Influence Guided Sampling for Domain Adaptation of Text Retrievers

TL;DR

This paper introduces Inf-DDS, a reinforcement learning-based sampling method that adaptively prioritizes training datasets to improve domain adaptation in text retrieval models, achieving significant performance gains and reduced computational costs.

Contribution

It presents a novel influence-guided sampling framework that adaptively reweighs datasets for training text retrievers, outperforming existing methods in effectiveness and efficiency.

Findings

Achieved a 5.03 absolute NDCG@10 improvement on a multilingual model.

Improved retrieval performance by 0.94 NDCG@10 on MiniLM-L6-v2.

Reduced GPU costs by 1.5 to 4 times compared to baseline methods.

Abstract

General-purpose open-domain dense retrieval systems are usually trained with a large, eclectic mix of corpora and search tasks. How should these diverse corpora and tasks be sampled for training? Conventional approaches sample them uniformly, proportional to their instance population sizes, or depend on human-level expert supervision. It is well known that the training data sampling strategy can greatly impact model performance. However, how to find the optimal strategy has not been adequately studied in the context of embedding models. We propose Inf-DDS, a novel reinforcement learning driven sampling framework that adaptively reweighs training datasets guided by influence-based reward signals and is much more lightweight with respect to GPU consumption. Our technique iteratively refines the sampling policy, prioritizing datasets that maximize model performance on a target development set. We evaluate the efficacy of our sampling strategy on a wide range of text retrieval tasks, demonstrating strong improvements in retrieval performance and better adaptation compared to existing gradient-based sampling methods, while also being 1.5x to 4x cheaper in GPU compute. Our sampling strategy achieves a 5.03 absolute NDCG@10 improvement while training a multilingual bge-m3 model and an absolute NDCG@10 improvement of 0.94 while training all-MiniLM-L6-v2, even when starting from expert-assigned weights on a large pool of training datasets.