pith. machine review for the scientific record. sign in

arxiv: 2604.20452 · v1 · submitted 2026-04-22 · 💻 cs.IR · cs.CL

Recognition: unknown

HaS: Accelerating RAG through Homology-Aware Speculative Retrieval

Peng Peng , Weiwei Lin , Wentai Wu , Xinyang Wang , Yongheng Liu
Authors on Pith no claims yet

Pith reviewed 2026-05-09 23:21 UTC · model grok-4.3

classification 💻 cs.IR cs.CL
keywords RAG accelerationspeculative retrievalhomology-aware validationretrieval latencyLLM knowledge augmentationmulti-hop reasoningagentic pipelinesplug-and-play retrieval
0
0 comments X

The pith

HaS accelerates RAG retrieval by running quick speculative searches in limited scopes and accepting the results when the current query matches a prior homologous one.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents HaS as a plug-and-play framework that first retrieves candidate documents through fast, narrow-scope searches and then checks whether those documents satisfy the incoming query. The check relies on identifying whether the new query is a homologous re-encounter of a previously seen query; if so, the earlier documents are used and the full-database retrieval step is skipped. A reader would care because retrieval time grows with database size and currently limits how often and how deeply RAG can be used inside LLMs. The reported outcome is a 24 to 37 percent reduction in retrieval latency across datasets, accompanied by only a 1 to 2 percent accuracy drop. The same mechanism also speeds up multi-hop reasoning chains without requiring changes to the rest of the pipeline.

Core claim

HaS performs low-latency speculative retrieval over restricted scopes to obtain candidate documents, followed by validating whether they contain the required knowledge. The validation, grounded in the homology relation between queries, is formulated as a homologous query re-identification task: once a previously observed query is identified as a homologous re-encounter of the incoming query, the draft is deemed acceptable, allowing the system to bypass slow full-database retrieval. Benefiting from the prevalence of homologous queries under real-world popularity patterns, HaS achieves substantial efficiency gains.

What carries the argument

The homologous query re-identification task that decides acceptance of a speculative document draft by matching the current query to a prior homologous query.

If this is right

  • Retrieval latency falls 23.74 percent to 36.99 percent on the evaluated datasets.
  • Answer accuracy declines by only 1 to 2 percent.
  • Complex multi-hop queries inside agentic RAG pipelines complete faster without any modification to the underlying retriever or generator.
  • The approach functions as a drop-in layer that leaves existing RAG code unchanged.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Query logs could be mined offline to pre-build larger sets of homology-linked drafts, further increasing the hit rate.
  • The method may combine naturally with approximate nearest-neighbor indexes, because the speculative stage already operates on restricted scopes.
  • In production settings the savings would grow when the same homology cache is shared across many users with overlapping interests.

Load-bearing premise

Real-world queries often share close homology with earlier ones, and matching to such an earlier query reliably shows that its documents already contain everything the new query needs.

What would settle it

A workload of mostly unique queries with no detectable homologues, or a direct count of cases in which the homology check accepts documents that later prove insufficient for correct generation.

Figures

Figures reproduced from arXiv: 2604.20452 by Peng Peng, Weiwei Lin, Wentai Wu, Xinyang Wang, Yongheng Liu.

Figure 1
Figure 1. Figure 1: Retrieval is much slower than generation, as revealed [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Illustration of different approaches for accelerating [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Framework of HaS. Given a query q, the two-channel fast retrieval is first performed. Documents A–F are retrieved, and the Top-3 (A, B, and D) form the draft. For validation, documents in the draft are indexed to cached queries by the document￾query inverted index. Frequencies of queries hit in the cache are used to compute their homology scores for threshold-based re-identification. If any quasi-homologou… view at source ↗
Figure 4
Figure 4. Figure 4: Estimated proportion of queries that have homologous [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: An illustration of (fully) homologous queries. [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Distributions of semantic similarity scores and homology scores for Easy Positives (Fully homologous), Hard Positives [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Distribution of the number of attributes queried per [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Dataset augmentation workflow. For entity mentions [PITH_FULL_IMAGE:figures/full_fig_p008_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Comparison on varying threshold settings. Point size in [PITH_FULL_IMAGE:figures/full_fig_p008_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Joint distribution of document rankings for two [PITH_FULL_IMAGE:figures/full_fig_p010_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: System performance under different k and varying thresholds. diminishing returns, causing unnecessary latency for marginal accuracy gains. In our setup, τ = 0.2 achieves a satisfactory balance. HaS is robust to the choice of encoders. In retrieval, the underlying encoder serves to extract features for ENNS, and variations across encoders can result in differing retrieval qual￾ity and preferences. In addit… view at source ↗
Figure 12
Figure 12. Figure 12: An illustrative example for a case study. [PITH_FULL_IMAGE:figures/full_fig_p011_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: System performance with and without HaS in the [PITH_FULL_IMAGE:figures/full_fig_p012_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: An illustrative case study demonstrating how HaS can [PITH_FULL_IMAGE:figures/full_fig_p012_14.png] view at source ↗
read the original abstract

Retrieval-Augmented Generation (RAG) expands the knowledge boundary of large language models (LLMs) at inference by retrieving external documents as context. However, retrieval becomes increasingly time-consuming as the knowledge databases grow in size. Existing acceleration strategies either compromise accuracy through approximate retrieval, or achieve marginal gains by reusing results of strictly identical queries. We propose HaS, a homology-aware speculative retrieval framework that performs low-latency speculative retrieval over restricted scopes to obtain candidate documents, followed by validating whether they contain the required knowledge. The validation, grounded in the homology relation between queries, is formulated as a homologous query re-identification task: once a previously observed query is identified as a homologous re-encounter of the incoming query, the draft is deemed acceptable, allowing the system to bypass slow full-database retrieval. Benefiting from the prevalence of homologous queries under real-world popularity patterns, HaS achieves substantial efficiency gains. Extensive experiments demonstrate that HaS reduces retrieval latency by 23.74% and 36.99% across datasets with only a 1-2% marginal accuracy drop. As a plug-and-play solution, HaS also significantly accelerates complex multi-hop queries in modern agentic RAG pipelines. Source code is available at: https://github.com/ErrEqualsNil/HaS.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper proposes HaS, a homology-aware speculative retrieval framework for RAG systems. It performs low-latency speculative retrieval over restricted scopes to obtain candidate documents, then validates them via a homologous query re-identification task that allows bypassing full-database retrieval when a prior query is deemed homologous. The approach leverages real-world query popularity patterns and is presented as plug-and-play, with experiments claiming 23.74% and 36.99% latency reductions across datasets at a 1-2% accuracy cost, plus gains on multi-hop agentic queries.

Significance. If the homology validation proves reliable, HaS could offer a practical efficiency boost for large-scale RAG without heavy accuracy trade-offs, particularly valuable as knowledge bases grow and for complex multi-hop pipelines. The open-source code at the provided GitHub link supports reproducibility and is a clear strength.

major comments (3)
  1. [§3] §3 (Method, homology definition and re-identification task): The central bypass decision rests on identifying 'homologous re-encounters' as sufficient to confirm that restricted-scope candidates contain all required knowledge. However, no formal definition of homology in terms of knowledge overlap or information need is provided, nor are precision/recall metrics for the re-identification classifier reported. This leaves the 1-2% accuracy claim unanchored and risks accepting insufficient candidate sets, especially for multi-hop queries where partial coverage can break reasoning chains.
  2. [§4] §4 (Experiments): The reported latency reductions (23.74% and 36.99%) and accuracy drops are presented as aggregate results without breakdowns by query similarity buckets, homology strength, or multi-hop vs. single-hop cases. No details on the validation procedure, raw data, or statistical tests are given, making it impossible to confirm that measurements support the claims or rule out post-hoc tuning.
  3. [§4.2] §4.2 (Multi-hop evaluation): The claim that HaS 'significantly accelerates complex multi-hop queries' is load-bearing for the agentic RAG use case, yet no per-hop accuracy or failure-mode analysis is shown. If homology detection accepts partial document sets, chain-level accuracy could degrade more than the marginal 1-2% aggregate suggests.
minor comments (2)
  1. [Abstract] The abstract and introduction use 'homology' without an initial informal definition or example, which could confuse readers unfamiliar with the term in this IR context.
  2. [§4] Table or figure captions for latency/accuracy results should explicitly state the number of runs, confidence intervals, and exact datasets used to improve clarity.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment below, indicating where revisions will be made to strengthen the manuscript.

read point-by-point responses

  1. Referee: [§3] §3 (Method, homology definition and re-identification task): The central bypass decision rests on identifying 'homologous re-encounters' as sufficient to confirm that restricted-scope candidates contain all required knowledge. However, no formal definition of homology in terms of knowledge overlap or information need is provided, nor are precision/recall metrics for the re-identification classifier reported. This leaves the 1-2% accuracy claim unanchored and risks accepting insufficient candidate sets, especially for multi-hop queries where partial coverage can break reasoning chains.

    Authors: We agree that a formal definition of homology grounded in knowledge overlap would improve clarity. In the revised manuscript we will add an explicit definition: two queries are homologous if the document set sufficient to answer the first query is also sufficient to answer the second (i.e., their information needs are covered by the same restricted-scope candidates). We will also report precision and recall of the re-identification classifier on a held-out query-pair validation set in Section 3, together with a brief discussion of failure cases for multi-hop queries. These additions will better anchor the reported accuracy figures. revision: yes

  2. Referee: [§4] §4 (Experiments): The reported latency reductions (23.74% and 36.99%) and accuracy drops are presented as aggregate results without breakdowns by query similarity buckets, homology strength, or multi-hop vs. single-hop cases. No details on the validation procedure, raw data, or statistical tests are given, making it impossible to confirm that measurements support the claims or rule out post-hoc tuning.

    Authors: We will expand Section 4 with breakdowns by query-similarity buckets, homology strength, and separate single-hop versus multi-hop results. We will also add a description of the validation procedure (training and evaluation of the re-identification model), report statistical significance (paired t-tests) for the latency and accuracy differences, and point readers to the already-public GitHub repository for raw per-query logs. A supplementary table summarizing per-bucket statistics will be included if space allows. revision: yes

  3. Referee: [§4.2] §4.2 (Multi-hop evaluation): The claim that HaS 'significantly accelerates complex multi-hop queries' is load-bearing for the agentic RAG use case, yet no per-hop accuracy or failure-mode analysis is shown. If homology detection accepts partial document sets, chain-level accuracy could degrade more than the marginal 1-2% aggregate suggests.

    Authors: We will augment Section 4.2 with per-hop accuracy metrics and a failure-mode analysis that isolates cases where homology detection may accept partial document sets. This will quantify whether the aggregate 1-2% drop masks larger per-hop degradations and will include concrete examples of successful and unsuccessful multi-hop chains. Any observed limitations will be reported transparently. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical claims rest on external measurements

full rationale

The paper presents HaS as an empirical framework whose latency and accuracy gains are demonstrated through experiments on datasets, with no derivation chain, fitted parameters renamed as predictions, or load-bearing self-citations. The homology re-identification task is introduced as a formulation whose validity is checked experimentally rather than by construction from prior results or definitions within the paper itself. No equations or self-referential reductions appear in the provided description.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The approach depends on domain assumptions about query similarity patterns and the sufficiency of homology validation; no free parameters or invented entities are explicitly named in the abstract.

axioms (2)
  • domain assumption Homologous queries are prevalent under real-world popularity patterns
    Abstract states that HaS benefits from this prevalence to achieve efficiency gains.
  • domain assumption Homology-based validation accurately identifies when speculative candidates contain the required knowledge
    This is the condition that allows bypassing full-database retrieval.

pith-pipeline@v0.9.0 · 5538 in / 1336 out tokens · 68089 ms · 2026-05-09T23:21:46.505011+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

39 extracted references · 11 canonical work pages · 8 internal anchors

  1. [1]

    Retrieval-augmented generation for knowledge-intensive nlp tasks,

    P. Lewis, E. Perez, A. Piktus, F. Petroni, V . Karpukhin, N. Goyal, H. K¨uttler, M. Lewis, W.-t. Yih, T. Rockt¨aschel, S. Riedel, and D. Kiela, “Retrieval-augmented generation for knowledge-intensive nlp tasks,” in Advances in Neural Information Processing Systems, vol. 33, 2020, pp. 9459–9474

    2020

  2. [2]

    Cacheblend: Fast large language model serving for rag with cached knowledge fusion,

    J. Yao, H. Li, Y . Liu, S. Ray, Y . Cheng, Q. Zhang, K. Du, S. Lu, and J. Jiang, “Cacheblend: Fast large language model serving for rag with cached knowledge fusion,” inProceedings of the Twentieth European Conference on Computer Systems, 2025, p. 94–109

    2025

  3. [3]

    RAG- Cache: Efficient knowledge caching for retrieval-augmented generation,

    C. Jin, Z. Zhang, X. Jiang, F. Liu, S. Liu, X. Liu, and X. Jin, “RAG- Cache: Efficient knowledge caching for retrieval-augmented generation,” ACM Trans. Comput. Syst., vol. 44, no. 1, Nov. 2025

    2025

  4. [4]

    xrag: Extreme context compression for retrieval-augmented generation with one token,

    X. Cheng, X. Wang, X. Zhang, T. Ge, S.-Q. Chen, F. Wei, H. Zhang, and D. Zhao, “xrag: Extreme context compression for retrieval-augmented generation with one token,” inAdvances in Neural Information Process- ing Systems, vol. 37, 2024, pp. 109 487–109 516

    2024

  5. [5]

    Accelerating retrieval-augmented generation,

    D. Quinn, M. Nouri, N. Patel, J. Salihu, A. Salemi, S. Lee, H. Za- mani, and M. Alian, “Accelerating retrieval-augmented generation,” in Proceedings of the 30th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 1, 2025, p. 15–32

    2025

  6. [6]

    Auto-RAG: Autonomous Retrieval-Augmented Generation for Large Language Models,

    T. Yu, S. Zhang, and Y . Feng, “Auto-RAG: Autonomous retrieval- augmented generation for large language models,” 2024. [Online]. Available: https://arxiv.org/abs/2411.19443

    work page arXiv 2024

  7. [7]

    RA-ISF: Learning to answer and understand from retrieval augmentation via it- erative self-feedback,

    Y . Liu, X. Peng, X. Zhang, W. Liu, J. Yin, J. Cao, and T. Du, “RA-ISF: Learning to answer and understand from retrieval augmentation via it- erative self-feedback,” inFindings of the Association for Computational Linguistics: ACL 2024, 2024, pp. 4730–4749

    2024

  8. [8]

    Im-rag: Multi-round retrieval-augmented generation through learning inner monologues,

    D. Yang, J. Rao, K. Chen, X. Guo, Y . Zhang, J. Yang, and Y . Zhang, “Im-rag: Multi-round retrieval-augmented generation through learning inner monologues,” inProceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval, 2024, p. 730–740

    2024

  9. [9]

    Modular RAG: Transforming RAG systems into LEGO-like reconfigurable frameworks,

    Y . Gao, Y . Xiong, M. Wang, and H. Wang, “Modular RAG: Transforming RAG systems into LEGO-like reconfigurable frameworks,” 2024. [Online]. Available: https://arxiv.org/abs/2407.21059

    work page arXiv 2024

  10. [10]

    Accelerating large-scale inference with anisotropic vector quantization,

    R. Guo, P. Sun, E. Lindgren, Q. Geng, D. Simcha, F. Chern, and S. Kumar, “Accelerating large-scale inference with anisotropic vector quantization,” inProceedings of the 37th International Conference on Machine Learning, vol. 119, 2020, pp. 3887–3896

    2020

  11. [11]

    Caching historical embeddings in conversational search,

    O. Frieder, I. Mele, C. I. Muntean, F. M. Nardini, R. Perego, and N. Tonellotto, “Caching historical embeddings in conversational search,” ACM Trans. Web, vol. 18, no. 4, 2024

    2024

  12. [12]

    Leveraging approximate caching for faster retrieval- augmented generation,

    S. A. Bergman, Z. Ji, A.-M. Kermarrec, D. Petrescu, R. Pires, M. Randl, and M. de V os, “Leveraging approximate caching for faster retrieval- augmented generation,” inProceedings of the 5th Workshop on Machine Learning and Systems, 2025, p. 66–73

    2025

  13. [13]

    Cache-craft: Managing chunk-caches for efficient retrieval-augmented generation,

    S. Agarwal, S. Sundaresan, S. Mitra, D. Mahapatra, A. Gupta, R. Sharma, N. J. Kapu, T. Yu, and S. Saini, “Cache-craft: Managing chunk-caches for efficient retrieval-augmented generation,”Proc. ACM Manag. Data, vol. 3, no. 3, 2025

    2025

  14. [14]

    Qwen3 Technical Report

    A. Yang, A. Li, B. Yang, B. Zhang, B. Hui, B. Zheng, B. Yu, C. Gao, C. Huang, C. Lvet al., “Qwen3 technical report,” 2025. [Online]. Available: https://arxiv.org/abs/2505.09388

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  15. [15]

    The Llama 3 Herd of Models

    A. Grattafiori, A. Dubey, A. Jauhri, A. Pandey, A. Kadian, A. Al-Dahle, A. Letman, A. Mathur, A. Schelten, A. Vaughan et al., “The Llama 3 herd of models,” 2024. [Online]. Available: https://arxiv.org/abs/2407.21783

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  16. [16]

    Mistral 7B

    A. Q. Jiang, A. Sablayrolles, A. Mensch, C. Bamford, D. S. Chaplot, D. de las Casas, F. Bressand, G. Lengyel, G. Lample, L. Saulnier, L. R. Lavaud, M.-A. Lachaux, P. Stock, T. L. Scao, T. Lavril, T. Wang, T. Lacroix, and W. E. Sayed, “Mistral 7b,” 2023. [Online]. Available: https://arxiv.org/abs/2310.06825

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  17. [17]

    The faiss library,

    M. Douze, A. Guzhva, C. Deng, J. Johnson, G. Szilvasy, P.-E. Mazar ´e, M. Lomeli, L. Hosseini, and H. J ´egou, “The faiss library,”IEEE Transactions on Big Data, 2025

    2025

  18. [18]

    Dense passage retrieval for open-domain question an- swering,

    V . Karpukhin, B. Oguz, S. Min, P. Lewis, L. Wu, S. Edunov, D. Chen, and W.-t. Yih, “Dense passage retrieval for open-domain question an- swering,” inProceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), 2020, pp. 6769–6781

    2020

  19. [19]

    Narrowing the knowledge eval- uation gap: Open-domain question answering with multi-granularity answers,

    G. Yona, R. Aharoni, and M. Geva, “Narrowing the knowledge eval- uation gap: Open-domain question answering with multi-granularity answers,” inProceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 2024, pp. 6737– 6751

    2024

  20. [20]

    Simple entity-centric questions challenge dense retrievers,

    C. Sciavolino, Z. Zhong, J. Lee, and D. Chen, “Simple entity-centric questions challenge dense retrievers,” inProceedings of the 2021 Con- ference on Empirical Methods in Natural Language Processing, 2021, pp. 6138–6148

    2021

  21. [21]

    When not to trust language models: Investigating effectiveness of para- metric and non-parametric memories,

    A. Mallen, A. Asai, V . Zhong, R. Das, D. Khashabi, and H. Hajishirzi, “When not to trust language models: Investigating effectiveness of para- metric and non-parametric memories,” inProceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 2023, pp. 9802–9822

    2023

  22. [22]

    Self-rag: Learning to retrieve, generate, and critique through self-reflection,

    A. Asai, Z. Wu, Y . Wang, A. Sil, and H. Hajishirzi, “Self-rag: Learning to retrieve, generate, and critique through self-reflection,” inThe Twelfth International Conference on Learning Representations, 2023

    2023

  23. [23]

    Min- Cache: A hybrid cache system for efficient chatbots with hierarchical embedding matching and LLM,

    K. Haqiq, M. V . Jahan, S. A. Farimani, and S. M. F. Masoom, “Min- Cache: A hybrid cache system for efficient chatbots with hierarchical embedding matching and LLM,”Future Generation Computer Systems, vol. 170, p. 107822, 2025

    2025

  24. [24]

    Corrective Retrieval Augmented Generation

    S.-Q. Yan, J.-C. Gu, Y . Zhu, and Z.-H. Ling, “Corrective retrieval augmented generation,” 2024. [Online]. Available: https://arxiv.org/abs/ 2401.15884

    work page internal anchor Pith review arXiv 2024

  25. [25]

    Triviaqa: A large scale distantly supervised challenge dataset for reading comprehension,

    M. Joshi, E. Choi, D. S. Weld, and L. Zettlemoyer, “Triviaqa: A large scale distantly supervised challenge dataset for reading comprehension,”

  26. [26]

    TriviaQA: A Large Scale Distantly Supervised Challenge Dataset for Reading Comprehension

    [Online]. Available: https://arxiv.org/abs/1705.03551

    work page internal anchor Pith review arXiv

  27. [27]

    SQuAD: 100,000+ questions for machine comprehension of text,

    P. Rajpurkar, J. Zhang, K. Lopyrev, and P. Liang, “SQuAD: 100,000+ questions for machine comprehension of text,” inProceedings of the 2016 Conference on Empirical Methods in Natural Language Process- ing, 2016, pp. 2383–2392

    2016

  28. [28]

    Long-context LLMs meet RAG: Overcoming challenges for long inputs in RAG,

    B. Jin, J. Yoon, J. Han, and S. O. Arik, “Long-context LLMs meet RAG: Overcoming challenges for long inputs in RAG,” inThe Thirteenth International Conference on Learning Representations, 2025

    2025

  29. [29]

    Unsupervised Dense Information Retrieval with Contrastive Learning

    G. Izacard, M. Caron, L. Hosseini, S. Riedel, P. Bojanowski, A. Joulin, and E. Grave, “Unsupervised dense information retrieval with contrastive learning,” 2022. [Online]. Available: https://arxiv.org/abs/2112.09118

    work page internal anchor Pith review arXiv 2022

  30. [30]

    C-Pack: Packed Resources For General Chinese Embeddings

    S. Xiao, Z. Liu, P. Zhang, N. Muennighoff, D. Lian, and J.-Y . Nie, “C-Pack: Packed resources for general chinese embeddings,” 2024. [Online]. Available: https://arxiv.org/abs/2309.07597

    work page internal anchor Pith review arXiv 2024

  31. [31]

    Text Embeddings by Weakly-Supervised Contrastive Pre-training

    L. Wang, N. Yang, X. Huang, B. Jiao, L. Yang, D. Jiang, R. Majumder, and F. Wei, “Text embeddings by weakly-supervised contrastive pre- training,” 2024. [Online]. Available: https://arxiv.org/abs/2212.03533

    work page internal anchor Pith review arXiv 2024

  32. [32]

    RECOMP: Improving retrieval-augmented LMs with context compression and selective augmentation,

    F. Xu, W. Shi, and E. Choi, “RECOMP: Improving retrieval-augmented LMs with context compression and selective augmentation,” inThe Twelfth International Conference on Learning Representations, 2024

    2024

  33. [33]

    Speculative RAG: Enhancing retrieval augmented generation through drafting,

    Z. Wang, Z. Wang, L. Le, S. Zheng, S. Mishra, V . Perot, Y . Zhang, A. Mattapalli, A. Taly, J. Shang, C.-Y . Lee, and T. Pfister, “Speculative RAG: Enhancing retrieval augmented generation through drafting,” in The Thirteenth International Conference on Learning Representations, 2025

    2025

  34. [34]

    TurboRAG: Ac- celerating retrieval-augmented generation with precomputed KV caches for chunked text,

    S. Lu, H. Wang, Y . Rong, Z. Chen, and Y . Tang, “TurboRAG: Ac- celerating retrieval-augmented generation with precomputed KV caches for chunked text,” inProceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, 2025, pp. 6588–6601

    2025

  35. [35]

    Efficient and robust approxi- mate nearest neighbor search using hierarchical navigable small world graphs,

    Y . A. Malkov and D. A. Yashunin, “Efficient and robust approxi- mate nearest neighbor search using hierarchical navigable small world graphs,”IEEE Transactions on Pattern Analysis and Machine Intelli- gence, vol. 42, no. 4, pp. 824–836, 2020

    2020

  36. [36]

    DR-RAG: Applying dynamic document relevance to retrieval-augmented generation for question-answering,

    Z. Hei, W. Liu, W. Ou, J. Qiao, J. Jiao, G. Song, T. Tian, and Y . Lin, “DR-RAG: Applying dynamic document relevance to retrieval-augmented generation for question-answering,” 2024. [Online]. Available: https://arxiv.org/abs/2406.07348

    work page arXiv 2024

  37. [37]

    Accelerating inference of retrieval- augmented generation via sparse context selection,

    Y . Zhu, J.-C. Gu, C. Sikora, H. Ko, Y . Liu, C.-C. Lin, L. Shu, L. Luo, L. Meng, B. Liu, and J. Chen, “Accelerating inference of retrieval- augmented generation via sparse context selection,” inThe Thirteenth International Conference on Learning Representations, 2025

    2025

  38. [38]

    EACO-RAG: Edge- Assisted and Collaborative RAG with Adaptive Knowledge Update,

    J. Li, C. Xu, L. Jia, F. Wang, C. Zhang, and J. Liu, “EACO-RAG: Edge- Assisted and Collaborative RAG with Adaptive Knowledge Update,” Oct. 2024

    2024

  39. [39]

    Federated retrieval augmented generation for multi-product question answering,

    P. Shojaee, S. S. Harsha, D. Luo, A. Maharaj, T. Yu, and Y . Li, “Federated retrieval augmented generation for multi-product question answering,” inProceedings of the 31st International Conference on Computational Linguistics: Industry Track, 2025, pp. 387–397

    2025