxiv published 2025/01/30 · arxiv updated 2025/01/30 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Retrieval-augmented generation (RAG) enhances large language models (LLMs) by integrating external knowledge retrieved from a knowledge base. However, its effectiveness is fundamentally constrained by the reliability of both the retriever and the knowledge base. In real-world scenarios, imperfections in these components often lead to the retrieval of noisy, irrelevant, or misleading counterfactual information, ultimately undermining the trustworthiness of RAG systems. To address this challenge, we propose Robust Fine-Tuning (RbFT), a method designed to enhance the resilience of LLMs against retrieval defects through two targeted fine-tuning tasks. Experimental results demonstrate that RbFT significantly improves the robustness of RAG systems across diverse retrieval conditions, surpassing existing methods while maintaining high inference efficiency and compatibility with other robustness techniques.

Cited by

• MetaSyn: A Benchmark for LLM Agents on Meta-Analysis Articles from Nature Portfolio
• Generalized Pseudo-Relevance Feedback
• RWGBench: Evaluating Scholarly Positioning in Related Work Generation
• LLM-based Agents Suffer from Hallucinations: A Survey of Taxonomy, Methods, and Directions
• Do Large Language Models Favor Recent Content? A Study on Recency Bias in LLM-Based Reranking
• SurGE: A Benchmark and Evaluation Framework for Scientific Survey Generation
• Confident RAG: Enhancing the Performance of LLMs for Mathematics Question Answering through Multi-Embedding and Confidence Scoring
• Augmenting Multi-Agent Communication with State Delta Trajectory
• Joint Evaluation of Answer and Reasoning Consistency for Hallucination Detection in Large Reasoning Models
• Magic Mushroom: A Customizable Benchmark for Fine-grained Analysis of Retrieval Noise Erosion in RAG Systems
• Decoupling Reasoning and Knowledge Injection for In-Context Knowledge Editing
• Curriculum Guided Reinforcement Learning for Efficient Multi Hop Retrieval Augmented Generation
• Out of Style: RAG's Fragility to Linguistic Variation

Discussions

• A method for retrieval augmented generation that adds robustness against irrelevant information: 1. Defects Detection: Evaluating the existence of misinformation in the retrieval results. 2. Utility E [bsky, 0 points, 0 comments]

Related

• Astute RAG: Overcoming Imperfect Retrieval Augmentation and Knowledge Conflicts for Large Language Models
• Finetune-RAG: Fine-Tuning Language Models to Resist Hallucination in Retrieval-Augmented Generation
• RAGGED: Towards Informed Design of Scalable and Stable RAG Systems
• Investigating the Robustness of Retrieval-Augmented Generation at the Query Level
• Evaluating the Retrieval Robustness of Large Language Models