Pith. sign in

REVIEW 4 major objections 7 minor 92 references

Advancing Decoding Strategies: Enhancements in Locally Typical Sampling for LLMs

T0 review · 4 major / 7 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Adaptive Semantic-Aware Typicality Sampling is claimed to beat existing decoding methods on repetition, semantic alignment, and fluency.

desk verdict A well-meaning but unreproducible proposal: the ASTS algorithm is internally inconsistent in its own worked example, and the empirical claims rest on unspecified components. read the letter →

arxiv 2506.05387 v2 pith:7VXXFDZR submitted 2025-06-03 cs.CL cs.AI

Jaydip Sen , Saptarshi Sengupta , Subhasis Dasgupta This is my paper
classification cs.CLcs.AI
keywords locallytypicalsamplingadaptivesemantic-awaretypicalitylargelanguagemodelsdecodingstrategiesentropy-basedmulti-objectivescoringrepetitionreductiontextgeneration
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Large language models generate text by sampling each next token from a probability distribution, and the choice rule decides whether the output is fluent, varied, or repetitive. The paper argues that locally typical sampling, which admits tokens whose surprise matches the local entropy, can be improved by making the entropy window dynamic and by re-weighting tokens according to semantic fit, diversity, and repetition. The proposed rule, ASTS, is claimed to generate stories and summaries closer to human-written text: lower repetition, better MAUVE scores, and perplexity nearer the human reference. A sympathetic reader would care because the improvement is purely in the decoding step and requires no model retraining.

What carries the argument

The load-bearing object is the modified sampling distribution over the locally typical set $R$. The original rule keeps a token $x_t$ when $|-\log p(x_t) - H| \le \varepsilon$; ASTS replaces the fixed $\varepsilon$ with time-dependent bounds $\alpha_t = H - k_1 \sigma_H$ and $\beta_t = H + k_2 \sigma_H$, where $\sigma_H$ is the standard deviation of entropy over previous tokens. Each eligible token gets a composite score $S(x_t) = \lambda_1 C(x_t) + \lambda_2 SA(x_t) + \lambda_3 D(x_t)$ combining typicality, cosine-based semantic alignment with the context, and inverse-frequency diversity. The probabilities are then reweighted by $P'(x_t) = P(x_t)\exp(R(x_t) - P(x_t))$, with reward $R(x_t) = \mu_1 SA(x_t) + \mu_2 Relv(x_t) - \mu_3 RP(x_t)$, normalized, temperature-scaled, and finally sampled. The claim is that these reweightings, not just the entropy window, are what push generated text closer to human writing.

What would settle it

Re-run ASTS and locally typical sampling on the WritingPrompts story and CNN/DailyMail summarization benchmarks with every hyperparameter, embedding model, and relevance function fixed before looking at the test set, then compare MAUVE and REP scores. If the advantages over locally typical sampling shrink or disappear under these fixed conditions, the central claim fails; if they persist, it is supported.

Watch

Extended reading notes

Core claim

The paper sets out to establish that ASTS, a modified locally typical sampling rule, outperforms existing sampling strategies on story generation and abstractive summarization. In the reported results, ASTS obtains the highest MAUVE (0.97 on story, 0.99 on summarization), the lowest repetition scores (0.25 and 0.12), and perplexity values closest to human-written reference text, on both benchmarks. The claim is that the mechanism, adaptive entropy thresholds, multi-objective scoring, and reward-penalty adjustment, is responsible for these gains, and that it does so while remaining robust to the hyperparameter choices that destabilize top-k, nucleus, and even locally typical sampling.

Load-bearing premise

The central claim collapses if the reported gains were produced by unstated choices in how semantic alignment and relevance scores are computed, rather than by the ASTS mechanism itself, because the paper never defines those components.

Editorial extensions

If this is right

  • If the reported results are correct, switching a deployed LLM from locally typical sampling to ASTS lowers the repetition score and raises MAUVE without changing the model weights.
  • The robustness to the hyperparameter $\tau$ reported for ASTS means the method would need less per-task tuning than top-k, nucleus, or locally typical sampling in practice.
  • The closeness of ASTS perplexity to human reference values would make fluency-based discrimination between human and machine text harder, provided that closeness is not an artifact of unreported tuning.
  • ASTS should transfer to other open-ended generation tasks such as dialogue and story continuation, since the mechanism is task-agnostic, though the paper only tests story generation and summarization.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper never specifies the embedding model, the context-embedding construction, the relevance function, or the experimental values of $k_1, k_2, \lambda_1, \lambda_2, \lambda_3, \mu_1, \mu_2, \mu_3$, and $T$; if those were tuned on the same test benchmarks, the closeness of ASTS's perplexity and Zipf values to human references would be expected rather than predictive.
  • Because the reward-penalty step exponentiates $R(x_t) - P(x_t)$, the final distribution depends on the raw probability scale in a way the paper does not analyze, so a formal derivation of the induced distribution would clarify when the adjustment is well behaved.
  • A component-ablation study, using dynamic thresholds only, semantic alignment only, and repetition penalty only on the same benchmarks, would show which piece actually drives the reported gains.
  • A natural stress test is to apply ASTS to constrained domains such as legal or medical text, where semantic alignment and relevance weighting must obey domain-specific constraints; the authors list this as future work, but it is also the fastest way to see whether the mechanism generalizes.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 7 minor

Summary. The paper proposes Adaptive Semantic-Aware Typicality Sampling (ASTS), an extension of Locally Typical Sampling for LLM decoding. ASTS combines dynamic entropy thresholds, a composite objective score over coherence, diversity, and semantic alignment, and a reward-penalty probability adjustment. The authors report experiments on story generation (GPT-2 on WritingPrompts) and abstractive summarization (BART on CNN/DailyMail), claiming that ASTS outperforms existing sampling strategies on perplexity closeness to human text, MAUVE, repetition, Zipf coefficient, and diversity. The central claim is an empirical superiority result presented in Tables 5 and 6.

Significance. If substantiated, an improved sampling scheme that reduces repetition while increasing semantic alignment would be practically useful for open-ended generation and summarization. The paper identifies a real limitation of Locally Typical Sampling and its proposed direction is plausible. However, the significance as presented is severely weakened by the fact that the algorithm is not fully specified and the reported experiments lack the reproducibility infrastructure (code, seeds, variance, hyperparameter values) that would let a reader verify the claimed gains. The paper's strengths are its explicit metric definitions and its acknowledgment of limitations in Section 6, but these do not compensate for the missing method details.

major comments (4)
  1. [Section 4.6, Step 3; Eq. (13)] The algorithm is internally inconsistent. Equation (13) defines the adjusted probability as P'(x) = P(x)·exp(R(x) − P(x)), while the worked example in Section 4.6, Step 3, computes P'(x) = P(x)·exp(S(x) + R(x)). These are different formulas: one subtracts the original probability inside the exponent, the other adds the composite score S(x). Because Eq. (13) is the operative definition of the method, the ASTS algorithm is ambiguous as written and cannot be reproduced without deciding which formula is intended.
  2. [Section 4.6, Table 3] The arithmetic in the worked example is incorrect even under the example's own formula. For "analyze", 0.175·exp(0.89 + 0.83) = 0.175·exp(1.72) ≈ 0.98, not the reported 0.45. The other rows show similar discrepancies. Consequently, the normalization factor 1.63 and the final sampling probabilities in Table 4 are invalid. This is not a cosmetic issue; it undermines the only concrete illustration of how ASTS is supposed to work.
  3. [Sections 4.2–4.3, Eq. (11), Eq. (14), Eq. (15)] The semantic components that distinguish ASTS from Locally Typical Sampling are never defined. Eq. (11) and Eq. (15) use a token embedding E(x_t) and a contextual embedding E(Y_<t), and Eq. (14) uses a relevance term Relv(x_t), but the paper gives no embedding model, no procedure for computing the context embedding, and no definition of relevance. Since these components drive the reward-penalty adjustment, the reported results cannot be independently implemented or checked.
  4. [Section 5.2, Tables 5 and 6; Section 6] The abstract and the conclusion claim that ASTS outperforms top-k and nucleus sampling, but Tables 5 and 6 compare ASTS only against Mirostat and Locally Typical Sampling variants. No experimental results for top-k or nucleus sampling are reported anywhere, so the claimed superiority over those baselines is unsupported. Moreover, the paper does not report values for the ASTS hyperparameters k1, k2, λ1–λ3, μ1–μ3, or temperature T, and no code or seeds are provided; Section 5's statement that the "same experimental setup" as [9] was followed does not supply these missing details. It is therefore unclear whether the hyperparameters were tuned on the test benchmarks, which would make the closeness of ASTS's perplexity and Zipf values to human references a tuning artifact rather than a predictive result.
minor comments (7)
  1. [Section 4.1, heading] The heading reads "Dynamic Entropy Thresholds" but the body text says "Locally Typical Samling" in the opening sentence; this typo should be corrected.
  2. [Section 4.5] There are two subsections numbered 4.5: "Temperature Scaling" and "Sampling the Next Token". The latter should be renumbered.
  3. [Tables] Table 1 appears twice: once as the Section 2 summary of decoding techniques and again in Section 4.6 for composite scores. The tables should be renumbered sequentially throughout the manuscript.
  4. [Section 5.1] The metric name is spelled inconsistently as "Zipf's score", "Zip's coefficient", and "Zipf coefficient"; please standardize.
  5. [Figures 1–2 captions] The captions read "ASTS if found to be least sensitive"; this should be "ASTS is found".
  6. [Eq. (18)] In Eq. (18), the summation in the denominator is written as ∑(x|Y<t)^(1/T) without the probability argument; it should be ∑ P_R'(x|Y<t)^(1/T) over x∈R.
  7. [Section 6, Limitations] The limitations list computational complexity and domain-specific performance, but it does not acknowledge the missing reproducibility items (code, hyperparameter values, and definitions of the embedding and relevance functions), which are the most immediate obstacles to verification.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: ASTS's performance claims are empirical evaluations against external benchmarks, not derivations from its own inputs; the paper's serious defects are reproducibility gaps, not circular reasoning.

full rationale

No circular step can be identified from the paper's equations or citations. The central claim is an empirical superiority claim: ASTS is evaluated on external benchmarks (WritingPrompts, CNN/DailyMail) with metrics PPL, MAUVE, REP, Zipf, and diversity, and the reported numbers are presented as measurements, not as derivations from the method's objective. The fact that ASTS explicitly contains a diversity reward and a repetition penalty (Eqs. 10 and 16) makes low REP and high diversity consistent with the design, but reporting those outcomes is validation, not circularity. The reuse of Meister et al.'s experimental setup is an external comparison, and no load-bearing premise is justified by the authors' own prior work or by a self-citation chain. The statement that hyperparameters are 'tuned based on the specific task' (Section 4.2) is not accompanied by values or a protocol, so any claim that the favorable PPL, MAUVE, or Zipf numbers are fitted to the test benchmarks would be speculation; the hard rules require exhibiting a specific reduction, which is not present. The genuine problems in the manuscript are reproducibility and correctness: Eq. 13 conflicts with the worked example in Section 4.6, the semantic embeddings and relevance term are undefined, and ASTS hyperparameters are omitted. These are serious completeness and verification concerns, but they are not circularity, because the claimed predictions are not equivalent to the method's inputs by construction.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The paper's central claim rests on a large set of hand-set weights and unspecified scoring functions. The free parameters (k, lambda, mu, T) are described as task-dependent, but no values, search procedure, or validation split is reported. No new physical or ontological entity is introduced; ASTS is a procedure built from existing components.

free parameters (5)
  • k1, k2 (entropy threshold scaling) = k1 = k2 = 0.3 in the illustrative example; experimental values not reported
    Set by hand in Eqs. 7-8 to control how far the typical set extends around H; directly determines which tokens are sampled.
  • lambda1, lambda2, lambda3 (composite score weights) = 0.4, 0.4, 0.2 in the example; experimental values not reported
    Weight coherence, semantic alignment, and diversity in Eq. 12; no tuning procedure or validation split is described.
  • mu1, mu2, mu3 (reward-penalty weights) = 0.5, 0.3, 0.2 in the example; experimental values not reported
    Weight semantic alignment, relevance, and repetition penalty in Eq. 14; relevance itself is not formally defined.
  • Temperature T = 1 in the example; experimental value not reported
    Controls distribution sharpness in Eq. 18; no value or schedule is given for the benchmark experiments.
  • tau for ASTS in Figures 1-2 = not defined in ASTS equations
    The text and figures report ASTS behavior across tau values, but tau is not a parameter of the ASTS equations in Sections 4.1-4.5; this makes the sensitivity plots under-specified.
assumptions (5)
  • domain assumption Token and context embeddings E(x) and E(Y<t) exist and cosine similarity is a valid measure of semantic alignment.
    Invoked in Eq. 11 and used in the composite and reward scores; no embedding model is specified.
  • domain assumption The standard deviation of past token entropies sigma_H is a meaningful prior for choosing the current typical set.
    Eqs. 7-8 set thresholds as H plus or minus k times sigma_H; no justification is given for this distributional assumption.
  • domain assumption Closeness of generated-text perplexity and Zipf coefficient to human-written references indicates better generation quality.
    Section 5.1 defines the optimal PPL and Zipf by deviation from reference; this is a modeling choice, not a theorem.
  • ad hoc to paper Linear combination of coherence, semantic alignment, and diversity scores (Eq. 12) monotonically improves generation quality.
    The composite score is proposed without derivation or ablation; its weights are hand-set.
  • standard math Softmax and entropy formulas from prior work are correct.
    Standard background used in Eqs. 1-2.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Advancing Decoding Strategies: Enhancements in Locally Typical Sampling for LLMs." pith.science (2026). https://pith.science/paper/7VXXFDZR

@misc{pith2026250605387,
  author       = {Pith},
  title        = {Pith review of: Advancing Decoding Strategies: Enhancements in Locally Typical Sampling for LLMs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7VXXFDZR}},
  note         = {Machine review of arXiv:2506.05387}
}
read the original abstract

This chapter explores advancements in decoding strategies for large language models (LLMs), focusing on enhancing the Locally Typical Sampling (LTS) algorithm. Traditional decoding methods, such as top-k and nucleus sampling, often struggle to balance fluency, diversity, and coherence in text generation. To address these challenges, Adaptive Semantic-Aware Typicality Sampling (ASTS) is proposed as an improved version of LTS, incorporating dynamic entropy thresholding, multi-objective scoring, and reward-penalty adjustments. ASTS ensures contextually coherent and diverse text generation while maintaining computational efficiency. Its performance is evaluated across multiple benchmarks, including story generation and abstractive summarization, using metrics such as perplexity, MAUVE, and diversity scores. Experimental results demonstrate that ASTS outperforms existing sampling techniques by reducing repetition, enhancing semantic alignment, and improving fluency.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

92 extracted references · 15 canonical work pages

  1. [9]

    In: Trans- actions of the Association for Computational Linguistics, Vol 11, pp

    Meister, C., Pimentel, T., Wiher, G., Cotterell, R.: Locally Typical Sampling. In: Trans- actions of the Association for Computational Linguistics, Vol 11, pp. 102 -121, MIT Press, Cambridge, MA, USA. DOI: https://doi.org/10.1162/tacl_a_00536 (2023)

    doi:10.1162/tacl_a_00536 2023

  2. [1]

    OpenAI Research Paper

    Radford, A., Narasimhan, K., Salimans, T., Sutskever, I.: Improving Language Under- standing by Generative Pre-Training. OpenAI Research Paper. (2018)

    2018

  3. [2]

    Radford, A., Wu, J., Child, R., Luan, D., Amodei, D., Sutskever, I.: Language Models are Unsupervised Multitask Learners. (2019)

    2019

  4. [3]

    W., Lee, K., Toutanova, K.: BERT: Pre-Training of Deep Bidirec- tional Transformers for Language Understanding

    Devlin, J., Chang, M. W., Lee, K., Toutanova, K.: BERT: Pre-Training of Deep Bidirec- tional Transformers for Language Understanding. In : Proceedings of the 2019 Confer- ence of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies , Vol 1, Long and Short Papers, pp. 4171–4186. DOI: https://doi.org/10.18...

    doi:10.18653/v1/n19-1423 2019

  5. [4]

    Raffel, C., Shazeer, N., Roberts, A., Lee, K., Narang, S., Matena, M., Zhou, Y., Li, W., Liu, P. J. Exploring the Limits of Transfer Learning with a Unified Text -to-Text Trans- former. Journal of Machine Learning Research, Vol 21, No 140, 1 –67. DOI: https://doi.org/10.48550/arXiv.1910.10683 (2019)

    arXiv doi:10.48550/arxiv.1910.10683 2019

  6. [5]

    Sutskever, I., Vinyals, O., Le, Q. V. Sequence to Sequence Learning with Neural Net- works. In: Proceedings of the 28 th International Conference on Neural Information Pro- cessing Systems, Vol 2, pp. 3104-3112. DOI: https://doi.org/10.48550/arXiv.1409.3215 (2014) Advancing Decoding Strategies: Enhancements in Locally Typical Sampling for LLMs 34

    arXiv doi:10.48550/arxiv.1409.3215 2014

  7. [6]

    J., Marcu, D

    Koehn, P., Och, F. J., Marcu, D. Statistical Phrase-Based Translation. In: Proceedings of the 2003 Human Language Technology Conference of the North American Chapter of the Association for Computational Linguistics, Vol 1, pp. 48 -54. DOI: https://doi.org/10.3115/1073445.107346 (2003)

    work page arXiv 2003

  8. [7]

    In: Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics, Long Pa- pers, pp

    Fan, A., Lewis, M., Dauphin, Y.: Hierarchical Neural Story Generation. In: Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics, Long Pa- pers, pp. 889-898. DOI: https://doi.org/10.48550/arXiv.1805.04833 (2018)

    arXiv doi:10.48550/arxiv.1805.04833 2018

Show all 92 references
  1. [8]

    In: Proceedings of the 8th International Conference on Learning Represen- tations (ICLR’20)

    Holtzman, A., Buys, J., Du, L., Forbes, M., Choi, Y.: The Curious Case of Neural Text Degeneration. In: Proceedings of the 8th International Conference on Learning Represen- tations (ICLR’20). DOI: https://doi.org/10.48550/arXiv.1904.09751 (2020)

    2020 doi arXiv

  2. [10]

    A Thorough Exam- ination of Decoding Methods in the Era of LLMs

    Shi, C., Yang, H., Cai, D., Zhang, Z., Wang, Y., Yang, Y., Lam, W. A Thorough Exam- ination of Decoding Methods in the Era of LLMs. In: Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pp. 8601 –8629. DOI: https://doi.org/10.18653/v1/2024...

    2024 doi

  3. [11]

    APAR: LLMs Can Do Auto- Parallel Auto -Regressive Decoding

    Liu, M., Zeng, A., Wang, B., Zhang, P., Tang, J., Dong, Y. APAR: LLMs Can Do Auto- Parallel Auto -Regressive Decoding. DOI: https://doi.org/10.48550/arXiv.2401.06761 (2024)

    2024 doi arXiv

  4. [12]

    E.: Think Big, Generate Quick: LLM -to-SLM for Fast Autoregressive Decoding

    Bergner, B., Skliar, A., Royer, A., Blankevoort, T., Asano, Y., Bejnordi, B. E.: Think Big, Generate Quick: LLM -to-SLM for Fast Autoregressive Decoding. arXiv:2402.16844. DOI: https://doi.org/10.48550/arXiv.2402.16844 (2024)

    2024 doi arXiv

  5. [13]

    In: Proceedings of the 202 4 Conference on Empirical Methods in Natural Language Pro- cessing, pp

    Jaiswal, A., Hu, B., Yin, L., Ro, Y., Liu, S., Chen, T., Akella, A.: FFN -SkipLLM: A Hidden Gem for Autoregressive Decoding with Adaptive Feed Forward Skipping. In: Proceedings of the 202 4 Conference on Empirical Methods in Natural Language Pro- cessing, pp. 16943 -16956. DOI...

    2024 doi

  6. [14]

    (C.).: When Linear Attention Meets Autoregressive Decoding: Towards More Effective and Efficient Linearized Large Language Models

    You, H., Fu, Y., Wang, Z., Yazdanbakhsh, A., Lin, Y. (C.).: When Linear Attention Meets Autoregressive Decoding: Towards More Effective and Efficient Linearized Large Language Models. In: Proceedings of the 41st International Conference on Machine Learning, pp. 57350-57366, DO...

    2024 doi arXiv

  7. [15]

    arXiv:2310.07177

    Liu, X., Hu, L., Bailis, P., Cheung, A., Deng, Z., Stoica, I., Zhang, H.: Online Speculative Decoding. arXiv:2310.07177. DOI: https://doi.org/10.48550/arXiv.2310.07177 (2023)

    2023 doi arXiv

  8. [16]

    In: Findings of the Association for Computational Linguistics, ACL 2024, pp

    Xia, H., Yang, Z., Dong, Q., Wang, P., Li, Y., Ge, T., Liu, T., Li, W., Sui, Z.: Unlocking Efficiency in Large Language Model Inference: A Comprehensive Survey of Speculative Decoding. In: Findings of the Association for Computational Linguistics, ACL 2024, pp. 7655-7671. DOI:...

    2024 doi

  9. [17]

    : Recursive Speculative De- coding: Accelerating LLM Inference via Sampling Without Replacement

    Jeon, W., Gagrani, M., Goel, R., Park, J., Lee, M., Lott, C. : Recursive Speculative De- coding: Accelerating LLM Inference via Sampling Without Replacement. arXiv:2402.14160. DOI: https://doi.org/10.48550/arXiv.2402.14160 (2024)

    2024 doi arXiv

  10. [18]

    V.: Nearest Neighbor Speculative Decoding for LLM Generation and Attribution

    Li, M., Chen, X., Holtzman, A., Chen, B., Lin, J., Yih, W., Lin, X. V.: Nearest Neighbor Speculative Decoding for LLM Generation and Attribution. arXiv:2405.19325. DOI: https://doi.org/10.48550/arXiv.2405.19325 (2024)

    2024 doi arXiv

  11. [19]

    DOI: https://doi.org/10.48550/arXiv.2412.00061 (2024)

    Wen, Z., Gui, S., Feng, Y.: Speculative Decoding with CTC-based Draft Model for LLM Inference Acceleration. DOI: https://doi.org/10.48550/arXiv.2412.00061 (2024)

    2024 doi arXiv

  12. [20]

    L., Holtzman, A., Fried, D., Liang, P., Eisner, J., Hashimoto, T., Zettlemoyer, L., Lewis, M.: Contrastive Decoding: Open-ended Text Generation as Optimization

    Li, X. L., Holtzman, A., Fried, D., Liang, P., Eisner, J., Hashimoto, T., Zettlemoyer, L., Lewis, M.: Contrastive Decoding: Open-ended Text Generation as Optimization. In: Pro- ceedings of the 61st Annual Meeting of the Association for Computational Linguistic, Vol 1, Long Pap...

    2023 doi

  13. [21]

    In: Proceedings of the 2024 Con- ference on North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Vol 1, Long Paper, pp

    Zhao, Z., Monti, E., Lehmann, J., Assem, H.: Enhancing Contextual Understanding in Large Language Models through Contrastive Decoding. In: Proceedings of the 2024 Con- ference on North American Chapter of the Association for Computational Linguistics: Human Language Technologi...

    2024 doi

  14. [22]

    arXiv:2309.09117

    O'Brien, S., Lewis, M.: Contrastive Decoding Improves Reasoning in Large Language Models. arXiv:2309.09117. DOI: https://doi.org/10.48550/arXiv.2309.09117 (2023)

    2023 doi arXiv

  15. [23]

    arXiv:2402.14874

    Phan, P., Tran, H., Phan, L.: Distillation Contrastive Decoding: Improving LLMs Rea- soning with Contrastive Decoding and Distillation. arXiv:2402.14874. DOI: https://doi.org/10.48550/arXiv.2402.14874 (2023)

    2023 doi arXiv

  16. [24]

    In: Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pp

    Chang, H.-S., Peng, N., Bansal, M., Ramakrishna, A., Chung, T.: Explaining and Im- proving Contrastive Decoding by Extrapolating the Probabilities of a Huge and Hypo- thetical LM. In: Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pp. 8...

    2024 doi

  17. [25]

    In : Pro- ceedings of the 62 nd Annual Meeting of the Association for Computational Linguistics, Vol 1, Long Papers, pp

    Zhang, J., Wang, J., Li, H., Shou, L., Chen, K., Chen, G., Mehrotra, S .: Draft & Verify: Lossless Large Language Model Acceleration via Self -Speculative Decoding. In : Pro- ceedings of the 62 nd Annual Meeting of the Association for Computational Linguistics, Vol 1, Long Pap...

    2024 doi

  18. [26]

    OPT-Tree: Spec- ulative Decoding with Adaptive Draft Tree Structure

    Wang, J., Su, Y., Li, J., Xia, Q., Ye, Z., Duan, X., Wang, Z., Zhang, M. OPT-Tree: Spec- ulative Decoding with Adaptive Draft Tree Structure. arXiv:2406.17276. DOI: https://doi.org/10.48550/arXiv.2406.17276 (2024)

    2024 doi arXiv

  19. [27]

    DOI: https://doi.org/10.48550/arXiv.2407.12021 (2024)

    Liu, X., Lei, B., Zhang, R., Xu, D.: Adaptive Draft-Verification for Efficient Large Lan- guage Model Decoding. DOI: https://doi.org/10.48550/arXiv.2407.12021 (2024)

    2024 doi arXiv

  20. [28]

    arXiv:2408.11850

    Liu, T., Li, Y., Lv, Q., Liu, K., Zhu, J., Hu, W.: Parallel Speculative Decoding with Adap- tive Draft Length. arXiv:2408.11850. DOI: https://doi.org/10.48550/arXiv.2408.11850 (2024)

    2024 doi arXiv

  21. [29]

    arXiv:2203.15721

    Wiher, G., Meister, C., Cotterell, R.: On Decoding Strategies for Neural Text Generators. arXiv:2203.15721. DOI: https://doi.org/10.48550/arXiv.2203.15721 (2022)

    2022 doi arXiv

  22. [30]

    In: Proceedings of the 61st Annual Meeting of the ACL, Vol 1, Long Papers, pp

    Xu, J., Xiong, C., Savarese, S., Zhou, Y.: Best-k Search Algorithm for Neural Text Gen- eration. In: Proceedings of the 61st Annual Meeting of the ACL, Vol 1, Long Papers, pp. 12385-12401. DOI: https://doi.org/10.18653/v1/2023.acl-long.692 (2020)

    2020 doi

  23. [31]

    In : Proceedings of the 1 st International Workshop on Multilin- gual, Multimodal, and Multitask Language Generation, pp

    Martínez-Murillo, I., Moreda, P., & Lloret, E.: Towards an efficient approach for con- trollable text generation. In : Proceedings of the 1 st International Workshop on Multilin- gual, Multimodal, and Multitask Language Generation, pp. 26-31, European Association for Machine T...

    2023

  24. [32]

    In: Proceedings of the 41 st International Conference on Machine Learning (ICML’24)

    Shao, C., Meng, F., Liu, Y., & Zhou, J.: Language Generation with Strictly Proper Scor- ing Rules. In: Proceedings of the 41 st International Conference on Machine Learning (ICML’24). DOI: https://doi.org/10.48550/arXiv.2405.18906 (2024)

    2024 doi arXiv

  25. [33]

    Z., Song, S., Yang, J., Niu, S., Hu, J., Liu, D., Yao, S., Xiong, F., Li, Z.: Controllable Text Generation for Large Language Models: A Survey

    Liang, X., Wang, H., Wang, Y. Z., Song, S., Yang, J., Niu, S., Hu, J., Liu, D., Yao, S., Xiong, F., Li, Z.: Controllable Text Generation for Large Language Models: A Survey. arXiv:2408.12599. DOI: https://doi.org/10.48550/arXiv.2408.12599 (2024)

    2024 doi arXiv

  26. [34]

    V., Rozière, B., Kahn, J., Li, D., Yih, W.-T., Weston, J., Li, X

    Sukhbaatar, S., Golovneva, O., Sharma, V., Xu, H., Lin, X. V., Rozière, B., Kahn, J., Li, D., Yih, W.-T., Weston, J., Li, X. Branch-Train-MiX: Mixing Expert LLMs into a Mix- ture-of-Experts LLM. DOI: https://doi.org/10.48550/arXiv.2403.07816 (2024)

    2024 doi arXiv

  27. [35]

    P., Zhang, H.: Toward Inference-Optimal Mixture- of-Expert Large Language Models

    Yun, L., Zhuang, Y., Fu, Y., Xing, E. P., Zhang, H.: Toward Inference-Optimal Mixture- of-Expert Large Language Models. DOI: https://doi.org/10.48550/arXiv.2404.02852 (2024)

    2024 doi arXiv

  28. [36]

    arXiv:2406.16554

    Zhu, T., Qu, X., Dong, D., Ruan, J., Tong, J., He, C., Cheng, Y.: LLaMA-MoE: Building Mixture-of-Experts from LLaMA with Continual Pre-Training. arXiv:2406.16554. DOI: https://doi.org/10.48550/arXiv.2406.16554 (2024) Advancing Decoding Strategies: Enhancements in Locally Typic...

    2024 doi arXiv

  29. [37]

    arXiv:2410.10814

    Li, Z., Zhou, T.: Your Mixture -of-Experts LLM is Secretly an Embedding Model for Free. arXiv:2410.10814. DOI: https://doi.org/10.48550/arXiv.2410.10814 (2024)

    2024 doi arXiv

  30. [38]

    In: Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC - COLING’24), pp

    Zhang, C., Yang, Y., Li, Q., Wang, J., Song, D.: Task -Agnostic Distillation of Encoder- Decoder Language Models. In: Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC - COLING’24), pp. 15629-15639. (2024)

    2024

  31. [39]

    In: Proceedings of the International Conference on Learning Representa- tion (ICLR’24)

    Gu, Y., Dong, L., Wei, F., Huang, M.: MiniLLM: Knowledge distillation of large lan- guage models. In: Proceedings of the International Conference on Learning Representa- tion (ICLR’24) . arXiv:2306.08543. DOI: https://doi.org/10.48550/arXiv.2306.08543 (2023)

    2023 doi arXiv

  32. [40]

    S., Menon, A

    Zhou, Y., Lyu, K., Rawat, A. S., Menon, A. K., Rostamizadeh, A., Kumar, S., Kagy, J. - F., Agarwal, R.: D istillSpec: Improving Speculative Decoding via Knowledge Distilla- tion. arXiv:2310.08461. DOI: https://doi.org/10.48550/arXiv.2310.08461 (2023)

    2023 doi arXiv

  33. [41]

    In: Proceedings of the ICLR’24 Workshop on Understanding of Foundation Models (ME -FoMo)

    Goel, R., Gagrani, M., Jeon, W., Park, J., Lee, M., Lott, C.: Direct Alignment of Draft Model for Speculative Decoding with Chat -Fine-Tuned LLMs. In: Proceedings of the ICLR’24 Workshop on Understanding of Foundation Models (ME -FoMo). arXiv:2403.00858. DOI: https://doi.org/1...

    2024 doi arXiv

  34. [42]

    In: ICLR’24 Poster Track , arXiv:2307.15337

    Ning, X., Lin, Z., Zhou, Z., Wang, Z., Yang, H., Wang, Y.: Skeleton-of-Thought: Prompt- ing LLMs for Efficient Parallel Generation. In: ICLR’24 Poster Track , arXiv:2307.15337. DOI: https://doi.org/10.48550/arXiv.2307.15337 (2024)

    2024 doi arXiv

  35. [43]

    arXiv:2404.12022

    Wu, P., Liu, J., Gong, Z., Wang, Q., Li, J., Wang, J., Cai, X., Zhao, D.: Parallel Decoding via Hidden Transfer for Lossless Large Language Model Acceleration. arXiv:2404.12022. DOI: https://doi.org/10.48550/arXiv.2404.12022 (2024)

    2024 doi arXiv

  36. [44]

    In: Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC - COLING’24), pp

    Sun, C., Zhang, H., Lin, Z., Zhang, J., Zhang, F., Wang, Z., Chen, B., Song, C., Zhang, D., Gai, K., Xiong, D.: Decoding at the Speed of Thought: Harnessing Parallel Decoding of Lexical Units for LLMs. In: Proceedings of the 2024 Joint International Conference on Computational...

    2024

  37. [45]

    In: Proceedings of ICML’23, arXiv:2210.15458

    Vilnis, L., Zemlyanskiy, Y., Murray, P., Passos, A., Sanghai, S.: Arithmetic Sampling: Parallel Diverse Decoding for Large Language Models. In: Proceedings of ICML’23, arXiv:2210.15458. DOI: https://doi.org/10.48550/arXiv.2210.15458 (2023)

    2023 doi arXiv

  38. [46]

    In: Findings of the Association for Computa- tional Linguistics, pp

    Zheng, C., Ke, P., Zhang, Z., Huang, M.: C lick: Controllable Text Generation with Se- quence Likelihood Contrastive Learning. In: Findings of the Association for Computa- tional Linguistics, pp. 1022-1040. DOI: https://doi.org/10.18653/v1/2023.findings-acl.65 (2023)

    2023 doi

  39. [47]

    In: Proceedings of the 36th Conference on Neural Information Processing Systems (NeurIPS’22)

    An, C., Feng, J., Lv, K., Kong, L., Qiu, X., & Huang, X.: CoNT: Contrastive Neural Text Generation. In: Proceedings of the 36th Conference on Neural Information Processing Systems (NeurIPS’22). DOI: https://doi.org/10.48550/arXiv.2205.14690 (2022)

    2022 doi arXiv

  40. [48]

    In: Proceedings of the 38th Conference on Uncertainty in Artifi- cial Intelligence, Vol 180, pp

    Teng, Z., Chen, C., Zhang, Y., & Zhang, Y.: Contrastive Latent Variable Models for Neural Text Generation. In: Proceedings of the 38th Conference on Uncertainty in Artifi- cial Intelligence, Vol 180, pp. 1928-1938. (2022)

    2022

  41. [49]

    Ghazvininejad, M., Levy, O., Liu, Y., Zettlemoyer, L.: Mask-Predict: Parallel Decoding of Conditional Masked Language Models. In: Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9 th International Joint Con- ference on Natural Lan...

    2019 doi

  42. [50]

    DOI: https://doi.org/10.48550/arXiv.2001.08785 (2020)

    Ghazvininejad, M., Levy, O., Zettlemoyer, L.: Semi -Autoregressive Training Improves Mask-Predict Decoding. DOI: https://doi.org/10.48550/arXiv.2001.08785 (2020)

    2020 doi arXiv

  43. [51]

    D., Brantley, K., Ramamurthy, R., Misra, D., Sun, W.: Learning to Generate Better than Your LLM

    Chang, J. D., Brantley, K., Ramamurthy, R., Misra, D., Sun, W.: Learning to Generate Better than Your LLM. DOI: https://doi.org/10.48550/arXiv.2306.11816 (2023) 37 Sen et al

    2023 doi arXiv

  44. [52]

    In: Proceedings of the International Conference on Machine Learning (ICML’24)

    Mudgal, S., Lee, J., Ganapathy, H., Li, Y., Wang, T., Huang, Y., Chen, Z., Cheng, H.-T., Collins, M., Strohman, T., Chen, J., Beutel, A., Beirami, A.: Controlled Decoding from Language Models. In: Proceedings of the International Conference on Machine Learning (ICML’24). DOI: ...

    2024 doi arXiv

  45. [53]

    In: Proceedings of the 2024 Con- ference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Vol 2: Short Papers, pp

    Yang, G., Chen, J., Lin, W., Byrne, B.: Direct Preference Optimization for Neural Ma- chine Translation with Minimum Bayes Risk Decoding. In: Proceedings of the 2024 Con- ference of the North American Chapter of the Association for Computational Linguistics: Human Language Tec...

    2024 doi

  46. [54]

    arXiv:2402.06700

    Wen, M., Liao, J., Deng, C., Wang, J., Zhang, W., Wen, Y.: Entropy-Regularized Token- Level Policy Optimization for Language Agent Reinforcement. arXiv:2402.06700. DOI: https://doi.org/10.48550/arXiv.2402.06700 (2024)

    2024 doi arXiv

  47. [55]

    A., Lange, L., Adel, H., Strötgen, J., Klakow, D.: A Survey on Recent Approaches for Natural Language Processing in Low -Resource Scenarios

    Hedderich, M. A., Lange, L., Adel, H., Strötgen, J., Klakow, D.: A Survey on Recent Approaches for Natural Language Processing in Low -Resource Scenarios. In: Proceed- ings of the 2021 Conference of the North American Chapter of the Association for Com- putational Linguistics:...

    2021 doi

  48. [56]

    Electronics, Vol 9, No 1 0, Article ID: 1562

    Park, C., Yang, Y., Park, K., & Lim, H.: Decoding Strategies for Improving Low -Re- source Machine Translation. Electronics, Vol 9, No 1 0, Article ID: 1562 . DOI: https://doi.org/10.3390/electronics9101562 (2020)

    2020 doi

  49. [57]

    Complex & Intelligent Systems, Vol 11, Article No: 57

    Xu, M., Zhang, J., Xu, L., Silamu, W., Li, Y.: Correlation-Guided Decoding Strategy for Low-Resource Uyghur Scene Text Recognition. Complex & Intelligent Systems, Vol 11, Article No: 57. DOI: https://doi.org/10.1007/s40747-024-01689-5 (2025)

    2025 doi

  50. [58]

    -H., Hsu, Y

    Tuli, S., Lin, C. -H., Hsu, Y. -C., Jha, N. K., Shen, Y., Jin, H.: DynaMo: Accelerating Language Model Inference with Dynamic Multi-Token Sampling. In: Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistic: Human Langu...

    2024 doi

  51. [59]

    DOI: https://doi.org/10.48550/arXiv.2405.04304 (2024)

    Mamou, J., Peregs, O., Korats, D., Berchanskys, M., Timor, N., Wasserblats, M., Schwartzp, R.: Dynamic Speculation Lookahead Accelerates Speculative Decoding of Large Language Models. DOI: https://doi.org/10.48550/arXiv.2405.04304 (2024)

    2024 doi arXiv

  52. [60]

    arXiv:2012.15416

    Pascual, D., Egressy, B., Bolli, F., & Wattenhofer, R.: Directed Beam Search: Plug-and- Play Lexically Constrained Language Generation. arXiv:2012.15416. DOI: https://doi.org/10.48550/arXiv.2012.15416 (2020)

    2020 doi arXiv

  53. [61]

    arXiv:2407.03951

    Grosse, J., Wu, R., Rashid, A., Hennig, P., Poupart, P., Kristiadi, A.: Uncertainty-Guided Optimization on Large Language Model Search Trees. arXiv:2407.03951. DOI: https://doi.org/10.48550/arXiv.2407.03951 (2024)

    2024 doi arXiv

  54. [62]

    DOI: https://doi.org/10.48550/arXiv.2409.16560 (2024)

    Qin, Z., He, Z., Prakriya, N., Cong, J., Sun, Y.: Dynamic -Width Speculative Beam De- coding for Efficient LLM Inference. DOI: https://doi.org/10.48550/arXiv.2409.16560 (2024)

    2024 doi arXiv

  55. [63]

    DOI: https://doi.org/10.48550/arXiv.2406.17519 (2024)

    Qiu, Z., Ou, Z., Wu, B., Li, J., Liu, A., King, I.: Entropy -Based Decoding for Retrieval- Augmented Large Language Models. DOI: https://doi.org/10.48550/arXiv.2406.17519 (2024)

    2024 doi arXiv

  56. [64]

    In: Findings of the Association for Computational Linguistics, pp

    Wang, R., Li, J., Li, P.: InfoDiffusion: Information Entropy Aware Diffusion Process for Non-Autoregressive Text Generation. In: Findings of the Association for Computational Linguistics, pp. 13757 -13770. DOI: https://doi.org/10.18653/v1/2023.findings- emnlp.919 (2023)

    2023 doi

  57. [65]

    J., Precup, D., Weston, J., Cheung, J

    Arora, K., O’Donnell, T. J., Precup, D., Weston, J., Cheung, J. C. K.: The Stable Entropy Hypothesis and Entropy-Aware Decoding: An Analysis and Algorithm for Robust Natu- ral Language Generation. DOI: https://doi.org/10.48550/arXiv.2302.06784 (2023) Advancing Decoding Strateg...

    2023 doi arXiv

  58. [66]

    Advances in Neural Information Pro- cessing Systems (NeurIPS’22), 35, 10656 -10668

    Qin, L., Welleck, S., Khashabi, D., Choi, Y.: COLD Decoding: Energy -based Con- strained Text Generation with Langevin Dynamics. Advances in Neural Information Pro- cessing Systems (NeurIPS’22), 35, 10656 -10668. arXiv:2202.11705. DOI: https://doi.org/10.48550/arXiv.2202.11705 (2023)

    2023 doi arXiv

  59. [67]

    In: Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (ACL’23), pp

    Liu, X., Khalifa, M., Wang, L.: BOLT: Fast Energy -based Controlled Text Generation with Tunable Biases. In: Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (ACL’23), pp. 186 -200. arXiv:2305.12018. DOI: https://doi.org/10.48550/arXiv.23...

    2023 doi arXiv

  60. [68]

    arXiv:2408.08146

    Zhang, K., Zhao, J., Chen, R.: KOALA: Enhancing Speculative Decoding for LLM via Multi-Layer Draft Heads with Adversarial Learning. arXiv:2408.08146. DOI: https://doi.org/10.48550/arXiv.2408.08146 (2024)

    2024 doi arXiv

  61. [69]

    In: Proceedings of the 32nd Conference on Neural Information Processing Systems (NIPS’18), Montreal, Can- ada, pp

    He, T., Tan, X., Xia, Y., He, D., Qin, T., Chen, Z., Liu, T.-Y.: Layer-Wise Coordination between Encoder and Decoder for Neural Machine Translation. In: Proceedings of the 32nd Conference on Neural Information Processing Systems (NIPS’18), Montreal, Can- ada, pp. 7955-7965. (2018)

    2018

  62. [70]

    et al.: Transformers: State-of-the-Art Natural Language Processing

    Wolf, T. et al.: Transformers: State-of-the-Art Natural Language Processing. In: Proceed- ings of the 2020 Conference on Empirical Methods in Natural Language Processing: Sys- tem Demonstrations, pp. 38 -45, Association for Computational Linguistics. DOI: https://doi.org/10.18...

    2020 doi

  63. [71]

    In: Proceedings of ICLR 2021

    Basu, S., Ramachandran, G.S., Keskar, N.S, Varshney, L.R.: Mirostat: A Neural Text Decoding Algorithm that Directly Controls Perplexity. In: Proceedings of ICLR 2021. DOI: https://doi.org/10.48550/arXiv.2007.14966 (2020)

    2020 doi arXiv

  64. [72]

    Journal of Machine Learning Research, Vol 24, No 1 -92

    Pillutla, K., Liu, L., Thickstun, J., Welleck, S., Swayamdipta, S., Zellers, R., Oh, S., Choi, Y., Harchaoui, Z.: MAUVE Scores for Generative Models: Theory and Prac-tice. Journal of Machine Learning Research, Vol 24, No 1 -92. DOI: https://doi.org/10.48550/arXiv.2212.14578 (2022)

    2022 doi arXiv

  65. [73]

    In: Proceedings of the 35th International Conference on Neural Infor- mation Processing Systems (NIPS’21), Article No: 368, pp

    Pillutla, K., Swayamdipta, S., Zellers, R., Thickstun, J., Welleck, S., Choi, Y., Harchaoui, Z.: MAUVE: Measuring the Gap between Neural Text and Human Text Using Diver- gence Frontiers. In: Proceedings of the 35th International Conference on Neural Infor- mation Processing Sy...

    2021

  66. [74]

    In: Proceeding of the Workshop on Human Evalua- tion of NLP Systems (HumEval’21), pp

    Zhang H., Duckworth D., Ippolito, D., Neelakantan, A.: Trading Off Diversity and Qual- ity in Natural Language Generation. In: Proceeding of the Workshop on Human Evalua- tion of NLP Systems (HumEval’21), pp. 25-33. Association for Computational Linguis- tics. (2021)

    2021

  67. [75]

    Nadeem, M., He, T., Cho, K., Glass, J.: A Systematic Characterization of Sampling Al- gorithms for Open-Ended Language Generation. In: Proceedings of the 1st Con -ference of the Asia-Pacific Chapter of the Association for Computational Linguistics and the 10th International Jo...

    2020 doi

  68. [76]

    In: Proceedings of the 40th Annual Meeting on Association for Computational Linguistics, pp

    Papineni, K., Roukos, S., Ward, T., Zhu, W.-J.: BLEU: A Method for Automatic Evalu- ation of Machine Translation. In: Proceedings of the 40th Annual Meeting on Association for Computational Linguistics, pp. 311-318, July 7-12, 2002, Philadelphia, Pennsylvania, USA. DOI: https:...

    2002

  69. [77]

    Text Summa- rization Branches Out, pp

    Lin, C.-Y.: ROUGE: A Package for Automatic Evaluation of Summaries. Text Summa- rization Branches Out, pp. 74-81. Association for Computational Linguistics. (2004)

    2004

  70. [78]

    In: Proceedings of the 8th International Conference on Learning Representation (ICLR’20), Addis Ababa, Ethiopia

    Welleck S., Kulikov, I., Roller, S., Dinan, E., Cho, K., Weston, J.: Neural Text Gen era- tion with Unlikelihood Training. In: Proceedings of the 8th International Conference on Learning Representation (ICLR’20), Addis Ababa, Ethiopia. (2020)

    2020

  71. [79]

    Addison -Wesley Press

    Zipf, G.K.: Human Behavior and the Principle of Least Effort. Addison -Wesley Press. (1949) 39 Sen et al

    1949

  72. [80]

    In: Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics, Vol1, Long Papers, pp

    Fan, A., Lewis, M., Dauphin, Y.: Hierarchical Neural Story Generation. In: Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics, Vol1, Long Papers, pp. 889 -898, Melbourne, Australia. Association for Computational Linguist ics. DOI: https://d...

    2018 doi

  73. [81]

    In: Pro- ceedings of the 5th International Conference on Learning Representation

    Merity, S., Xiong, C., Bradbury, J., Socher, R.: Pointer Sentinel Mixture Models. In: Pro- ceedings of the 5th International Conference on Learning Representation. DOI: https://doi.org/10.48550/arXiv.1609.07843 (2016)

    2016 doi arXiv

  74. [82]

    et al.: BART: Denoising Sequence-to-Sequence Pre-Training for Natural Lan- guage Generation, Translation, and Comprehension

    Lewis, M. et al.: BART: Denoising Sequence-to-Sequence Pre-Training for Natural Lan- guage Generation, Translation, and Comprehension. In: Proceedings of the 58 th Annual Meeting of the Association for Computational Linguistics, pp. 7871 -7880. Association for Computational Li...

    2020 doi

  75. [83]

    In: Proceedings of the 29 th SIGNLL Conference on Computational Natural Language Learning, pp

    Nallapati, R., Zhou, B., dos Santos, C., Gulcehre, C., Xiang, B.: Abstractive Text Sum- marization Using Sequence-to-Sequence RNNs and Beyond. In: Proceedings of the 29 th SIGNLL Conference on Computational Natural Language Learning, pp. 280-290, Berlin, Germany. DOI: https://...

    2016 doi

  76. [84]

    IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol 46, No 12, pp

    Cheng, H., Zhang, M., Shi, J.Q.: A Survey on Deep Neural Network Pruning: Taxonomy, Comparison, Analysis, and Recommendations. IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol 46, No 12, pp. 10558 -10578. DOI: https://doi.org/10.1109/TPAMI.2024.3447085 (2024)

    2024

  77. [85]

    In: Proceedings of the 26th International Con- ference on Computer Supported Cooperative Work in Design (CSCWD’23) pp

    Xie, Y., Luo, Y., She, H., Xiang, Z.: Neural Network Model Pruning without Additional Computation and Structure Requirements. In: Proceedings of the 26th International Con- ference on Computer Supported Cooperative Work in Design (CSCWD’23) pp. 1734 -

  78. [86]

    In: Proceedings of the ACM on Management of Data, Vol 1, No 1, Article No 54, pp

    Peng, Y., Choi, B., Chan, T.N., Yang, J., Xu, J.: Efficient Approximate Nearest Neighbor Search in Multi-Dimensional Databased. In: Proceedings of the ACM on Management of Data, Vol 1, No 1, Article No 54, pp. 1-27. DOI: https://doi.org/10.1145/3588908 (2023)

    2023 doi

  79. [87]

    In: Proceedings of the 41st International Conference on Ma- chine Learning (ICML’24), Article No 497, pp

    Engels, J., Landrum, B., Yu, S., Dhulipala, L., Shun, J.: Approximate Nearest Neighbor Search with Window Filters. In: Proceedings of the 41st International Conference on Ma- chine Learning (ICML’24), Article No 497, pp. 12469-12490 (2024)

    2024

  80. [88]

    arXiv:2404.16322

    Yang, M., Li, W., Jin, J., Zhong, X., Wang, X., Shen, Z., Jia, W., Wang, W.: Effective and General Distance Computation for Approximate Nearest Neighbor Search. arXiv:2404.16322. DOI: https://doi.org/10.48550/arXiv.2404.16322 (2024)

    2024 doi arXiv

  81. [89]

    IEEE Transactions on Signal Processing, Vol 67, No 20, pp

    Chi, Y., Lu, Y.M., Chen, Y.: Nonconvex Optimization Meets Low-Rank Matrix Factor- ization: An Overview. IEEE Transactions on Signal Processing, Vol 67, No 20, pp. 5239-

  82. [90]

    In: 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp

    Jiang, F., Oskarsson, M., Astrom, K.: On the Minimal Problems of Low -Rank Matrix Factorization. In: 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 2549-2557. DOI: https://doi.org/10.1109/CVPR.2015. (2025)

    2025 doi

  83. [1740]

    DOI: https://doi.org/10.1109/CSCWD57460.2023.10152777 (2023)

    2023

  84. [5269]

    DOI: https://doi.org/10.1109/TSP.2019.2937282 (2019)

    2019

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.