mlsanityData to 14 Oct 2026

When Should Dense Retrievers Be Updated in Evolving Corpora? Detecting Out-of-Distribution Corpora Using GradNormIR

arXiv 2506.01877 ↗First posted 2 Jun 2025cs.IR cs.CLcited by 3

Not on the map. Why?

Abstract (arXiv)

Dense retrievers encode texts into embeddings to efficiently retrieve relevant documents from large databases in response to user queries. However, real-world corpora continually evolve, leading to a shift from the original training distribution of the retriever. Without timely updates or retraining, indexing newly emerging documents can degrade retrieval performance for future queries. Thus, identifying when a dense retriever requires an update is critical for maintaining robust retrieval systems. In this paper, we propose a novel task of predicting whether a corpus is out-of-distribution (OOD) relative to a dense retriever before indexing. Addressing this task allows us to proactively manage retriever updates, preventing potential retrieval failures. We introduce GradNormIR, an unsupervised approach that leverages gradient norms to detect OOD corpora effectively. Experiments on the BEIR benchmark demonstrate that GradNormIR enables timely updates of dense retrievers in evolving document collections, significantly enhancing retrieval robustness and efficiency.

Key facts. Cited by 3 papers in the map (data through 14 Oct 2026). Builds on 29 of its 40 references that are in the map. First posted on arXiv 2 Jun 2025.

Builds on

29 of 40 references.

References · 29

  1. 2020Retrieval-Augmented Generation for Knowledge-Intensive NLP Taskscited by 9,297
  2. 2020Dense Passage Retrieval for Open-Domain Question Answeringcited by 3,824
  3. 2018HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answeringcited by 3,754
  4. 2021SimCSE: Simple Contrastive Learning of Sentence Embeddingscited by 2,173
  5. 2016MS MARCO: A Human Generated MAchine Reading COmprehension Datasetcited by 1,605
  6. 2022Text Embeddings by Weakly-Supervised Contrastive Pre-trainingcited by 1,253
  7. 2021BEIR: A Heterogenous Benchmark for Zero-shot Evaluation of Information Retrieval Modelscited by 1,252
  8. 2021Unsupervised Dense Information Retrieval with Contrastive Learningcited by 1,216
  9. 2024M3-Embedding: Multi-Linguality, Multi-Functionality, Multi-Granularity Text Embeddings Through Self-Knowledge Distillationcited by 1,073
  10. 2023Towards General Text Embeddings with Multi-stage Contrastive Learningcited by 723
  11. 2021Large Dual Encoders Are Generalizable Retrieverscited by 456
  12. 2021On the Importance of Gradients for Detecting Distributional Shifts in the Wildcited by 339
17 more
  1. 2020How Context Affects Language Models' Factual Predictionscited by 181
  2. 2020Recall and Learn: Fine-tuning Deep Pretrained Language Models with Less Forgettingcited by 170
  3. 2022Large Language Models with Controllable Working Memorycited by 169
  4. 2021Optimizing Dense Retrieval Model Training with Hard Negativescited by 155
  5. 2021GPL: Generative Pseudo Labeling for Unsupervised Domain Adaptation of Dense Retrievalcited by 119
  6. 2022Out-of-Domain Semantics to the Rescue! Zero-Shot Hybrid Retrieval Modelscited by 26
  7. 2023Continual Learning for Generative Retrieval over Dynamic Corporacited by 25
  8. 2024Leveraging LLMs for Unsupervised Dense Retriever Rankingcited by 19
  9. 2024Multi-Head RAG: Solving Multi-Aspect Problems with LLMscited by 18
  10. 2023Unsupervised Layer-wise Score Aggregation for Textual OOD Detectioncited by 15
  11. 2024Robust Neural Information Retrieval: An Adversarial and Out-of-distribution Perspectivecited by 15
  12. 2024GrowOVER: How Can LLMs Adapt to Growing Real-World Knowledge?cited by 11
  13. 2022Augmenting Document Representations for Dense Retrieval with Interpolation and Perturbationcited by 10
  14. 2023L^2R: Lifelong Learning for First-stage Retrieval with Backward-Compatible Representationscited by 8
  15. 2023Lightweight Learner for Shared Knowledge Lifelong Learningcited by 7
  16. 2023Selecting which Dense Retriever to use for Zero-Shot Searchcited by 7
  17. 2024DESIRE-ME: Domain-Enhanced Supervised Information REtrieval using Mixture-of-Expertscited by 7

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 1; 2026: 2.

3 papers.

  1. 2025Evaluating the Robustness of Dense Retrievers in Interdisciplinary Domainscited by 0
  2. 2026Embedding Surgery: Localized Updates for Adaptive Ranking Correction in Dense Retrievalcited by 0
  3. 2026ABAI at COLIEE 2026 Task 1: Multi-Stage Retrieval with GraphRAG-Enhanced Meta-Learning, and a Post-Hoc Study of the Cross-Validation-to-Test Gapcited by 0