Pith. sign in

REVIEW 2 major objections 2 minor 45 references

Prefix retention optimization closes the indexing-decoding gap in multimodal generative retrieval by aligning prefix rankings and reducing early pruning in beam search.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.3

2026-06-27 14:53 UTC pith:H2GOEYGJ

load-bearing objection The paper gives a concrete engineering fix for prefix loss in multimodal generative retrieval and reports gains on nine tasks, but the survival bound's handling of beam search dependencies is the part that needs the most checking. the 2 major comments →

arxiv 2606.09241 v2 pith:H2GOEYGJ submitted 2026-06-08 cs.IR

Closing the Indexing-Decoding Gap in Multimodal Generative Retrieval via Prefix Retention Optimization

Yufei Chen , Zihan Wang , Yubao Tang , Yukun Zhao , Maarten de Rijke , Zhaochun Ren This is my paper
classification cs.IR
keywords multimodal generative retrievalindexing-decoding gapprefix retentionresidual quantizationbeam search decodingprefix ranking distillationvocabulary scheduling
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper shows that multimodal generative retrieval, which turns retrieval into identifier generation via residual quantization and trie-constrained beam search, creates an indexing-decoding gap because standard reconstruction and contrastive losses do not enforce prefix discriminability. As a result, even well-learned identifiers can be discarded early in decoding. The authors derive a survival bound linking prefix retention to three controllable factors and introduce PRO, a framework with prefix ranking distillation, vocabulary scheduling, and geometric score fusion. Experiments across nine multimodal tasks confirm higher target-prefix retention and better retrieval performance than prior baselines.

Core claim

The central claim is that the indexing-decoding gap arises because identifier learning objectives fail to enforce prefix discriminability during decoding, causing low-rank prefixes to be pruned irreversibly in beam search; this gap is characterized by a derived survival bound relating retention to indexing and decoding factors, and it is closed by PRO, which combines prefix ranking distillation to align quantized prefixes with pre-quantization embeddings, vocabulary scheduling to grow codebook sizes across residual levels, and geometric score fusion to incorporate query-prefix similarity into beam scoring.

What carries the argument

The survival bound on prefix retention, which relates retention probability to three controllable factors in indexing and decoding and guides the three mechanisms of PRO.

Load-bearing premise

The derived survival bound correctly captures how prefix retention depends on the three factors, and the proposed mechanisms close the gap without introducing new mismatches.

What would settle it

An experiment that measures actual prefix retention rates during beam search on the nine tasks and finds that the observed rates deviate substantially from the rates predicted by the survival bound even after applying PRO.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • Target identifier prefixes are retained at higher rates during trie-constrained beam search.
  • Multimodal generative retrieval outperforms existing baselines on nine retrieval tasks.
  • The three mechanisms of distillation, scheduling, and fusion can be applied together in a single training and inference pipeline.
  • Identifier learning no longer needs separate post-hoc fixes for early pruning.

Where Pith is reading between the lines

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

  • The same survival-bound approach could be tested on purely textual generative retrieval to check whether the gap is modality-specific.
  • If vocabulary scheduling reduces early competition, it may also lower the required beam width in production systems.
  • Geometric score fusion adds a vector similarity term to discrete decoding; this hybrid scoring could be extended to other constrained generation tasks.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 2 minor

Summary. The paper claims that multimodal generative retrieval using residual quantization and trie-constrained beam search suffers from an indexing-decoding gap, where even optimized identifiers can be pruned early due to low-rank prefixes. It theoretically characterizes this gap, derives a survival bound relating prefix retention to three controllable factors, and proposes Prefix Retention Optimization (PRO) with three mechanisms—prefix ranking distillation via listwise loss, vocabulary scheduling across residual levels, and geometric score fusion in beam search—to close the gap. Experiments on nine multimodal retrieval tasks show improved target prefix retention and outperformance over existing baselines.

Significance. If the survival bound holds and the mechanisms demonstrably improve retention without new mismatches, the work provides a principled framework for addressing a practical mismatch in generative retrieval, potentially leading to more reliable multimodal systems. The multi-task empirical evaluation is a strength, as is the attempt at a theoretical characterization rather than purely heuristic fixes.

major comments (2)
  1. [§3] §3 (survival bound derivation): the bound is presented as relating prefix retention to the three factors, but the derivation appears to model the factors independently; this risks overlooking the sequential, cumulative-score dependencies inherent to trie-constrained beam search pruning, which could render the bound loose or mis-specified and weaken the causal link to the three PRO mechanisms.
  2. [§4.1–4.3] §4.1–4.3 (PRO mechanisms): the justification for distillation, scheduling, and fusion rests directly on the survival bound; if the bound does not fully capture beam-search dynamics, the mechanisms may address symptoms rather than the root gap, and additional analysis (e.g., ablation on bound tightness) is needed to confirm they do not introduce overfitting to the evaluation setup.
minor comments (2)
  1. [Abstract, §5] Abstract and §5: the claim of 'nine multimodal retrieval tasks' would benefit from an explicit list or table reference to the datasets and metrics used, to allow assessment of baseline fairness and post-hoc tuning.
  2. [§3] Notation: the survival bound and the three factors should be given explicit equation numbers and variable definitions in §3 to facilitate verification of the derivation steps.

Simulated Author's Rebuttal

2 responses · 0 unresolved

Thank you for the opportunity to respond to the referee's comments. We value the feedback on the theoretical characterization and empirical validation of our proposed Prefix Retention Optimization (PRO) framework. Below, we provide point-by-point responses to the major comments.

read point-by-point responses

  1. Referee: [§3] §3 (survival bound derivation): the bound is presented as relating prefix retention to the three factors, but the derivation appears to model the factors independently; this risks overlooking the sequential, cumulative-score dependencies inherent to trie-constrained beam search pruning, which could render the bound loose or mis-specified and weaken the causal link to the three PRO mechanisms.

    Authors: We acknowledge that the survival bound derivation employs an independence assumption among the three factors to obtain a tractable closed-form expression. This simplification may indeed overlook certain sequential dependencies in the beam search process. However, the bound serves as a guiding principle rather than an exact characterization, and our experiments demonstrate consistent improvements in prefix retention across tasks when optimizing the identified factors. In the revised manuscript, we will expand Section 3 to explicitly discuss the assumptions and limitations of the bound, including its potential looseness, and provide an empirical analysis of bound tightness on representative datasets. This will better substantiate the connection to the PRO mechanisms. revision: partial

  2. Referee: [§4.1–4.3] §4.1–4.3 (PRO mechanisms): the justification for distillation, scheduling, and fusion rests directly on the survival bound; if the bound does not fully capture beam-search dynamics, the mechanisms may address symptoms rather than the root gap, and additional analysis (e.g., ablation on bound tightness) is needed to confirm they do not introduce overfitting to the evaluation setup.

    Authors: The PRO mechanisms are motivated by both the theoretical bound and the practical observations of the indexing-decoding gap in Section 3. Prefix ranking distillation targets the prefix ranking factor, vocabulary scheduling addresses the vocabulary size factor to mitigate early pruning, and geometric score fusion incorporates query-prefix similarity to adjust decoding scores. To address the concern about potential overfitting and to verify they target the root gap, we will include additional ablations in the revision: (1) measuring the tightness of the survival bound before and after applying PRO, and (2) evaluating on additional held-out multimodal tasks. The current multi-task results on nine diverse tasks already provide evidence of generalization beyond a single evaluation setup. revision: partial

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper's central derivation is a theoretical survival bound relating prefix retention to three factors in indexing and decoding, presented as derived from gap analysis rather than fitted to data or reduced to self-citations. No equations or claims in the provided text show self-definitional loops, fitted inputs renamed as predictions, or load-bearing self-citations that collapse the result to inputs by construction. The PRO mechanisms are motivated by the bound but remain independent proposals evaluated on external tasks. This is the common case of a self-contained theoretical step with no circular reduction.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The work rests on standard assumptions of residual quantization and trie-constrained beam search in generative retrieval; no new free parameters or invented entities are explicitly introduced in the abstract, though codebook sizes are scheduled.

free parameters (1)
  • codebook sizes per residual level
    Vocabulary scheduling increases sizes from shallow to deep levels; exact values are controllable but not specified as fitted in the abstract.
axioms (2)
  • domain assumption Identifier learning uses reconstruction and contrastive losses that do not enforce prefix discriminability
    Invoked to define the indexing-decoding gap in the abstract.
  • domain assumption Beam search with trie constraints is the decoding procedure
    Stated as the decoding method that exposes the gap.

pith-pipeline@v0.9.1-grok · 5764 in / 1315 out tokens · 19763 ms · 2026-06-27T14:53:56.759249+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of Closing the Indexing-Decoding Gap in Multimodal Generative Retrieval via Prefix Retention Optimization." pith.science (2026). https://pith.science/paper/H2GOEYGJ

@misc{pith2026260609241,
  author       = {Pith},
  title        = {Pith review of: Closing the Indexing-Decoding Gap in Multimodal Generative Retrieval via Prefix Retention Optimization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/H2GOEYGJ}},
  note         = {Machine review of arXiv:2606.09241}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Multimodal generative retrieval formulates multimodal retrieval as discrete identifier generation, eliminating the need for explicit similarity search over external embeddings. Existing approaches construct identifiers via residual quantization and decode them with trie-constrained beam search. This combination introduces an indexing-decoding gap: identifier learning objectives, including reconstruction and contrastive losses, do not explicitly enforce prefix discriminability during decoding. As a result, even well-optimized identifiers can be irreversibly pruned early in beam search due to low-rank prefixes. We theoretically characterize this gap and derive a survival bound that relates prefix retention to three controllable factors in indexing and decoding. Building on this bound, we propose PRO, prefix retention optimization, a unified framework comprising three mechanisms: (i) prefix ranking distillation aligns quantized prefix rankings with those induced by pre-quantization embeddings using a listwise loss; (ii) vocabulary scheduling increases codebook sizes from shallow to deep residual quantization levels to reduce early competition from non-target prefixes; and (iii) geometric score fusion vectorizes each candidate prefix and incorporates its similarity to the query into beam search scoring, further reducing the indexing-decoding mismatch. Experiments on nine multimodal retrieval tasks show that PRO improves retention of target identifier prefixes and outperforms existing multimodal generative retrieval baselines.

Figures

Figures reproduced from arXiv: 2606.09241 by Maarten de Rijke, Yubao Tang, Yufei Chen, Yukun Zhao, Zhaochun Ren, Zihan Wang.

Figure 1
Figure 1. Figure 1: The indexing-decoding gap. Top (decod￾ing): Beam search prunes irrevocably at each level; the target item (colored path) survives levels 1–2 but is lost at level 3. Bottom (indexing): Existing identifier learning objectives do not explicitly su￾pervise prefix survival during decoding. In this paper, we theoretically characterize this gap and show that existing indexing objectives are insufficient to determ… view at source ↗
Figure 2
Figure 2. Figure 2: Empirical diagnostics of the indexing–decoding gap on COCO under quantized oracle [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overview of PRO. During tokenizer training, [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Survival trends predicted by Thm. 2: ranking divergence quintile (Spearman ρ=−0.59), teacher margin quintile, and survival gap between the lowest and highest decoder mismatch quintiles. Ablation study [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Cross-dataset diagnostic analysis. Top row: [PITH_FULL_IMAGE:figures/full_fig_p024_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: End-to-end diagnostics and efficiency. (a) Per-level survival rate of the target identifier [PITH_FULL_IMAGE:figures/full_fig_p024_6.png] view at source ↗

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

45 extracted references · 5 canonical work pages · 4 internal anchors

  1. [1]

    Learning to rank: From pairwise approach to listwise approach

    Zhe Cao, Tao Qin, Tie-Yan Liu, Ming-Feng Tsai, and Hang Li. Learning to rank: From pairwise approach to listwise approach. InProceedings of the International Conference on Machine Learning (ICML), ACM International Conference Proceeding Series, pages 129–136. ACM, 2007

    2007

  2. [2]

    WebQA: Multihop and multimodal QA

    Yingshan Chang, Guihong Cao, Mridu Narang, Jianfeng Gao, Hisami Suzuki, and Yonatan Bisk. WebQA: Multihop and multimodal QA. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 16474–16483. IEEE, 2022

    2022

  3. [3]

    Yang Chen, Hexiang Hu, Yi Luan, Haitian Sun, Soravit Changpinyo, Alan Ritter, and Ming- Wei Chang. Can pre-trained vision and language models answer visual information-seeking questions? InProceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 14948–14968. Association for Computational Linguistics, 2023

    2023

  4. [4]

    Cover and Joy A

    Thomas M. Cover and Joy A. Thomas.Elements of Information Theory. Wiley, 2001

    2001

  5. [5]

    Autoregressive entity retrieval

    Nicola De Cao, Gautier Izacard, Sebastian Riedel, and Fabio Petroni. Autoregressive entity retrieval. InInternational Conference on Learning Representations (ICLR). OpenReview.net, 2021

    2021

  6. [6]

    Bottleneck-minimal indexing for generative doc- ument retrieval

    Xin Du, Lixin Xiu, and Kumiko Tanaka-Ishii. Bottleneck-minimal indexing for generative doc- ument retrieval. InProceedings of the International Conference on Machine Learning (ICML), Proceedings of Machine Learning Research, pages 11888–11904. PMLR / OpenReview.net, 2024

    2024

  7. [7]

    William H. R. Equitz and Thomas M. Cover. Successive refinement of information.IEEE Transactions on Information Theory, 37(2):269–275, 1991

    1991

  8. [8]

    Fedotov, Peter Harremoës, and Flemming Topsøe

    Alexei A. Fedotov, Peter Harremoës, and Flemming Topsøe. Refinements of Pinsker’s inequality. IEEE Transactions on Information Theory, 49(6):1491–1498, 2003

    2003

  9. [9]

    DreamSim: Learning new dimensions of human visual similarity using synthetic data

    Stephanie Fu, Netanel Tamir, Shobhita Sundaram, Lucy Chai, Richard Zhang, Tali Dekel, and Phillip Isola. DreamSim: Learning new dimensions of human visual similarity using synthetic data. InAdvances in Neural Information Processing Systems (NeurIPS), volume 36, 2023

    2023

  10. [10]

    On Using Monolingual Corpora in Neural Machine Translation

    Ça˘glar Gülçehre, Orhan Firat, Kelvin Xu, Kyunghyun Cho, Loïc Barrault, Huei-Chi Lin, Fethi Bougares, Holger Schwenk, and Yoshua Bengio. On using monolingual corpora in neural machine translation.CoRR, abs/1503.03535, 2015

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  11. [11]

    Distilling the Knowledge in a Neural Network

    Geoffrey E. Hinton, Oriol Vinyals, and Jeffrey Dean. Distilling the knowledge in a neural network.CoRR, abs/1503.02531, 2015

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  12. [12]

    Yupeng Hou, Jiacheng Li, Zhankui He, An Yan, Xiusi Chen, and Julian J. McAuley. Bridging language and items for retrieval and recommendation.CoRR, abs/2403.03952, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  13. [13]

    Open-domain visual entity recognition: Towards recognizing millions of Wikipedia entities

    Hexiang Hu, Yi Luan, Yang Chen, Urvashi Khandelwal, Mandar Joshi, Kenton Lee, Kristina Toutanova, and Ming-Wei Chang. Open-domain visual entity recognition: Towards recognizing millions of Wikipedia entities. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 12031–12041. IEEE, 2023. 10

    2023

  14. [14]

    LLM2CLIP: Powerful language model unlocks richer cross-modality representation

    Weiquan Huang, Aoqi Wu, Yifan Yang, Xufang Luo, Yuqing Yang, Usman Naseem, Chunyu Wang, Qi Dai, Xiyang Dai, Dongdong Chen, Chong Luo, Lili Qiu, and Liang Hu. LLM2CLIP: Powerful language model unlocks richer cross-modality representation. InProceedings of the AAAI Conference on Artificial Intelligence (AAAI), pages 5131–5139. AAAI Press, 2026

    2026

  15. [15]

    Product quantization for nearest neighbor search.IEEE Transactions on Pattern Analysis and Machine Intelligence, 33(1):117–128, 2011

    Hervé Jégou, Matthijs Douze, and Cordelia Schmid. Product quantization for nearest neighbor search.IEEE Transactions on Pattern Analysis and Machine Intelligence, 33(1):117–128, 2011

    2011

  16. [16]

    Supervised contrastive learning

    Prannay Khosla, Piotr Teterwak, Chen Wang, Aaron Sarna, Yonglong Tian, Phillip Isola, Aaron Maschinot, Ce Liu, and Dilip Krishnan. Supervised contrastive learning. InAdvances in Neural Information Processing Systems (NeurIPS), volume 33, pages 18661–18673, 2020

    2020

  17. [17]

    GENIUS: A generative framework for universal multimodal search

    Sungyeon Kim, Xinliang Zhu, Xiaofan Lin, Muhammet Bastan, Douglas Gray, and Suha Kwak. GENIUS: A generative framework for universal multimodal search. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 19659–19669. Computer Vision Foundation / IEEE, 2025

    2025

  18. [18]

    Breaking the hourglass phenomenon of residual quantization: Enhancing the upper bound of generative retrieval

    Zhirui Kuai, Zuxu Chen, Huimu Wang, Mingming Li, Dadong Miao, Binbin Wang, Xusong Chen, Li Kuang, Yuxing Han, Jiaxing Wang, Guoyu Tang, Lin Liu, Songlin Wang, and Jingwei Zhuo. Breaking the hourglass phenomenon of residual quantization: Enhancing the upper bound of generative retrieval. InProceedings of the Conference on Empirical Methods in Natural Langu...

    2024

  19. [19]

    Autoregressive image generation using residual quantization

    Doyup Lee, Chiheon Kim, Saehoon Kim, Minsu Cho, and Wook-Shin Han. Autoregressive image generation using residual quantization. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 11513–11522. IEEE, 2022

    2022

  20. [20]

    GLEN: Generative retrieval via lexical index learning

    Sunkyung Lee, Minjin Choi, and Jongwuk Lee. GLEN: Generative retrieval via lexical index learning. InProceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 7693–7704. Association for Computational Linguistics, 2023

    2023

  21. [21]

    SemCORE: A semantic-enhanced generative cross-modal retrieval framework with MLLMs.CoRR, abs/2504.13172, 2025

    Haoxuan Li, Yi Bin, Yunshan Ma, Guoyin Wang, Yang Yang, See-Kiong Ng, and Tat-Seng Chua. SemCORE: A semantic-enhanced generative cross-modal retrieval framework with MLLMs.CoRR, abs/2504.13172, 2025

    work page arXiv 2025

  22. [22]

    MINDER: Multiview identifiers enhanced generative retrieval

    Yongqi Li, Nan Yang, Liang Wang, Furu Wei, and Wenjie Li. MINDER: Multiview identifiers enhanced generative retrieval. InProceedings of the Annual Meeting of the Association for Computational Linguistics (ACL), pages 6636–6648. Association for Computational Linguistics, 2023

    2023

  23. [23]

    Generative cross-modal retrieval: Memorizing images in multimodal language models for retrieval and beyond

    Yongqi Li, Wenjie Wang, Leigang Qu, Liqiang Nie, Wenjie Li, and Tat-Seng Chua. Generative cross-modal retrieval: Memorizing images in multimodal language models for retrieval and beyond. InProceedings of the Annual Meeting of the Association for Computational Linguistics (ACL), pages 11851–11861. Association for Computational Linguistics, 2024

    2024

  24. [24]

    Revolutionizing text-to-image retrieval as autoregressive token-to-voken generation

    Yongqi Li, Hongru Cai, Wenjie Wang, Leigang Qu, Yinwei Wei, Wenjie Li, Liqiang Nie, and Tat-Seng Chua. Revolutionizing text-to-image retrieval as autoregressive token-to-voken generation. InProceedings of the International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR), pages 813–822. ACM, 2025

    2025

  25. [25]

    Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C

    Tsung-Yi Lin, Michael Maire, Serge J. Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C. Lawrence Zitnick. Microsoft COCO: Common objects in context. In European Conference on Computer Vision (ECCV), Lecture Notes in Computer Science, pages 740–755. Springer, 2014

    2014

  26. [26]

    Image retrieval on real-life images with pre-trained vision-and-language models

    Zheyuan Liu, Cristian Rodriguez Opazo, Damien Teney, and Stephen Gould. Image retrieval on real-life images with pre-trained vision-and-language models. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 2105–2114. IEEE, 2021

    2021

  27. [27]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agar- wal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya 11 Sutskever. Learning transferable visual models from natural language supervision. InProceed- ings of the International Conference on Machine Learning (ICML), Proceedings of Mach...

    2021

  28. [28]

    Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J. Liu. Exploring the limits of transfer learning with a unified text-to-text transformer.Journal of Machine Learning Research, 21:140:1–140:67, 2020

    2020

  29. [29]

    Tran, Jonah Samost, Maciej Kula, Ed H

    Shashank Rajput, Nikhil Mehta, Anima Singh, Raghunandan Hulikal Keshavan, Trung Vu, Lukasz Heldt, Lichan Hong, Yi Tay, Vinh Q. Tran, Jonah Samost, Maciej Kula, Ed H. Chi, and Mahesh Sathiamoorthy. Recommender systems with generative retrieval. InAdvances in Neural Information Processing Systems (NeurIPS), volume 36, 2023

    2023

  30. [30]

    Sequence level training with recurrent neural networks

    Marc’Aurelio Ranzato, Sumit Chopra, Michael Auli, and Wojciech Zaremba. Sequence level training with recurrent neural networks. InInternational Conference on Learning Representa- tions (ICLR), 2016

    2016

  31. [31]

    A historical perspective on Schützenberger-Pinsker inequalities

    Olivier Rioul. A historical perspective on Schützenberger-Pinsker inequalities. InGeometric Science of Information (GSI), Lecture Notes in Computer Science, pages 291–306. Springer, 2023

    2023

  32. [32]

    f-divergence inequalities.IEEE Transactions on Information Theory, 62(11):5973–6006, 2016

    Igal Sason and Sergio Verdú. f-divergence inequalities.IEEE Transactions on Information Theory, 62(11):5973–6006, 2016

    2016

  33. [33]

    Learning to tokenize for generative retrieval

    Weiwei Sun, Lingyong Yan, Zheng Chen, Shuaiqiang Wang, Haichao Zhu, Pengjie Ren, Zhumin Chen, Dawei Yin, Maarten de Rijke, and Zhaochun Ren. Learning to tokenize for generative retrieval. InAdvances in Neural Information Processing Systems (NeurIPS), volume 36, 2023

    2023

  34. [34]

    Gen- erative retrieval meets multi-graded relevance

    Yubao Tang, Ruqing Zhang, Jiafeng Guo, Maarten de Rijke, Wei Chen, and Xueqi Cheng. Gen- erative retrieval meets multi-graded relevance. InAdvances in Neural Information Processing Systems (NeurIPS), volume 38, 2024

    2024

  35. [35]

    Cohen, and Donald Metzler

    Yi Tay, Vinh Tran, Mostafa Dehghani, Jianmo Ni, Dara Bahri, Harsh Mehta, Zhen Qin, Kai Hui, Zhe Zhao, Jai Prakash Gupta, Tal Schuster, William W. Cohen, and Donald Metzler. Transformer memory as a differentiable search index. InAdvances in Neural Information Processing Systems (NeurIPS), volume 35, 2022

    2022

  36. [36]

    A neural corpus indexer for document retrieval

    Yujing Wang, Yingyan Hou, Haonan Wang, Ziming Miao, Shibin Wu, Qi Chen, Yuqing Xia, Chengmin Chi, Guoshuai Zhao, Zheng Liu, Xing Xie, Hao Sun, Weiwei Deng, Qi Zhang, and Mao Yang. A neural corpus indexer for document retrieval. InAdvances in Neural Information Processing Systems (NeurIPS), volume 35, 2022

    2022

  37. [37]

    UniIR: Training and benchmarking universal multimodal information retrievers

    Cong Wei, Yang Chen, Haonan Chen, Hexiang Hu, Ge Zhang, Jie Fu, Alan Ritter, and Wenhu Chen. UniIR: Training and benchmarking universal multimodal information retrievers. In European Conference on Computer Vision (ECCV), Lecture Notes in Computer Science, pages 387–404. Springer, 2024

    2024

  38. [38]

    Williams and David Zipser

    Ronald J. Williams and David Zipser. A learning algorithm for continually running fully recurrent neural networks.Neural Computation, 1(2):270–280, 1989

    1989

  39. [39]

    Sam Wiseman and Alexander M. Rush. Sequence-to-sequence learning as beam-search op- timization. InProceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1296–1306. Association for Computational Linguistics, 2016

    2016

  40. [40]

    From image descriptions to visual denotations: New similarity metrics for semantic inference over event descriptions

    Peter Young, Alice Lai, Micah Hodosh, and Julia Hockenmaier. From image descriptions to visual denotations: New similarity metrics for semantic inference over event descriptions. Transactions of the Association for Computational Linguistics, 2:67–78, 2014

    2014

  41. [41]

    APAO: Bridging the Training-Inference Gap in Generative Recommendation via Adaptive Prefix-Aware Optimization

    Yuanqing Yu, Yifan Wang, Weizhi Ma, Zhiqiang Guo, and Min Zhang. APAO: Adaptive prefix-aware optimization for generative recommendation.CoRR, abs/2603.02730, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  42. [42]

    Scalable and effective generative information retrieval

    Hansi Zeng, Chen Luo, Bowen Jin, Sheikh Muhammad Sarwar, Tianxin Wei, and Hamed Zamani. Scalable and effective generative information retrieval. InProceedings of the ACM Web Conference (WWW), pages 1441–1452. ACM, 2024. 12

    2024

  43. [43]

    IRGen: Generative modeling for image retrieval

    Yidan Zhang, Ting Zhang, Dong Chen, Yujing Wang, Qi Chen, Xing Xie, Hao Sun, Weiwei Deng, Qi Zhang, Fan Yang, Mao Yang, Qingmin Liao, Jingdong Wang, and Baining Guo. IRGen: Generative modeling for image retrieval. InEuropean Conference on Computer Vision (ECCV), Lecture Notes in Computer Science, pages 21–41. Springer, 2024

    2024

  44. [44]

    Enhancing generative retrieval with reinforcement learning from relevance feedback

    Yujia Zhou, Zhicheng Dou, and Ji-Rong Wen. Enhancing generative retrieval with reinforcement learning from relevance feedback. InProceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 12481–12490. Association for Computational Linguistics, 2023. 13 Supplementary Material Our supplementary material has the followin...

    2023

  45. [45]

    All pairwise comparisons between our full system and GENIUS are statistically significant (p <0.01 , two-sample t-test), including the smallest gap (InfoSeek R@1, ∆=1.2, p <0.01)

    are below 0.40 in all cases. All pairwise comparisons between our full system and GENIUS are statistically significant (p <0.01 , two-sample t-test), including the smallest gap (InfoSeek R@1, ∆=1.2, p <0.01). Dense baselines (CLIP-SF, BLIP-FF) use deterministic pretrained weights. Table 15: Hyperparameter sensitivity (COCO t→i). Default values inbold. Eac...