Adam: Dense Retrieval Distillation with Adaptive Dark Examples

TL;DR

This paper introduces ADAM, a novel knowledge distillation framework for dense retrieval that uses adaptive dark examples and self-paced learning to enhance the transfer of dark knowledge from teacher to student, improving retrieval performance.

Contribution

The paper proposes a new method for constructing dark examples and a self-paced distillation strategy to better transfer dark knowledge in dense retrieval models.

Findings

ADAM outperforms existing methods on benchmark datasets.

Dark examples with moderate relevance improve knowledge transfer.

Self-paced distillation enhances student model learning.

Abstract

To improve the performance of the dual-encoder retriever, one effective approach is knowledge distillation from the cross-encoder ranker. Existing works construct the candidate passages following the supervised learning setting where a query is paired with a positive passage and a batch of negatives. However, through empirical observation, we find that even the hard negatives from advanced methods are still too trivial for the teacher to distinguish, preventing the teacher from transferring abundant dark knowledge to the student through its soft label. To alleviate this issue, we propose ADAM, a knowledge distillation framework that can better transfer the dark knowledge held in the teacher with Adaptive Dark exAMples. Different from previous works that only rely on one positive and hard negatives as candidate passages, we create dark examples that all have moderate relevance to the query through mixing-up and masking in discrete space. Furthermore, as the quality of knowledge held in different training instances varies as measured by the teacher's confidence score, we propose a self-paced distillation strategy that adaptively concentrates on a subset of high-quality instances to conduct our dark-example-based knowledge distillation to help the student learn better. We conduct experiments on two widely-used benchmarks and verify the effectiveness of our method.