Align then Train: Efficient Retrieval Adapter Learning

2026-04-03

AI Summary

This work addresses the semantic gap between complex queries and simple documents, as well as the high computational cost of fine-tuning large embedding models, by proposing the Efficient Retrieval Adapter (ERA) framework. ERA introduces, for the first time, the pretrain–finetune paradigm from large language models into retrieval adapter learning through a two-stage strategy: it first aligns the embedding spaces of a strong query encoder and a lightweight document encoder via self-supervised learning, then performs query-side supervised adaptation using only a small amount of labeled data—without requiring document re-indexing. Evaluated on the MAIR benchmark spanning six domains and 126 tasks, ERA significantly outperforms annotation-intensive methods under low-label settings, effectively bridging the representation gap and enhancing cross-domain retrieval performance.

Abstract

Dense retrieval systems increasingly need to handle complex queries. In many realistic settings, users express intent through long instructions or task-specific descriptions, while target documents remain relatively simple and static. This asymmetry creates a retrieval mismatch: understanding queries may require strong reasoning and instruction-following, whereas efficient document indexing favors lightweight encoders. Existing retrieval systems often address this mismatch by directly improving the embedding model, but fine-tuning large embedding models to better follow such instructions is computationally expensive, memory-intensive, and operationally burdensome. To address this challenge, we propose Efficient Retrieval Adapter (ERA), a label-efficient framework that trains retrieval adapters in two stages: self-supervised alignment and supervised adaptation. Inspired by the pre-training and supervised fine-tuning stages of LLMs, ERA first aligns the embedding spaces of a large query embedder and a lightweight document embedder, and then uses limited labeled data to adapt the query-side representation, bridging both the representation gap between embedding models and the semantic gap between complex queries and simple documents without re-indexing the corpus. Experiments on the MAIR benchmark, spanning 126 retrieval tasks across 6 domains, show that ERA improves retrieval in low-label settings, outperforms methods that rely on larger amounts of labeled data, and effectively combines stronger query embedders with weaker document embedders across domains.

Problem

• dense retrieval
• instruction-following
• retrieval mismatch
• embedding models
• asymmetric queries

Innovation

• retrieval adapter
• dense retrieval
• embedding alignment
• label-efficient learning
• asymmetric encoding