Pith. sign in

REVIEW 2 major objections 1 minor 300 references

A segment-wise rejection sampling method prevents hallucination snowballing in long-form LLM outputs.

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

2026-06-28 10:01 UTC pith:CK4LHPZI

load-bearing objection SHARS introduces segment-level rejection sampling with adapted semantic uncertainty to curb hallucination snowballing, but the abstract supplies no numbers or modification details to evaluate the claim. the 2 major comments →

arxiv 2606.03628 v1 pith:CK4LHPZI submitted 2026-06-02 cs.CL cs.AIcs.LG

Building Reliable Long-Form Generation via Hallucination Rejection Sampling

Lin Li , Georgia Channing , Suhaas M Bhat , Gabriel Davis Jones , Yarin Gal This is my paper
classification cs.CL cs.AIcs.LG
keywords hallucination mitigationlong-form generationrejection samplingsemantic uncertaintylarge language modelsfactual consistencyinference-time methods
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.

Large language models often produce hallucinations that compound across long generations because early errors influence later ones. The paper proposes SHARS to detect hallucinated segments using semantic uncertainty, reject them, and resample until each segment is faithful before continuing. This builds the full output only on confident content. A sympathetic reader would care because the approach lets models self-correct during generation without external knowledge sources.

Core claim

The authors claim that their Segment-wise Hallucination Rejection Sampling framework, instantiated with a modified semantic uncertainty detector, identifies hallucinated segments during generation, rejects them, and resamples until faithful segments are produced, thereby mitigating error propagation and enhancing factual consistency in long-form text while preserving informativeness.

What carries the argument

The SHARS framework that applies a hallucination detector to individual segments and resamples rejected ones to prevent accumulation of errors.

Load-bearing premise

Semantic uncertainty after the modifications serves as a reliable detector that rejects hallucinations without discarding valid content or letting errors through.

What would settle it

Observing no reduction in hallucination metrics or a drop in informativeness when applying the method to benchmark tasks compared to standard sampling.

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

If this is right

  • Long-form generations have reduced hallucination rates on standard benchmarks.
  • The method maintains or improves the amount of informative content produced.
  • It operates without external resources like search engines.
  • It is compatible with various hallucination detectors for future use.

Where Pith is reading between the lines

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

  • This approach might apply to other error-prone sequential tasks such as multi-step reasoning.
  • Further work could explore optimal segment lengths for the detector.
  • Integration with training-time methods could compound the benefits.

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 manuscript proposes Segment-wise HAllucination Rejection Sampling (SHARS), an inference-time framework that detects hallucinated segments in long-form LLM outputs using a modified semantic uncertainty detector, rejects them, and resamples until faithful content is obtained. This is intended to prevent hallucination snowballing. The abstract claims that empirical evaluations on standardized hallucination benchmarks show substantial reductions in hallucinations while preserving or improving informativeness, without needing external resources.

Significance. If the central empirical claims hold and the detector modifications are validated, the work could provide a practical self-correction method for long-form generation reliability. The provision of code at the GitHub link supports reproducibility.

major comments (2)
  1. [Abstract] Abstract: the claim that 'vital modifications' to semantic uncertainty address its limitations is unsupported by any description of those modifications or validation that they raise uncertainty scores on consistently generated hallucinations (the core stress-test concern). This is load-bearing for the detector's reliability at segment level.
  2. [Evaluation] Evaluation section: the abstract asserts 'substantial' reductions but supplies no quantitative results, baselines, dataset details, or ablations; without these, the central claim that the method reduces hallucinations while preserving informativeness cannot be assessed.
minor comments (1)
  1. The method is presented as compatible with external resources for future extensions; a brief discussion of how this compatibility would work would improve clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback highlighting issues with the abstract and evaluation claims. We address each point below and will revise the manuscript to strengthen clarity and support for the central claims.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that 'vital modifications' to semantic uncertainty address its limitations is unsupported by any description of those modifications or validation that they raise uncertainty scores on consistently generated hallucinations (the core stress-test concern). This is load-bearing for the detector's reliability at segment level.

    Authors: We agree the abstract lacks detail on the modifications and their validation. The full manuscript (Section 3.2) describes the adaptations: computing semantic uncertainty at the segment level rather than token level, using a calibrated threshold based on segment length, and incorporating consistency checks across multiple samples. However, we did not include a dedicated stress-test validation showing elevated uncertainty scores specifically on consistently generated hallucinations. We will revise the abstract to briefly summarize the key modifications and add a targeted validation experiment or ablation in the revised manuscript. revision: yes

  2. Referee: [Evaluation] Evaluation section: the abstract asserts 'substantial' reductions but supplies no quantitative results, baselines, dataset details, or ablations; without these, the central claim that the method reduces hallucinations while preserving informativeness cannot be assessed.

    Authors: The evaluation section reports results on standardized hallucination benchmarks, but we acknowledge it does not sufficiently detail quantitative numbers, exact baselines, dataset splits, or ablations in a way that allows full assessment from the abstract alone. We will expand the evaluation section in the revision to include specific quantitative reductions, baseline comparisons, dataset descriptions, and ablation studies demonstrating the contribution of each component while preserving informativeness. revision: yes

Circularity Check

0 steps flagged

No circularity: inference-time procedure defined independently of inputs

full rationale

The paper proposes SHARS as an inference-time framework that applies any hallucination detector to identify and reject segments during generation, then resamples until faithful content is obtained. Semantic uncertainty is instantiated as the detector after unspecified modifications for long-form use, but the abstract and description contain no equations, fitted parameters, or self-citations that reduce any claim to its own inputs by construction. The central steps (segment detection, rejection, resampling) are procedurally specified and evaluated on external benchmarks, with no renaming of known results, uniqueness theorems imported from the authors, or predictions that are statistically forced by the fitting process. This matches the reader's assessment of minimal circularity risk.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Abstract-only review provides no explicit free parameters, invented entities, or detailed axioms beyond the core domain assumption that a detector can identify hallucinated segments.

axioms (1)
  • domain assumption Semantic uncertainty after modifications can serve as an effective hallucination detector for text segments
    The framework is instantiated by adopting semantic uncertainty as the detector.

pith-pipeline@v0.9.1-grok · 5764 in / 1177 out tokens · 34018 ms · 2026-06-28T10:01:12.452706+00:00 · methodology

0 comments
read the original abstract

Large language models (LLMs) have achieved remarkable progress in open-ended text generation, yet they remain prone to hallucinating incorrect or unsupported content, which undermines their reliability. This issue is exacerbated in long-form generation due to hallucination snowballing, a phenomenon where early errors propagate and compound into subsequent outputs. To address this challenge, we propose a novel inference-time hallucination mitigation framework, named Segment-wise HAllucination Rejection Sampling (SHARS), which uses an arbitrary hallucination detector to identify and reject hallucinated segments during generation and resample until faithful content is produced. By retaining only confident information and building subsequent generations upon it, the framework mitigates hallucination accumulation and enhances factual consistency. To instantiate this framework, we adopt semantic uncertainty as the detector and introduce several vital modifications to address its limitations and better adapt it to long-form text. Our method enables models to self-correct hallucinations without requiring external resources such as web search or knowledge bases, while remaining compatible with them for future extensions. Empirical evaluations on standardized hallucination benchmarks demonstrate that our method substantially reduces hallucinations in long-form generation while preserving or even improving the informativeness of generation. Code is available at: https://github.com/TreeLLi/hallucination-rejection-sampling.

Figures

Figures reproduced from arXiv: 2606.03628 by Gabriel Davis Jones, Georgia Channing, Lin Li, Suhaas M Bhat, Yarin Gal.

Figure 1
Figure 1. Figure 1: (a) Comparison of biographies generated by Greedy decoding and our method. Unlike Greedy decoding, our method rejects hallucinated content, preserves factual information, and acquires additional factual content (the last two sentences in the displayed generation) beyond the original information space. (b) and (c) Scaling of factual precision with respect to inference-time computation on the FactScore bench… view at source ↗
Figure 2
Figure 2. Figure 2: Illustration of naive long-form semantic entropy method and our proposed HalluSE. Different colors under ‘Answer Sampling & Response’ denote distinct semantic clusters of generated responses. A green check indicates low semantic entropy (high agreement, reliable answers), while a red cross marks high semantic uncertainty (likely hallucinated content). be found in Farquhar et al. (2024). Low semantic entrop… view at source ↗
Figure 3
Figure 3. Figure 3: The change of response rate and factual precision with the semantic entropy threshold θ [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The change of response rate and factual precision with the number of answers, A. sequent outputs upon verified information. Together with HalluSE, our improved uncertainty-based detection method, SHARS provides an effective and flexible approach to mit￾igating hallucinations while maintaining or enhancing in￾formativeness. Extensive evaluations across multiple long￾form factuality benchmarks demonstrate th… view at source ↗
Figure 5
Figure 5. Figure 5: Set of prompts for various parts of the pipeline. 14 [PITH_FULL_IMAGE:figures/full_fig_p014_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Additional prompt if rewrite is enabled. 15 [PITH_FULL_IMAGE:figures/full_fig_p015_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

300 extracted references · 1 canonical work pages · 1 internal anchor

  1. [1]

    The Llama 3 Herd of Models

    The llama 3 herd of models , author=. arXiv preprint arXiv:2407.21783 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  2. [2]

    2016 , eprint=

    Neural Text Generation from Structured Data with Application to the Biography Domain , author=. 2016 , eprint=

    2016

  3. [3]

    Advances in neural information processing systems , volume=

    Retrieval-augmented generation for knowledge-intensive nlp tasks , author=. Advances in neural information processing systems , volume=

  4. [4]

    Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , pages=

    Forag: Factuality-optimized retrieval augmented generation for web-enhanced long-form question answering , author=. Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , pages=

  5. [5]

    SWE-bench: Can Language Models Resolve Real-world Github Issues? , author=

  6. [6]

    Proceedings of the 41st International Conference on Machine Learning , pages=

    How language model hallucinations can snowball , author=. Proceedings of the 41st International Conference on Machine Learning , pages=

  7. [7]

    and Steiner, Benoit and Tucker, Paul and Vasudevan, Vijay and Warden, Pete and Wicke, Martin and Yu, Yuan and Zheng, Xiaoqiang , journal =

    Abadi, Martin and Barham, Paul and Chen, Jianmin and Chen, Zhifeng and Davis, Andy and Dean, Jeffrey and Devin, Matthieu and Ghemawat, Sanjay and Irving, Geoffrey and Isard, Michael and Kudlur, Manjunath and Levenberg, Josh and Monga, Rajat and Moore, Sherry and Murray, Derek G. and Steiner, Benoit and Tucker, Paul and Vasudevan, Vijay and Warden, Pete an...

  8. [8]

    Interpreting

    Abbasi-Asl, Reza and Yu, Bin , booktitle =. Interpreting

  9. [9]

    and Awad, Mohammed Ibrahim and Munim, Hossam E

    AbdElKhalek, Youssef M. and Awad, Mohammed Ibrahim and Munim, Hossam E. Abd El and Maged, Shady A. , journal =. Trajectory-based fast ball detection and tracking for an autonomous industrial robot system , year =

  10. [10]

    Are you still on track!?

    Abdelnabi, Sahar and Fay, Aideen and Cherubin, Giovanni and Salem, Ahmed and Fritz, Mario and Paverd, Andrew , journal =. Are you still on track!?

  11. [11]

    Abdelnabi, Sahar and Greshake, Kai and Mishra, Shailesh and Endres, Christoph and Holz, Thorsten and Fritz, Mario , booktitle =. Not

  12. [12]

    Large language models associate

    Abid, Abubakar and Farooqi, Maheen and Zou, James , journal =. Large language models associate

  13. [13]

    Emergence of invariance and disentanglement in deep representations , year =

    Achille, Alessandro and Soatto, Stefano , journal =. Emergence of invariance and disentanglement in deep representations , year =

  14. [14]

    Acikgoz, Emre Can and Qian, Cheng and Ji, Heng and Hakkani-Tür, Dilek and Tur, Gokhan , journal =. Self-

  15. [15]

    Differentially

    Acs, Gergely and Melis, Luca and Castelluccia, Claude and De Cristofaro, Emiliano , journal =. Differentially

  16. [16]

    Adadi, Amina and Berrada, Mohammed , journal =. Peeking

  17. [17]

    Forecasting the

    Adämmer, Philipp and Schüssler, Rainer A , journal =. Forecasting the

  18. [18]

    Addepalli, Sravanti and Varun, Yerram and Suggala, Arun and Shanmugam, Karthikeyan and Jain, Prateek , booktitle =. Does

  19. [19]

    Efficient and

    Addepalli, Sravanti and Jain, Samyak and Radhakrishnan, Venkatesh Babu , booktitle =. Efficient and

  20. [20]

    , booktitle =

    Addepalli, Sravanti and Jain, Samyak and Sriramanan, Gaurang and Venkatesh Babu, R. , booktitle =. Scaling

  21. [21]

    Debugging

    Adebayo, Julius and Muelly, Michael and Liccardi, Ilaria and Kim, Been , booktitle =. Debugging

  22. [22]

    Post hoc

    Adebayo, Julius and Muelly, Michael and Abelson, Harold and Kim, Been , journal =. Post hoc

  23. [23]

    and Nix, Tionney and Rybeck, Gabriel and Scheidegger, Carlos and Smith, Brandon and Venkatasubramanian, Suresh , journal =

    Adler, Philip and Falk, Casey and Friedler, Sorelle A. and Nix, Tionney and Rybeck, Gabriel and Scheidegger, Carlos and Smith, Brandon and Venkatasubramanian, Suresh , journal =. Auditing black-box models for indirect influence , year =

  24. [24]

    Afchar, Darius and Guigue, Vincent and Hennequin, Romain , booktitle =. Towards

  25. [25]

    Cognitive data augmentation for adversarial defense via pixel masking , year =

    Agarwal, Akshay and Vatsa, Mayank and Singh, Richa and Ratha, Nalini , journal =. Cognitive data augmentation for adversarial defense via pixel masking , year =

  26. [26]

    Agarwal, Rishabh and Melnick, Levi and Frosst, Nicholas and Zhang, Xuezhou and Lengerich, Ben and Caruana, Rich and Hinton, Geoffrey , journal =. Neural

  27. [27]

    Agarwal, Rishabh and Melnick, Levi and Frosst, Nicholas and Zhang, Xuezhou and Lengerich, Ben and Caruana, Rich and Hinton, Geoffrey E , booktitle =. Neural

  28. [28]

    Agarwal, Shivam and Zhang, Zimin and Yuan, Lifan and Han, Jiawei and Peng, Hao , journal =. The

  29. [29]

    Agashe, Saaket and Han, Jiuzhou and Gan, Shuyu and Yang, Jiachen and Li, Ang and Wang, Xin Eric , journal =. Agent

  30. [30]

    Aggarwal, Gunjan and Sinha, Abhishek and Kumari, Nupur and Singh, Mayank , journal =. On the

  31. [31]

    Analyzing the

    Agrawal, Pulkit and Girshick, Ross and Malik, Jitendra , booktitle =. Analyzing the

  32. [32]

    and Goldfarb, Avi , journal =

    Agrawal, Ajay and Gans, Joshua S. and Goldfarb, Avi , journal =. Artificial

  33. [33]

    Agrawal, Garima and Kumarage, Tharindu and Alghamdi, Zeyad and Liu, Huan , booktitle =. Can

  34. [34]

    Agrawal, Aishwarya and Batra, Dhruv and Parikh, Devi and Kembhavi, Aniruddha , journal =. Don't

  35. [35]

    Agrawal, Ayush and Suzgun, Mirac and Mackey, Lester and Kalai, Adam , booktitle =. Do

  36. [36]

    , booktitle =

    Ahdritz, Gustaf and Qin, Tian and Vyas, Nikhil and Barak, Boaz and Edelman, Benjamin L. , booktitle =. Distinguishing the

  37. [37]

    and Arriaga, Rosa I

    Aher, Gati V. and Arriaga, Rosa I. and Kalai, Adam Tauman , booktitle =. Using

  38. [38]

    Improving

    Aichberger, Lukas and Schweighofer, Kajetan and Ielanskyi, Mykyta and Hochreiter, Sepp , booktitle =. Improving

  39. [39]

    Aichberger, Lukas and Paren, Alasdair and Li, Guohao and Torr, Philip and Gal, Yarin and Bibi, Adel , journal =

  40. [40]

    Rethinking

    Aichberger, Lukas and Schweighofer, Kajetan and Hochreiter, Sepp , booktitle =. Rethinking

  41. [41]

    and Maini, Pratyush and Lipton, Zachary Chase and Kolter, J

    Aithal, Sumukh K. and Maini, Pratyush and Lipton, Zachary Chase and Kolter, J. Zico , journal =. Understanding

  42. [42]

    Advances in

    Akhtar, Naveed and Mian, Ajmal and Kardan, Navid and Shah, Mubarak , journal =. Advances in

  43. [43]

    Adversarial

    Al-Maliki, Shawqi and Qayyum, Adnan and Ali, Hassan and Abdallah, Mohamed and Qadir, Junaid and Hoang, Dinh Thai and Niyato, Dusit and Al-Fuqaha, Ala , journal =. Adversarial

  44. [44]

    and Schaar, Mihaela van der , booktitle =

    Alaa, Ahmed and Breugel, Boris Van and Saveliev, Evgeny S. and Schaar, Mihaela van der , booktitle =. How

  45. [45]

    and Neyshabur, Behnam and Zhai, Xiaohua , booktitle =

    Alabdulmohsin, Ibrahim M. and Neyshabur, Behnam and Zhai, Xiaohua , booktitle =. Revisiting

  46. [46]

    Alayrac, Jean-Baptiste and Uesato, Jonathan and Huang, Po-Sen and Fawzi, Alhussein and Stanforth, Robert and Kohli, Pushmeet , booktitle =. Are

  47. [47]

    Flamingo: a

    Alayrac, Jean-Baptiste and Donahue, Jeff and Luc, Pauline and Miech, Antoine and Barr, Iain and Hasson, Yana and Lenc, Karel and Mensch, Arthur and Millican, Katie and Reynolds, Malcolm and Ring, Roman and Rutherford, Eliza and Cabi, Serkan and Han, Tengda and Gong, Zhitao and Samangooei, Sina and Monteiro, Marianne and Menick, Jacob and Borgeaud, Sebasti...

  48. [48]

    Adversarial example detection for

    Aldahdooh, Ahmed and Hamidouche, Wassim and Fezza, Sid Ahmed and Déforges, Olivier , journal =. Adversarial example detection for

  49. [49]

    News and

    Aleti, Saketh and Bollerslev, Tim , journal =. News and

  50. [50]

    What is an object? , year =

    Alexe, Bogdan and Deselaers, Thomas and Ferrari, Vittorio , booktitle =. What is an object? , year =

  51. [51]

    and Thabet, Ali and Bibi, Adel and Torr, Philip H

    Alfarra, Motasem and Perez, Juan C. and Thabet, Ali and Bibi, Adel and Torr, Philip H. S. and Ghanem, Bernard , booktitle =. Combating

  52. [52]

    Alfarra, Motasem and Bibi, Adel and Torr, Philip H. S. and Ghanem, Bernard , booktitle =. Data dependent randomized smoothing , year =

  53. [53]

    Alfarra, Motasem and Bibi, Adel and Khan, Naeemullah and Torr, Philip H. S. and Ghanem, Bernard , booktitle =

  54. [54]

    Superintelligence

    Alfonseca, Manuel and Cebrian, Manuel and Anta, Antonio Fernandez and Coviello, Lorenzo and Abeliuk, Andrés and Rahwan, Iyad , journal =. Superintelligence

  55. [55]

    Generalizability of

    Alhamoud, Kumail and Hammoud, Hasan Abed Al Kader and Alfarra, Motasem and Ghanem, Bernard , journal =. Generalizability of

  56. [56]

    Entropy-

    Ali, Riccardo and Caso, Francesco and Irwin, Christopher and Liò, Pietro , journal =. Entropy-

  57. [57]

    Allen-Zhu, Zeyuan and Li, Yuanzhi and Song, Zhao , booktitle =. A

  58. [58]

    Allen-Zhu, Zeyuan and Li, Yuanzhi , journal =. Feature

  59. [59]

    , journal =

    Alom, Md Zahangir and Hasan, Mahmudul and Yakopcic, Chris and Taha, Tarek M. , journal =. Inception

  60. [60]

    Mind the

    Alsallakh, Bilal and Kokhlikyan, Narine and Miglani, Vivek and Yuan, Jun and Reblitz-Richardson, Orion , journal =. Mind the

  61. [61]

    and Restificar, Angelo C

    Altendorf, Eric E. and Restificar, Angelo C. and Dietterich, Thomas G. , booktitle =. Learning from sparse data by exploiting monotonicity constraints , year =

  62. [62]

    A causal framework for explaining the predictions of black-box sequence-to-sequence models , year =

    Alvarez-Melis, David and Jaakkola, Tommi , booktitle =. A causal framework for explaining the predictions of black-box sequence-to-sequence models , year =

  63. [63]

    , booktitle =

    Alvarez-Melis, David and Jaakkola, Tommi S. , booktitle =. On the

  64. [64]

    , booktitle =

    Alzantot, Moustafa and Sharma, Yash and Chakraborty, Supriyo and Zhang, Huan and Hsieh, Cho-Jui and Srivastava, Mani B. , booktitle =

  65. [65]

    Generating

    Alzantot, Moustafa and Sharma, Yash and Elgohary, Ahmed and Ho, Bo-Jhang and Srivastava, Mani and Chang, Kai-Wei , booktitle =. Generating

  66. [66]

    Understanding and

    Amir, Dan and Weiss, Yair , journal =. Understanding and

  67. [67]

    Concrete

    Amodei, Dario and Olah, Chris and Steinhardt, Jacob and Christiano, Paul and Schulman, John and Mané, Dan , journal =. Concrete

  68. [68]

    Amorim, J. P. and Abreu, P. H. and Reyes, M. and Santos, J. , booktitle =. Interpretability vs

  69. [69]

    An, Xiao and Sun, Jiaxing and Gui, Zihan and He, Wei , booktitle =

  70. [70]

    Modeling human decisions in coupled human and natural systems:

    An, Li , journal =. Modeling human decisions in coupled human and natural systems:

  71. [71]

    Gradient-

    Ancona, Marco and Ceolini, Enea and Öztireli, Cengiz and Gross, Markus , booktitle =. Gradient-

  72. [72]

    Towards better understanding of gradient-based attribution methods for

    Ancona, Marco and Ceolini, Enea and Öztireli, Cengiz and Gross, Markus , journal =. Towards better understanding of gradient-based attribution methods for

  73. [73]

    Zico and Fredrikson, Matt and Gal, Yarin and Davies, Xander , journal =

    Andriushchenko, Maksym and Souly, Alexandra and Dziemian, Mateusz and Duenas, Derek and Lin, Maxwell and Wang, Justin and Hendrycks, Dan and Zou, Andy and Kolter, J. Zico and Fredrikson, Matt and Gal, Yarin and Davies, Xander , journal =

  74. [74]

    Andriushchenko, Maksym and Flammarion, Nicolas , journal =. Does

  75. [75]

    Jailbreaking

    Andriushchenko, Maksym and Croce, Francesco and Flammarion, Nicolas , booktitle =. Jailbreaking

  76. [76]

    Andriushchenko, Maksym and Croce, Francesco and Flammarion, Nicolas and Hein, Matthias , booktitle =. Square

  77. [77]

    Understanding and

    Andriushchenko, Maksym and Flammarion, Nicolas , booktitle =. Understanding and

  78. [78]

    Ang (Ming), Gary , journal =

  79. [79]

    and Bates, Stephen , journal =

    Angelopoulos, Anastasios N. and Bates, Stephen , journal =. A

  80. [80]

    , booktitle =

    Angelucci, Alessandra and Bressloff, Paul C. , booktitle =. Contribution of feedforward, lateral and feedback connections to the classical receptive field center and extra-classical receptive field surround of primate

Showing first 80 references.