Bridging the Training-Inference Gap for Dense Phrase Retrieval

Gyuwan Kim, Jinhyuk Lee, Barlas Oguz, Wenhan Xiong, Yizhe Zhang, Yashar Mehdad, William Yang Wang


Abstract
Building dense retrievers requires a series of standard procedures, including training and validating neural models and creating indexes for efficient search. However, these procedures are often misaligned in that training objectives do not exactly reflect the retrieval scenario at inference time. In this paper, we explore how the gap between training and inference in dense retrieval can be reduced, focusing on dense phrase retrieval (Lee et al., 2021) where billions of representations are indexed at inference. Since validating every dense retriever with a large-scale index is practically infeasible, we propose an efficient way of validating dense retrievers using a small subset of the entire corpus. This allows us to validate various training strategies including unifying contrastive loss terms and using hard negatives for phrase retrieval, which largely reduces the training-inference discrepancy. As a result, we improve top-1 phrase retrieval accuracy by 2 3 points and top-20 passage retrieval accuracy by 2 4 points for open-domain question answering. Our work urges modeling dense retrievers with careful consideration of training and inference via efficient validation while advancing phrase retrieval as a general solution for dense retrieval.
Anthology ID:
2022.findings-emnlp.272
Volume:
Findings of the Association for Computational Linguistics: EMNLP 2022
Month:
December
Year:
2022
Address:
Abu Dhabi, United Arab Emirates
Editors:
Yoav Goldberg, Zornitsa Kozareva, Yue Zhang
Venue:
Findings
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
3713–3724
Language:
URL:
https://aclanthology.org/2022.findings-emnlp.272
DOI:
10.18653/v1/2022.findings-emnlp.272
Bibkey:
Cite (ACL):
Gyuwan Kim, Jinhyuk Lee, Barlas Oguz, Wenhan Xiong, Yizhe Zhang, Yashar Mehdad, and William Yang Wang. 2022. Bridging the Training-Inference Gap for Dense Phrase Retrieval. In Findings of the Association for Computational Linguistics: EMNLP 2022, pages 3713–3724, Abu Dhabi, United Arab Emirates. Association for Computational Linguistics.
Cite (Informal):
Bridging the Training-Inference Gap for Dense Phrase Retrieval (Kim et al., Findings 2022)
Copy Citation:
PDF:
https://aclanthology.org/2022.findings-emnlp.272.pdf
Video:
 https://aclanthology.org/2022.findings-emnlp.272.mp4