# Adaptive-RAG: Learning to Adapt Retrieval-Augmented Large Language Models through Question Complexity

**Authors:** Soyeong Jeong, Jinheon Baek, Sukmin Cho, Sung Ju Hwang, Jong C. Park  
**arXiv:** [2403.14403](https://arxiv.org/abs/2403.14403)  
**Date:** 2024-03-21  
**Summary page:** https://www.summarizepaper.com/en/paper/2403.14403v1/

## Key points

- Jeong et al. introduce Adaptive-RAG, a novel framework for adapting retrieval-augmented Large Language Models (LLMs) to varying complexities in Question-Answering tasks.
- The framework selects different LLMs based on the complexity of each query using a classifier that predicts query complexity levels.
- Adaptive-RAG seamlessly switches between iterative and single-step retrieval-augmented LLMs, as well as no-retrieval methods, offering a balanced strategy for responding to diverse query complexities.
- Validation on open-domain QA datasets shows enhanced efficiency and accuracy compared to relevant baselines, including adaptive retrieval approaches.
- The Adaptive-RAG framework represents a significant advancement in QA systems by effectively handling queries with varying complexity levels, improving response accuracy, and providing flexibility and adaptability across different types of queries.

## Summary

, , , , 

In their work titled "Adaptive-RAG: Learning to Adapt Retrieval-Augmented Large Language Models through Question Complexity," Jeong et al. introduce a novel adaptive framework for handling varying complexities in Question-Answering (QA) tasks. Traditional one-size-fits-all approaches may not adequately cover the diverse complexity levels present in real-world queries, which is why the researchers propose a tailored approach where different retrieval-augmented Large Language Models (LLMs) are selected based on the specific complexity of each query. This selection process is operationalized with a classifier that predicts the complexity level of incoming queries using labels collected from actual model outcomes and inherent biases in datasets. By seamlessly adapting between iterative and single-step retrieval-augmented LLMs, as well as no-retrieval methods, the proposed framework offers a balanced strategy for responding to a wide range of query complexities. The researchers validate their model on open-domain QA datasets encompassing multiple complexity levels and demonstrate its enhanced efficiency and accuracy compared to relevant baselines, including adaptive retrieval approaches. Overall, Jeong et al. 's Adaptive-RAG framework represents a significant advancement in QA systems by effectively addressing the challenge of handling queries with varying levels of complexity. Their approach not only improves response accuracy but also offers a more flexible and adaptive solution for optimizing performance across different types of queries. The availability of code further facilitates reproducibility and implementation of their innovative methodology in future research endeavors.