Pith. sign in

REVIEW 2 major objections 1 minor 2 cited by

Annotation-free tool-wise rewards let a fine-tuned 4B model outperform strong baselines on long-horizon preference-grounded shopping, where even frontier models stay under 70% success.

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-4.5

2026-07-14 20:51 UTC pith:L6C4LIYU

load-bearing objection We only have the Shopping Companion abstract; the attached full text is a different paper (EnSF data assimilation), so the shopping claims cannot be audited. the 2 major comments →

arxiv 2603.14864 v3 pith:L6C4LIYU submitted 2026-03-16 cs.CL

Shopping Companion: Benchmarking and Training LLM Agents for Long-Horizon Preference-Grounded E-Commerce Tasks

Zijian Yu , Kejun Xiao , Huaipeng Zhao , Tao Luo , Xiaoyi Zeng This is my paper
classification cs.CL
keywords LLM agentse-commercepreference memoryprocess supervisiontool-wise rewardsshopping benchmarklong-horizon tasks
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.

LLM shopping agents must track user preferences across long, multi-session conversations, yet the field has lacked both hard evaluation suites and dense training signals. This paper introduces Shopping Companion Bench: two tasks that require cross-session preference memory over a pool of more than 1.2 million real products. Failure analysis isolates two main error sources—cascading preference hallucination and weak verification of product attributes against stated requirements. To fix them, the authors design annotation-free rewards that score each tool call, giving process supervision instead of a single sparse terminal signal. A lightweight 4B model trained with these rewards consistently beats strong baselines on preference capture and task success, while even state-of-the-art models remain below 70% success, showing both the benchmark’s difficulty and the reward design’s value.

Core claim

Annotation-free, tool-wise process rewards alleviate reward sparsity on long-horizon preference-grounded shopping tasks, enabling a fine-tuned 4B model to outperform strong baselines on Shopping Companion Bench—a new two-task suite requiring cross-session preference memory over 1.2 million real items—where even models such as GPT-5 stay below 70% success.

What carries the argument

Shopping Companion Bench plus annotation-free tool-wise rewards that assign process supervision to every tool call; these rewards specifically target preference hallucination and attribute-verification failures so credit is no longer delayed until task end.

Load-bearing premise

That success on the two constructed tasks under the authors’ tool-wise rewards measures real long-horizon shopping competence rather than artifacts of synthetic user trajectories, product-pool construction, or reward hacking of the tool interface.

What would settle it

An independent evaluation with real human shoppers or a held-out product pool in which the 4B model no longer beats baselines, or in which tool-wise rewards fail to reduce preference hallucination relative to terminal-only rewards, would falsify the central claim.

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

If this is right

  • Long-horizon shopping agents can be trained without expensive human process annotations.
  • Lightweight models become competitive for preference-grounded e-commerce once dense tool-level rewards are available.
  • Preference hallucination and attribute verification become first-class failure modes that future agents must explicitly address.
  • Cross-session memory benchmarks over large real catalogs will stay hard for frontier models until process supervision improves.
  • Tool-level reward design can transfer to other multi-step agent domains that suffer from sparse terminal rewards.

Where Pith is reading between the lines

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

  • The same tool-wise process rewards could improve agents for multi-session preference tasks outside shopping, such as travel planning or medical intake.
  • Larger models may close or reverse the gap if they are also trained with the same dense tool rewards rather than only terminal success.
  • Reported gains will largely depend on how faithfully the synthetic user trajectories match real shopper behavior.

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 / 1 minor

Summary. The submission is presented under the title and abstract of “Shopping Companion: Benchmarking and Training LLM Agents for Long-Horizon Preference-Grounded E-Commerce Tasks” (arXiv:2603.14864, cs.CL). That abstract claims a new long-horizon preference-grounded shopping benchmark (Shopping Companion Bench) over a 1.2M-item product pool, a failure-mode analysis (preference hallucination and attribute verification), annotation-free tool-wise process rewards, and experimental results in which GPT-5 stays below 70% success while a fine-tuned 4B model outperforms strong baselines. The body that was supplied, however, is an entirely different manuscript: “A Score Filter Enhanced Data Assimilation Framework for Data-Driven Dynamical Systems” (content matching arXiv:2603.14863), which develops an Ensemble Score Filter (EnSF) hybrid with LSTM and R-DeepONet surrogates and evaluates it on Lorenz–96 and the KdV equation. No Shopping Companion methods, tasks, rewards, or results appear in the body.

Significance. If the abstract’s claims were supported by a matching manuscript, a hard long-horizon preference-memory shopping benchmark with process-level, annotation-free tool rewards and a competitive lightweight agent would be of clear interest to the LLM-agent and e-commerce communities. As submitted, that contribution cannot be assessed: the body contains no benchmark definition, no agent training pipeline, and no e-commerce experiments. The EnSF–ML assimilation work in the body is a separate, technically coherent line of research on uncertainty reduction for data-driven dynamical systems, but it is not the paper announced by the title and abstract. Significance of the claimed Shopping Companion result is therefore indeterminate from this package.

major comments (2)
  1. [Title, Abstract, and full body (Sections 1–5)] Title/abstract vs. body mismatch: the abstract and paper_id announce Shopping Companion Bench, tool-wise rewards, GPT-5 <70% success, and a fine-tuned 4B agent, but Sections 1–5 and all figures/tables develop EnSF-enhanced LSTM/R-DeepONet forecasting on Lorenz–96 and KdV. There is no section defining the two shopping tasks, the 1.2M product pool, preference memory, tool-wise rewards, or any LLM-agent experiment. The central claims of 2603.14864 are therefore unsupported by the supplied manuscript and cannot be audited for soundness, baselines, or ablations.
  2. [Abstract claims vs. Sections 4.1–4.2 (which report only Lorenz–96/KdV RMSE)] Because the body is a different paper, load-bearing Shopping Companion claims (preference hallucination as a cascading failure mode; annotation-free tool-wise rewards alleviating reward sparsity; 4B outperformance of strong baselines; GPT-5 success below 70%) have no methods, equations, tables, or error bars against which to check them. Evaluation of those claims is not possible without the correct manuscript body.
minor comments (1)
  1. [Section 2.1, Eq. (1); Figures 3–15] The supplied body (EnSF paper) itself has ordinary presentation issues—e.g., garbled math glyphs in the LSTM equations (Section 2.1), inconsistent notation for observation operators, and figure captions that sometimes restate results without quantitative detail—but these are secondary to the abstract/body identity failure and would only matter if that paper were the submission under review.

Circularity Check

0 steps flagged

No significant circularity: EnSF+ML forecasts are evaluated against external numerical ground truth (Lorenz-96, KdV) and EnKF baselines; self-citations supply the filter method, not a tautological result.

full rationale

The supplied full manuscript is the EnSF data-assimilation paper (arXiv:2603.14863), not the Shopping Companion abstract (2603.14864). On the manuscript that is actually present: the load-bearing chain is (i) train LSTM / R-DeepONet surrogates, (ii) roll them out in long-term prediction mode where errors accumulate, (iii) correct the forecast ensemble with EnSF (or EnKF) using noisy/partial observations, (iv) report RMSE against trajectories from classical numerical solvers. That chain is not circular. EnSF is imported from the authors’ prior work as a filtering method, but the claimed improvement is an empirical comparison to EnKF and to uncorrected LTP/SSP on external dynamical systems, not a quantity forced by definition or by a uniqueness theorem. Score approximations (mini-batch score, likelihood injection with damping h(t)=1−t) are algorithmic choices, not fits renamed as predictions. No self-definitional loop, no fitted-input-as-prediction, and no load-bearing uniqueness import appear in the derivation. Honest non-finding: score 0; steps empty.

Axiom & Free-Parameter Ledger

0 free parameters · 3 axioms · 2 invented entities

Abstract-only review of Shopping Companion Bench. Load-bearing premises are domain assumptions about what constitutes long-horizon preference-grounded shopping competence, that tool-call process rewards without annotation are valid supervision, and that the constructed product pool and tasks are representative. No free parameters or invented physical entities are stated in the abstract; the main invented construct is the benchmark/reward package itself.

axioms (3)
  • domain assumption Cross-session preference memory and attribute verification against user requirements are the critical bottlenecks for e-commerce LLM agents on long-horizon tasks.
    Abstract frames these as the two major failure sources motivating the benchmark and rewards.
  • ad hoc to paper Annotation-free tool-wise rewards provide sufficient process supervision to alleviate reward sparsity without human step labels.
    Central training claim; abstract asserts effectiveness via 4B fine-tune gains but does not derive necessity or optimality.
  • domain assumption Success rates on the two Shopping Companion Bench tasks are a meaningful proxy for real preference-grounded shopping performance over large catalogs.
    Benchmark validity assumption; required for interpreting GPT-5 <70% and 4B outperformance as field progress.
invented entities (2)
  • Shopping Companion Bench no independent evidence
    purpose: Provide evaluation tasks requiring cross-session preference memory over a 1.2M+ real-product pool.
    Primary contribution named in the abstract; independent evidence outside this paper is not shown here.
  • Annotation-free tool-wise rewards no independent evidence
    purpose: Supply dense process supervision per tool call to train shopping agents without step annotations.
    Methodological construct claimed to fix preference hallucination and weak attribute verification; formulas not given in abstract.

pith-pipeline@v1.1.0-grok45 · 25478 in / 2463 out tokens · 26893 ms · 2026-07-14T20:51:26.325698+00:00 · methodology

0 comments
read the original abstract

In e-commerce, LLM agents show promise for shopping tasks such as recommendations, budget management, and bundle deals, where accurately capturing user preferences from long-horizon conversations is critical. However, progress is limited by two key challenges: (1) the absence of benchmarks for evaluating long-term preference-aware shopping tasks, and (2) the lack of fine-grained supervision for shopping agent training. To fill the benchmark gap, we introduce Shopping Companion Bench, a novel benchmark comprising two shopping tasks that require cross-session preference memory, grounded in a product pool of over 1.2 million real-world items. Our analysis further identifies two major sources of failure on this benchmark: cascading errors caused by preference hallucination, and insufficient verification of product attributes against user requirements. To address these failure modes, we design annotation-free, tool-wise rewards that provide process supervision for each tool call, alleviating reward sparsity in long-horizon tasks. Experimental results demonstrate that even state-of-the-art models such as GPT-5 achieve success rates below 70%, highlighting the difficulty of our benchmark. Notably, our fine-tuned lightweight 4B model consistently outperforms strong baselines in both preference capture and task performance, suggesting the effectiveness of our reward design.

discussion (0)

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

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. ShopGym: An Integrated Framework for Realistic Simulation and Scalable Benchmarking of E-Commerce Web Agents

    cs.AI 2026-05 conditional novelty 7.0

    ShopGym introduces ShopArena to convert live storefronts into self-contained sandbox shops and ShopGuru to synthesize 224 benchmark tasks, with validation showing structural preservation and positive correlation of ag...

  2. Paying to Know: Micro-Transaction Markets for Verified Product Information in Agentic E-Commerce

    cs.CL 2026-06 unverdicted novelty 5.0

    Agentic e-commerce should operate as a micro-transaction market for verified information unlocked progressively by buyer agents, redirecting NLP research toward cost-optimal acquisition, data pricing, and related problems.

Reference graph

Works this paper leans on

88 extracted references · 3 linked inside Pith · cited by 2 Pith papers

  1. [1]

    Allen, S

    Anna A. Allen, S. Markou, W. Tebbutt, J. Requeima, W. P. Br uinsma, T. R. Andersson, M. Herzog, N. D. Lane, M .Chantry, J. S. Hosking, and R. E. Turn er. End-to-end data-driven weather prediction. ������, 641(8065):1172–1179, 5 2025. 21

    2025

  2. [2]

    Andrieu, A

    C. Andrieu, A. Doucet, and R. Holenstein. Particle marko v chain monte carlo methods. �� �� �������� ���� �, 72(3):269–342, 2010

    2010

  3. [3]

    Archibald and F

    R. Archibald and F. Bao. A kernel learning method for back ward sde filter. ����� �������� ����������������, 2022

    2022

  4. [4]

    Archibald, F

    R. Archibald, F. Bao, and X. Tu. A direct method for parame ter estimations. �� ������� �����, 398:108871, 2019

    2019

  5. [5]

    F. Bao, Y. Cao, and X. Han. Forward backward doubly stocha stic differential equations and optimal filtering of diffusion processes. �������������� �� ������������ ��������, 18(3):635– 661, 2020

    2020

  6. [6]

    F. Bao, Y. Cao, and J. Yong. Data informed solution estima tion for forward backward stochastic differential equations. �������� ��� ������������, 19:439–464, 2021

    2021

  7. [7]

    F. Bao, Y. Cao, and W. Zhao. A backward doubly stochastic d ifferential equation approach for nonlinear filtering problems. ����� ����� ���� , 23:1573–1601, 201

  8. [8]

    F. Bao, H. G. Chipilski, S. Liang, G. Zhang, and J. S. Whita ker. Nonlinear ensemble filtering with diffusion models: Application to the surface quasigeost rophic dynamics. ������� ������� ������, 153(7):1155–1169, 2025

    2025

  9. [9]

    F. Bao, Z. Zhang, and G. Zhang. An ensemble score filter for tracking high-dimensional nonlinear dynamical systems. �������� ������� �� ������� ��������� ��� ����������� , 432:117447, 2024

    2024

  10. [10]

    F. Bao, Z. Zhang, and G. Zhang. A score-based filter for no nlinear data assimilation. ������� �� ������������� �������, 514:113207, 2024

    2024

  11. [11]

    United filter for jointly esti mating state and parameters of stochastic dynamical systems

    F Bao, Z Zhang, and G Zhang. United filter for jointly esti mating state and parameters of stochastic dynamical systems. �����, 39(3):747–774, 2026

    2026

  12. [12]

    Beneˇ s, M

    K. Beneˇ s, M. K. Baskar, and L. Burget. Residual memory n etworks in language modeling: Improving the reputation of feed-forward networks. ����� �����������, pages 284–288, 2017

    2017

  13. [13]

    Beucler, M

    T. Beucler, M. Pritchard, S. Rasp, J. Ott, P. Baldi, and P . Gentine. Enforcing analytic con- straints in neural networks emulating physical systems. ����� ���� �����, 126:098302, 2021

    2021

  14. [14]

    M. Buehner. Local ensemble transform kalman filter with cross validation. ��� , 148:2265– 2282, 2020

    2020

  15. [15]

    � �������������� �������� �� �������� �������� ���� ������������ �� ����� ������ ��������

    Michael C Burkhart. � �������������� �������� �� �������� �������� ���� ������������ �� ����� ������ ��������. ProQuest Dissertations Publishing, 2019

    2019

  16. [16]

    Chai and A

    J. Chai and A. Li. Deep learning in natural language proc essing: A state-of-the-art survey. In ���� ������������� ���������� �� ������� �������� ��� ������� ���� ������� , pages 1–6, 2019

    2019

  17. [17]

    J. Chai, H. Zeng, A. Li, and E. W.T. Ngai. Deep learning in computer vision: A critical review of emerging techniques and application scenarios. ������� �������� ���� ������������, 6:100134, 2021. 22

    2021

  18. [18]

    On the brownian displacements and ther mal diffusion of grains suspended in a non-uniform fluid

    Sydney Chapman. On the brownian displacements and ther mal diffusion of grains suspended in a non-uniform fluid. ����������� �� ��� ����� ������� �� ������� ������ �� ���������� ������ �� � ������������ ��� �������� ���������, pages 34–54, 1928

    1928

  19. [19]

    Chattopadhyay, P

    A. Chattopadhyay, P. Hassanzadeh, and D. Subramanian. Data-driven predictions of a multi- scale lorenz 96 chaotic system using machine-learning meth ods: reservoir computing, artificial neural network, and long short-term memory network. ��������� ��������� �� ����������, 27(3):373–389, 2020

    2020

  20. [20]

    Chen and H

    T. Chen and H. Chen. Universal approximation to nonline ar operators by neural networks with arbitrary activation functions and its application to dyna mical systems. ���� ������������ �� ������ ��������, 6(4):911–917, 1995

    1995

  21. [21]

    Chung, C

    J. Chung, C. Gulcehre, K. Cho, and Y. Yere. Empirical eva luation of gated recurrent neural networks on sequence modeling. 2014

    2014

  22. [22]

    An introduction to estimation theory (g tspecial issueltdata assimilation in meteology and oceanography: theory and practice)

    Stephen E Cohn. An introduction to estimation theory (g tspecial issueltdata assimilation in meteology and oceanography: theory and practice). ������� �� ��� �������������� ������� �� ������ ���� ��, 75(1B):257–288, 1997

    1997

  23. [23]

    G. Evensen. The ensemble kalman filter: Theoretical for mulation and practical implementation. ����� �������� , 53(4):343–367, 2003

    2003

  24. [24]

    Sequential data assimilation with a nonl inear quasi-geostrophic model using monte carlo methods to forecast error statistics

    Geir Evensen. Sequential data assimilation with a nonl inear quasi-geostrophic model using monte carlo methods to forecast error statistics. ������� �� ����������� ��������� ������, 99(C5):10143–10162, 1994

    1994

  25. [25]

    ���� ������������� ��� �������� ������ �����

    Gier Evensen. ���� ������������� ��� �������� ������ �����. Springer Science & Business Media, 2009

    2009

  26. [26]

    F. A. Gers, J. Schmidhuber, and F. Cummins. Learning to f orget: continual prediction with lstm. ������ �����������, 12(8):2451–2471, 2000

    2000

  27. [27]

    Learning to forget: Continual prediction with lstm

    Felix A Gers, J¨ urgen Schmidhuber, and Fred Cummins. Learning to forget: Continual prediction with lstm. ������ �����������, 12(10):2451–2471, 2000

    2000

  28. [28]

    Goodfellow, Y

    I. Goodfellow, Y. Bengio, and A. Courville. ���� ��������. MIT Press, 2016. deeplearning- book.org

    2016

  29. [29]

    N. J. Gordon, D. J. Salmond, and A. F. M. Smith. Novel appr oach to nonlinear/non-gaussian bayesian state estimation. ��� ���������� � , 140(2):107–113, 1993

    1993

  30. [30]

    Long short-term memory

    Alex Graves. Long short-term memory. ���������� �������� ��������� ���� ��������� ������ ��������, pages 37–45, 2012

    2012

  31. [31]

    ���������� �������� ��������� ���� ��������� ������ ��������

    Alex Graves. ���������� �������� ��������� ���� ��������� ������ ��������. Studies in Com- putational Intelligence. Springer, Berlin, Heidelberg, 2 012

  32. [32]

    Generating sequences with recurrent neur al networks

    Alex Graves. Generating sequences with recurrent neur al networks. 2013

    2013

  33. [33]

    K. Greff, R. K. Srivastava, J. Koutn ´ ık, B. R. Steunebrink, and J. Schmidhuber. Lstm: A search space odyssey. ���� ������������ �� ������ �������� ��� �������� �������, 28(10):2222–2232, 2016. 23

    2016

  34. [34]

    G. J. Hakim and S. Masanam. Dynamical tests of a deep lear ning weather prediction model. ��������� ������������ ��� ��� ����� �������, 3(3):e230090, 2024

    2024

  35. [35]

    Z. Hao, J. Yao, C. Su, H. Su, Z. Wang, F. Lu, Z. Xia, Y. Zhang , S. Liu, Lu Lu, and J. Zhu. Pinnacle: A comprehensive benchmark of physics-informed n eural networks for solving pdes. In A. Globerson, L. Mackey, D. Belgrave, A. Fan, U. Paquet, J. To mczak, and C. Zhang, editors, �������� �� ������ ����������� ���������� �������, volume 37, pages 76721–7677...

    2024

  36. [36]

    Sequential deep operator networks (s-deeponet) for predic ting full-field solutions under time- dependent loads

    Junyan He, Shashank Kushwaha, Jaewan Park, Seid Koric, Diab Abueidda, and Iwona Jasiuk. Sequential deep operator networks (s-deeponet) for predic ting full-field solutions under time- dependent loads. ����������� ������������ �� ��������� ������������, 127:107258, 2024

    2024

  37. [37]

    Hochreiter and J

    S. Hochreiter and J. Schmidhuber. Long short-term memo ry. ������ ����������� , 9(8):1735– 1780, 1997

    1997

  38. [38]

    Long short-t erm memory

    Sepp Hochreiter and J¨ urgen Schmidhuber. Long short-t erm memory. ������ �����������, 9(8):1735–1780, 1997

    1997

  39. [39]

    Ensemble ka lman filtering

    Peter L Houtekamer and Herschel L Mitchell. Ensemble ka lman filtering. ��������� ������� �� ��� ����� �������������� �������� � ������� �� ��� ����������� ��������� ������� ����������� ��� �������� ������������, 131(613):3269–3289, 2005

    2005

  40. [40]

    W. W. Hsieh and B. Tang. Applying neural network models t o prediction and data analysis in meteorology and oceanography. �������� �� ��� �������� �������������� �������, 79(9):1855– 1870, 1998

    1998

  41. [41]

    Hu and P

    C.-C. Hu and P. J. van Leeuwen. A particle flow filter for hi gh-dimensional system applications. ����� , 147:2352–2374, 2021

    2021

  42. [42]

    B. R. Hunt, E. J. Kostelich, and I. Szunyogh. Efficient dat a assimilation for spatiotemporal chaos: A local ensemble transform kalman filter. ������� �� ��������� ��������� , 230(1- 2):112–126, 2007

    2007

  43. [43]

    Kolen J and S

    F. Kolen J and S. C. Kremer. �������� ���� �� ��������� ����� ��� �������� �� �������� �������� ������������ , pages 237–243. IEEE, 2001

    2001

  44. [44]

    A. Jiao, H. He, R. Ranade, J. Pathak, and Lu Lu. One-shot l earning for solution operators of partial differential equations. ���� ������� , 16:8386, 2025

    2025

  45. [45]

    R. E. Kalman. A new approach to linear filtering and predi ction problems. �� ����� ���� , 82(1):35–45, 1960

    1960

  46. [46]

    K. Kang, V. Maroulas, I. Schizas, and F. Bao. Improved di stributed particle filters for tracking in a wireless sensor network. ������� �������� ���� �����, 117:90–108, 2018

    2018

  47. [47]

    D. P. Kingma and J. Ba. Adam: A method for stochastic opti mization. In ����������� �� ��� ��� ������������� ���������� �� �������� ��������������� ������, Banff, 14-16 April 2014

    2014

  48. [48]

    ¨Uber die analytischen methoden in der wahrscheinlichkeits rechnung

    Andrei Kolmogoroff. ¨Uber die analytischen methoden in der wahrscheinlichkeits rechnung. ����� ��������� �������, pages 415 – 458, 1931. 24

    1931

  49. [49]

    N. B. Kovachki, S. Lanthaler, and A. M. Stuart. Operator learning: Algorithms and analysis. arXiv preprint, arXiv:2402.15715, 2024

    Pith/arXiv arXiv 2024

  50. [50]

    Tutorial: The kalman filter

    Tony Lacey. Tutorial: The kalman filter. ������� ��������� �� ����������, 139, 1998

    1998

  51. [51]

    Data ass imilation

    Kody Law, Andrew Stuart, and Kostas Zygalakis. Data ass imilation. ����� ������������ ��������, 214:52, 2015

    2015

  52. [52]

    Le and M

    V.-A. Le and M. Dik. A mathematical analysis of neural op erator behaviors. arXiv preprint arXiv:2410.21481, 2024

    arXiv 2024

  53. [53]

    LeCun, Y

    Y. LeCun, Y. Bengio, and G. Hinton. Deep learning. ������, (521):436, 2015

    2015

  54. [54]

    P. J. Van Leeuwen, H. R. K¨ unsch, L. Nerger, R. Potthast, and S. Reich. Particle filters for high- dimensional geoscience applications: A review. ��������� ������� �� ��� ����� �������������� �������, 145(723):2335–2365, 2019

    2019

  55. [55]

    Liang, H

    S. Liang, H. Tran, F. Bao, H. G. Chipilski, P. Jan van Leeu wen, and G. Zhang. Ensemble score filter with image inpainting for data assimilation in tracki ng surface quasi-geostrophic dynamics with partial observations. ����������������

  56. [56]

    Z. C. Lipton, J. Berkowitz, and C. Elkan. A critical revi ew of recurrent neural networks for sequence learning. ����� �������� ����������������, 2015

    2015

  57. [57]

    Y. Liu, M. Yang, Z. Zhang, Y. Cao, G. Zhang, and F. Bao. Diffu sion-model-assisted supervised learning of generative models for density estimation. ������� �� ������� �������� ��� �������� ��� ��������� , 5(1):25–38, 2024

    2024

  58. [58]

    E. N. Lorenz. Predictability: A problem partly solved. In ����� ������� �� ��������������, volume 1, pages 1–18. Reading, 1996

    1996

  59. [59]

    L. Lu, , G. Pang, P. Jin, Z. Zhang, and G. E. Karniadakis. L earning nonlinear operators via deeponet based on the universal approximation theorem o f operators. ���� ����� �������, 3:218–229, 2021

    2021

  60. [60]

    L. Lu, P. Jin, and G. E. Karniadakis. Deeponet: Learning nonlinear operators for identifying differential equations based on the universal approximation theorem of operators. arXiv preprint, arXiv:1910.03193, 2019

    Pith/arXiv arXiv 1910

  61. [61]

    L. Lu, X. Meng, S. Cai, Z. Mao, S. Goswami, Z. Zhang, and G. E. Karniadakis. A comprehensive and fair comparison of two neural operators (with practical extensions) based on fair data. �������� ������� �� ������� ��������� ��� ����������� , 393:114778, 2022

    2022

  62. [62]

    Lu Lu, X. Meng, Z. Mao, and G. E. Karniadakis. Deepxde: A d eep learning library for solving differential equations. ���� ������ , 63(1):208–228, 2021

    2021

  63. [63]

    S. Mao, A. M. Naser, S. N. Buoy, K. K. Brock, and K. A. Hutch eson. Optimizing modified barium swallow exam workflow: Automating pre-analysis vide o sorting in swallowing function assessment. In ���� ���� ������ ������������� ���������� �� ��� ���� ����������� �� �������� ��� ������� ������� ������ , pages 1–4. IEEE, 2024. 25

    2024

  64. [64]

    Neural operator learning for long-time integrat ion in dynamical systems with re- current neural networks

    Katarzyna Micha/suppress lowska, Somdatta Goswami, George Em Karniadakis, and Signe Riemer- Sørensen. Neural operator learning for long-time integrat ion in dynamical systems with re- current neural networks. In ���� ������������� ����� ���������� �� ������ �������� �������, pages 1–8. IEEE, 2024

    2024

  65. [65]

    Morzfeld, X

    M. Morzfeld, X. Tu, E. Atkins, and A. J. Chorin. A random m ap implementation of implicit filters. �� ������� ����� , 231(4):2049–2066, 2012

    2049

  66. [66]

    H. H. Nguyen and C. W. Chan. Multiple neural networks for a long term time series forecast. ������ ��������� � ������������, 13(1):90–98, 2004

    2004

  67. [67]

    D. W. Otter, J. R. Medina, and J. K. Kalita. A survey of the usages of deep learning for natural language processing. ���� ������������ �� ������ �������� ��� �������� ������� , 32(2):604–624, 2021

    2021

  68. [68]

    Pascanu, T

    R. Pascanu, T. Mikolov, and Y. Bengio. On the difficulty of training recurrent neural networks. In ������������� ���������� �� ������� ��������, pages 1310–1318. Pmlr, 2013

    2013

  69. [69]

    Pathak, B

    J. Pathak, B. Hunt, M. Girvan, Z. Lu, and E. Ott. Model-fr ee prediction of large spatiotempo- rally chaotic systems from data: A reservoir computing appr oach. ����� ���� �����, 120:024102, 1 2018

    2018

  70. [70]

    M. K. Pitt and N. Shephard. Filtering via simulation: Au xiliary particle filters. ������� �� ��� �������� ����������� �����������, 94(446):590–599, 1999

    1999

  71. [71]

    Radosavovic, R

    I. Radosavovic, R. P. Kosaraju, R. Girshick, K. He, and P . Doll´ ar. Designing network design spaces. In ���� �������� ���������� �� �������� ������ ��� ������� ������� ���� ������ , pages 10425–10433, 2020

    2020

  72. [72]

    Raissi, P

    M. Raissi, P. Perdikaris, and G. E. Karniadakis. Physic s-informed neural networks: A deep learning framework for solving forward and inverse problem s involving nonlinear partial differ- ential equations. ������� �� ������������� �������, 378:686–707, 2019

    2019

  73. [73]

    Fundamentals of recurrent neural ne twork (rnn) and long short-term memory (lstm) network

    Alex Sherstinsky. Fundamentals of recurrent neural ne twork (rnn) and long short-term memory (lstm) network. ������� �� ��������� ��������� , 404:132306, 2020

    2020

  74. [74]

    Snyder, T

    C. Snyder, T. Bengtsson, P. Bickel, and J. Anderson. Obs tacles to high-dimensional particle filtering. ���� ���� ���� , 136:4629–4640, 2008

    2008

  75. [75]

    Performance bounds for particle filters using the optimal proposal

    Chris Snyder, Thomas Bengtsson, and Mathias Morzfeld. Performance bounds for particle filters using the optimal proposal. ������� ������� ������ , 143(11):4750–4761, 2015

    2015

  76. [76]

    Deep unsuper- vised learning using nonequilibrium thermodynamics

    Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranat han, and Surya Ganguli. Deep unsuper- vised learning using nonequilibrium thermodynamics. In ������������� ���������� �� ������� ��������, pages 2256–2265. pmlr, 2015

    2015

  77. [77]

    Song and S

    Y. Song and S. Ermon. Generative modeling by estimating gradients of the data distribution. In ����� ���� ���� , pages 11918–11930, Vancouver, Canada, Dec. 2019

    2019

  78. [78]

    Y. Song, J. Sohl-Dickstein, D. P Kingma, A. Kumar, S. Erm on, and B. Poole. Score-based generative modeling through stochastic differential equati ons. In ������������� ���������� �� �������� ���������������, 2021. 26

    2021

  79. [79]

    Tan and Q

    M. Tan and Q. Le. Efficientnet: Rethinking model scaling f or convolutional neural networks. In ������������� ���������� �� ������� ��������, pages 6105–6114. PMLR, 2019

    2019

  80. [80]

    Tsubaki, K

    M. Tsubaki, K. Tomii, and J. Sese. Compound–protein int eraction prediction with end-to-end learning of neural networks for graphs and sequences. ��������������, 35(2):309–318, 2019

    2019

Showing first 80 references.