Saved in:
Bibliographic Details
Main Authors: Huang, Siyuan, Ma, Zhiyuan, Du, Jintao, Meng, Changhua, Wang, Weiqiang, Leng, Jingwen, Guo, Minyi, Lin, Zhouhan
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2502.11116
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910993972461568
author Huang, Siyuan
Ma, Zhiyuan
Du, Jintao
Meng, Changhua
Wang, Weiqiang
Leng, Jingwen
Guo, Minyi
Lin, Zhouhan
author_facet Huang, Siyuan
Ma, Zhiyuan
Du, Jintao
Meng, Changhua
Wang, Weiqiang
Leng, Jingwen
Guo, Minyi
Lin, Zhouhan
contents RAG systems rely on rerankers to identify relevant documents. However, fine-tuning these models remains challenging due to the scarcity of annotated query-document pairs. Existing distillation-based approaches suffer from training-inference misalignment and fail to capture interdependencies among candidate documents. To overcome these limitations, we reframe the reranking process as an attention-mask problem and propose Gumbel Reranking, an end-to-end training framework for rerankers aimed at minimizing the training-inference gap. In our approach, reranker optimization is reformulated as learning a stochastic, document-wise Top-$k$ attention mask using the Gumbel Trick and Relaxed Top-$k$ Sampling. This formulation enables end-to-end optimization by minimizing the overall language loss. Experiments across various settings consistently demonstrate performance gains, including a 10.4\% improvement in recall on HotpotQA for distinguishing indirectly relevant documents.
format Preprint
id arxiv_https___arxiv_org_abs_2502_11116
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Gumbel Reranking: Differentiable End-to-End Reranker Optimization
Huang, Siyuan
Ma, Zhiyuan
Du, Jintao
Meng, Changhua
Wang, Weiqiang
Leng, Jingwen
Guo, Minyi
Lin, Zhouhan
Computation and Language
Information Retrieval
RAG systems rely on rerankers to identify relevant documents. However, fine-tuning these models remains challenging due to the scarcity of annotated query-document pairs. Existing distillation-based approaches suffer from training-inference misalignment and fail to capture interdependencies among candidate documents. To overcome these limitations, we reframe the reranking process as an attention-mask problem and propose Gumbel Reranking, an end-to-end training framework for rerankers aimed at minimizing the training-inference gap. In our approach, reranker optimization is reformulated as learning a stochastic, document-wise Top-$k$ attention mask using the Gumbel Trick and Relaxed Top-$k$ Sampling. This formulation enables end-to-end optimization by minimizing the overall language loss. Experiments across various settings consistently demonstrate performance gains, including a 10.4\% improvement in recall on HotpotQA for distinguishing indirectly relevant documents.
title Gumbel Reranking: Differentiable End-to-End Reranker Optimization
topic Computation and Language
Information Retrieval
url https://arxiv.org/abs/2502.11116