The field of retrieval-augmented generation is moving towards more efficient and effective methods for integrating external knowledge sources into large language models. Recent research has focused on developing novel frameworks and techniques for improving the reliability and utility of these models. One key direction is the use of hierarchical and multi-granular approaches to document retrieval, which enable more precise and context-aware retrieval of relevant information. Another area of innovation is the development of methods for identifying and mitigating the negative impact of irrelevant or distracting passages on the generation process. Notable papers in this area include ArtRAG, which proposes a novel framework for multi-perspective artwork explanation, and MacRAG, which introduces a hierarchical retrieval framework for multi-scale adaptive context retrieval. Additionally, papers such as CAFE and LongRefiner have made significant contributions to enhancing multi-document question-answering capabilities and improving the efficiency of long-context retrieval-augmented generation.
Advances in Retrieval-Augmented Generation
Sources
Lost in OCR Translation? Vision-Based Approaches to Robust Document Retrieval
ArtRAG: Retrieval-Augmented Generation with Structured Context for Visual Art Understanding
References Indeed Matter? Reference-Free Preference Optimization for Conversational Query Reformulation
MacRAG: Compress, Slice, and Scale-up for Multi-Scale Adaptive Context RAG
The Distracting Effect: Understanding Irrelevant Passages in RAG
Pre-training vs. Fine-tuning: A Reproducibility Study on Dense Retrieval Knowledge Acquisition
Reproducibility, Replicability, and Insights into Visual Document Retrieval with Late Interaction
Patchwork: A Unified Framework for RAG Serving
OMGM: Orchestrate Multiple Granularities and Modalities for Efficient Multimodal Retrieval
Efficient and Reproducible Biomedical Question Answering using Retrieval Augmented Generation
Enhancing Cache-Augmented Generation (CAG) with Adaptive Contextual Compression for Scalable Knowledge Integration
Optimizing Retrieval-Augmented Generation: Analysis of Hyperparameter Impact on Performance and Efficiency
IterKey: Iterative Keyword Generation with LLMs for Enhanced Retrieval Augmented Generation
PsOCR: Benchmarking Large Multimodal Models for Optical Character Recognition in Low-resource Pashto Language
CAFE: Retrieval Head-based Coarse-to-Fine Information Seeking to Enhance Multi-Document QA Capability
Hierarchical Document Refinement for Long-context Retrieval-augmented Generation
CL-RAG: Bridging the Gap in Retrieval-Augmented Generation with Curriculum Learning
Enhancing Multi-Image Question Answering via Submodular Subset Selection