Retriever Portfolios: A Principled Approach to Adaptive RAG
This work provides a principled approach to improve the adaptability and performance of RAG systems for users facing diverse query types, offering a practical solution to enhance accuracy and efficiency.
This paper addresses the challenge of heterogeneous queries in RAG systems by proposing a method to automatically select a small, diverse subset of retrievers (a portfolio) from a large pool. The learned portfolios and router pipeline consistently outperform single-retriever and naive multi-retriever baselines on retrieval metrics and answer quality across multiple QA benchmarks.
Retrieval-augmented generation (RAG) systems typically rely on a single retriever and a single set of hyperparameters, despite facing highly heterogeneous queries that range from simple factoid questions to complex multi-hop reasoning. We propose a method that automatically selects a small, diverse subset of retrievers (a portfolio) from a large pool of candidates, to cover different regions of the target query distribution. We formalize this setting via an expected best-of-$k$ objective over the query distribution and show that it admits an efficient portfolio construction algorithm with near-optimal guarantees. Across multiple QA benchmarks, our learned portfolios and router pipeline consistently outperform single-retriever and naive multi-retriever baselines on both retrieval metrics and answer quality. In addition, compared to inference-time hyperparameter tuning approaches, fixed portfolios enable parallel retrieval and LLM calls, achieving comparable (and sometimes better) accuracy with substantially lower latency and token cost.