pith. sign in

arxiv: 2606.17649 · v1 · pith:JT55P42Dnew · submitted 2026-06-16 · 💻 cs.LG · cs.AI

A Risk Decomposition Framework for Pre-Hoc Fine-Tuning Prediction

Yuxiang Luo , Chen Wang , Nan Tang This is my paper

Pith reviewed 2026-06-27 01:46 UTC · model grok-4.3

classification 💻 cs.LG cs.AI
keywords pre-hoc predictionrisk decompositionoptimization variancephase diagramLLM fine-tuningperformance predictionstochastic estimation
0
0 comments X

The pith

Pre-hoc LLM fine-tuning performance prediction faces a lower bound on how fast its uncertainty can decrease.

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

The paper models the task of forecasting an LLM's post-fine-tuning performance without running the fine-tuning as a stochastic estimation problem that respects information limits. It splits the total prediction risk into an intrinsic limit fixed by data-model compatibility and a reducible component called optimization variance. A proof shows that the variance term cannot decay faster than a specific rate no matter which predictor is chosen, which directly caps how quickly uncertainty about fine-tuning outcomes can shrink. This rate limit produces a phase diagram that places tasks into Static-Sufficient, Dynamic-Critical, or Noise-Dominant categories and yields a budget-optimal probing rule. The framing matters because fine-tuning large models is costly, so knowing these hard speed limits helps decide when pre-hoc checks are worth the investment.

Core claim

We formulate pre-hoc performance prediction as a stochastic estimation problem under information constraints, decomposing prediction risk into an intrinsic limit set by static data-model compatibility and a reducible optimization variance. We prove that optimization variance admits a necessary lower bound on its decay rate, implying fundamental constraints on how quickly uncertainty dissipates regardless of the predictor used. From these dynamics we derive a budget-optimal probing principle and a predictability phase diagram that organizes tasks into Static-Sufficient, Dynamic-Critical, and Noise-Dominant regimes.

What carries the argument

Risk decomposition into intrinsic limit and optimization variance, with the proven lower bound on the variance decay rate.

If this is right

  • Uncertainty dissipates at a bounded rate independent of the choice of predictor.
  • Tasks fall into one of three predictability regimes according to the phase diagram.
  • A budget-optimal probing strategy follows directly from the derived dynamics.

Where Pith is reading between the lines

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

  • The same decomposition could be applied to decide whether pre-hoc checks are economical for other expensive training procedures.
  • The phase diagram offers a way to pre-screen tasks before committing compute to any predictor.
  • Real-world validation on the reported benchmarks indicates the three regimes are observable rather than purely theoretical.

Load-bearing premise

The pre-hoc performance prediction problem can be accurately formulated as a stochastic estimation problem under information constraints.

What would settle it

An experiment in which optimization variance for a concrete fine-tuning predictor decays faster than the derived lower bound on any benchmark would falsify the central claim.

Figures

Figures reproduced from arXiv: 2606.17649 by Chen Wang, Nan Tang, Yuxiang Luo.

Figure 1
Figure 1. Figure 1: Perspective shift. We move from treating pre-hoc pre￾diction as a black-box regression problem (Left) to a structural decomposition of risk into intrinsic limits and optimization vari￾ance (Right). This motivates the problem of pre-hoc fine-tuning predic￾tion: estimating the final fine-tuning performance of a fine-tuning task before executing full training. More concretely, given a pretrained model, a data… view at source ↗
Figure 2
Figure 2. Figure 2: Population-level uncertainty decay across regimes. The figure shows normalized observable uncertainty as a function of probing depth c in log-log scale, aggregated by regime. Solid curves represent regime-wise means across tasks, with shaded regions indicating bootstrap confidence intervals. Vertical dashed lines denote the stable probing interval used for fitting. reflects that probing is largely unnecess… view at source ↗
Figure 4
Figure 4. Figure 4: Regime-dependent marginal gain of probing. The figure plots the normalized marginal reduction in observable uncertainty ∆(c) as a function of probing depth c for different regimes. All curves exhibit diminishing returns, but at markedly different rates. These patterns empirically validate the regime-aware stop￾ping principles derived in Theorem 6.1. They also sug￾gest that the practical value of probing li… view at source ↗
Figure 5
Figure 5. Figure 5: Observable uncertainty decay and effective intrinsic floors across regimes. Mean Ub(c) with confidence intervals is shown as a function of probing depth. Noise-Dominant regimes plateau at higher levels, indicating larger intrinsic ambiguity. 0 1 2 3 4 5 0 20 40 60 80 100 Count [PITH_FULL_IMAGE:figures/full_fig_p028_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Distribution of estimated uncertainty decay rates across regimes. Dynamic-Critical tasks exhibit slower decay, while Static￾Sufficient tasks show rapid contraction. Regime I (Static-Sufficient) Regime II (Dynamic-Critical) Regime III (Noise-Dominant) 0.0000 0.0002 0.0004 0.0006 0.0008 Pre dictiv e risk (C V M S E o n R) Ablation A1: static vs dynamic vs hybrid @ c=200 Static-only Dynamic-only Hybrid [PITH… view at source ↗
Figure 7
Figure 7. Figure 7: Ablation of information sources under a fixed probing budget. Dynamic signals are informative only in Dynamic-Critical regimes. 28 [PITH_FULL_IMAGE:figures/full_fig_p028_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Localization of hard cases on the empirical phase diagram. Hard-to-predict tasks cluster in Noise-Dominant regimes or near regime boundaries. 29 [PITH_FULL_IMAGE:figures/full_fig_p029_8.png] view at source ↗
read the original abstract

The high cost of fine-tuning LLMs poses a significant economic barrier; pre-hoc performance prediction offers a critical solution to substantially reduce this expense. However, the theoretical limits of pre-hoc performance prediction remain unexplored. We formulate it as a stochastic estimation problem under information constraints, decomposing prediction risk into two components: an intrinsic limit (static data-model compatibility) and a reducible optimization variance. We prove that optimization variance admits a necessary lower bound on its decay rate, implying fundamental constraints on how quickly uncertainty dissipates, regardless of the predictor used. Based on these dynamics, we derive a budget-optimal probing principle and introduce a predictability phase diagram that organizes tasks into three distinct regimes: Static-Sufficient, Dynamic-Critical, and Noise-Dominant. Extensive experiments on synthetic and real-world benchmarks validate these theoretical regimes and demonstrate the efficiency of our probing strategy.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 0 minor

Summary. The manuscript presents a risk decomposition framework for pre-hoc fine-tuning prediction of large language models. It formulates the problem as a stochastic estimation task under information constraints, decomposes the risk into an intrinsic limit and reducible optimization variance, proves a lower bound on the decay rate of the optimization variance, derives a budget-optimal probing principle, and introduces a predictability phase diagram with three regimes: Static-Sufficient, Dynamic-Critical, and Noise-Dominant. The theoretical findings are supported by experiments on synthetic and real-world benchmarks.

Significance. If the central claims hold, this work would be significant for providing theoretical foundations for pre-hoc performance prediction in LLM fine-tuning, potentially leading to more efficient use of computational resources. The proof of a necessary lower bound on variance decay and the phase diagram could influence how practitioners approach probing strategies. The experimental validation adds practical value, though its strength depends on the fidelity of the model to real fine-tuning dynamics.

major comments (3)
  1. [Abstract] Abstract: The proof that optimization variance admits a necessary lower bound on its decay rate is load-bearing for the central claim of fundamental constraints independent of the predictor. However, this bound is derived from the stochastic estimation formulation under information constraints; it is unclear if this bound is a genuine necessary limit or if it is induced by the specific modeling assumptions about information constraints.
  2. [Abstract] Abstract: The predictability phase diagram with regimes Static-Sufficient, Dynamic-Critical, and Noise-Dominant is derived from the decay-rate dynamics. The manuscript should explicitly show how these regimes are defined and distinguished, including any thresholds or conditions based on the lower bound, to allow assessment of whether they organize tasks in a non-trivial way.
  3. [Abstract] Abstract: The experiments are claimed to validate the theoretical regimes and the efficiency of the probing strategy. Without details on the specific benchmarks, how the regimes were identified, or error analysis, it is difficult to evaluate if the results support the claims or if confounding factors in the data affect the validation.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments on the abstract. We address each point below with clarifications on the theoretical claims and indicate where revisions will strengthen the presentation.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The proof that optimization variance admits a necessary lower bound on its decay rate is load-bearing for the central claim of fundamental constraints independent of the predictor. However, this bound is derived from the stochastic estimation formulation under information constraints; it is unclear if this bound is a genuine necessary limit or if it is induced by the specific modeling assumptions about information constraints.

    Authors: The stochastic estimation formulation under information constraints directly models the pre-hoc setting, where predictors have restricted access to fine-tuning dynamics. The lower bound is a necessary consequence of these constraints and is independent of the predictor; it is not an artifact but follows from the information-theoretic limits in the problem formulation. We will revise the abstract to explicitly note that the bound holds under this modeling framework. revision: partial

  2. Referee: [Abstract] Abstract: The predictability phase diagram with regimes Static-Sufficient, Dynamic-Critical, and Noise-Dominant is derived from the decay-rate dynamics. The manuscript should explicitly show how these regimes are defined and distinguished, including any thresholds or conditions based on the lower bound, to allow assessment of whether they organize tasks in a non-trivial way.

    Authors: The regimes are distinguished by the relationship between the optimization variance decay rate and the intrinsic limit, using the proven lower bound: Static-Sufficient when decay exceeds the bound sufficiently to reach the limit, Dynamic-Critical when the bound governs the transition, and Noise-Dominant when decay is constrained below effective reduction. We will add explicit definitions, thresholds, and conditions based on the lower bound to the phase diagram section. revision: yes

  3. Referee: [Abstract] Abstract: The experiments are claimed to validate the theoretical regimes and the efficiency of the probing strategy. Without details on the specific benchmarks, how the regimes were identified, or error analysis, it is difficult to evaluate if the results support the claims or if confounding factors in the data affect the validation.

    Authors: The full manuscript details the synthetic benchmarks (controlled variance parameters) and real-world NLP tasks, regime identification via empirical decay plots against the theoretical bound, and error analysis with repeated runs and confidence intervals. The abstract is concise, but we will add a brief validation summary. The experiments isolate the modeled effects and support the claims. revision: partial

Circularity Check

0 steps flagged

No significant circularity; derivation is a standard mathematical consequence of the stochastic model

full rationale

The paper's central result is a mathematical proof of a lower bound on optimization variance decay, obtained after explicitly formulating the problem as stochastic estimation under information constraints and decomposing risk into intrinsic limit plus reducible variance. This is a first-principles derivation internal to the chosen model rather than a tautology, fitted parameter renamed as prediction, or load-bearing self-citation. No equations or steps in the provided text reduce the bound to its inputs by construction; the bound follows from the information constraints assumed in the formulation. The modeling choice itself is an assumption whose validity is separate from circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests primarily on the domain assumption that pre-hoc prediction is a stochastic estimation problem under information constraints; no free parameters or invented entities with independent evidence are described in the abstract.

axioms (1)
  • domain assumption Pre-hoc fine-tuning performance prediction can be formulated as a stochastic estimation problem under information constraints.
    This modeling choice is the starting point for the risk decomposition and the proof of the variance decay bound.
invented entities (1)
  • Predictability phase diagram with regimes Static-Sufficient, Dynamic-Critical, Noise-Dominant no independent evidence
    purpose: Organizes tasks according to the dynamics of prediction risk.
    Introduced based on the derived lower bound and budget-optimal probing principle.

pith-pipeline@v0.9.1-grok · 5669 in / 1303 out tokens · 50394 ms · 2026-06-27T01:46:37.804042+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

175 extracted references · 17 canonical work pages · 2 internal anchors

  1. [1]

    Langley , title =

    P. Langley , title =. Proceedings of the 17th International Conference on Machine Learning (ICML 2000) , address =. 2000 , pages =

    2000

  2. [2]

    T. M. Mitchell. The Need for Biases in Learning Generalizations. 1980

    1980

  3. [3]

    M. J. Kearns , title =

  4. [4]

    Xiaotian Lin and Yanlin Qi and Yizhang Zhu and Themis Palpanas and Chengliang Chai and Nan Tang and Yuyu Luo , title =. Proc. 2025 , url =

    2025

  5. [5]

    Machine Learning: An Artificial Intelligence Approach, Vol. I. 1983

    1983

  6. [6]

    R. O. Duda and P. E. Hart and D. G. Stork. Pattern Classification. 2000

    2000

  7. [7]

    Suppressed for Anonymity , author=

  8. [8]

    Newell and P

    A. Newell and P. S. Rosenbloom. Mechanisms of Skill Acquisition and the Law of Practice. Cognitive Skills and Their Acquisition. 1981

    1981

  9. [9]

    A. L. Samuel. Some Studies in Machine Learning Using the Game of Checkers. IBM Journal of Research and Development. 1959

    1959

  10. [10]

    Scaling Learning Algorithms Towards

    Bengio, Yoshua and LeCun, Yann , booktitle =. Scaling Learning Algorithms Towards

  11. [11]

    and Osindero, Simon and Teh, Yee Whye , journal =

    Hinton, Geoffrey E. and Osindero, Simon and Teh, Yee Whye , journal =. A Fast Learning Algorithm for Deep Belief Nets , volume =

  12. [12]

    2016 , publisher=

    Deep learning , author=. 2016 , publisher=

    2016

  13. [13]

    2025 , eprint=

    COSMOS: Predictable and Cost-Effective Adaptation of LLMs , author=. 2025 , eprint=

    2025

  14. [14]

    2024 , eprint=

    Fine-Tuning or Fine-Failing? Debunking Performance Myths in Large Language Models , author=. 2024 , eprint=

    2024

  15. [15]

    2025 , eprint=

    LENSLLM: Unveiling Fine-Tuning Dynamics for LLM Selection , author=. 2025 , eprint=

    2025

  16. [16]

    2017 , eprint=

    A Unified Approach to Interpreting Model Predictions , author=. 2017 , eprint=

    2017

  17. [17]

    2020 , eprint=

    Meta-Dataset: A Dataset of Datasets for Learning to Learn from Few Examples , author=. 2020 , eprint=

    2020

  18. [18]

    2021 , eprint=

    Dataset2Vec: Learning Dataset Meta-Features , author=. 2021 , eprint=

    2021

  19. [19]

    2022 , eprint=

    Model Zoos: A Dataset of Diverse Populations of Neural Network Models , author=. 2022 , eprint=

    2022

  20. [20]

    Proceedings of the 24th International Conference on Artificial Intelligence , pages =

    Domhan, Tobias and Springenberg, Jost Tobias and Hutter, Frank , title =. Proceedings of the 24th International Conference on Artificial Intelligence , pages =. 2015 , isbn =

    2015

  21. [21]

    2020 , eprint=

    Scaling Laws for Neural Language Models , author=. 2020 , eprint=

    2020

  22. [22]

    2024 , eprint=

    ProxyLM: Predicting Language Model Performance on Multilingual Tasks via Proxy Models , author=. 2024 , eprint=

    2024

  23. [23]

    2020 , eprint=

    Dataset Cartography: Mapping and Diagnosing Datasets with Training Dynamics , author=. 2020 , eprint=

    2020

  24. [24]

    2019 , eprint=

    Data Shapley: Equitable Valuation of Data for Machine Learning , author=. 2019 , eprint=

    2019

  25. [25]

    2025 , eprint=

    Refine-n-Judge: Curating High-Quality Preference Chains for LLM-Fine-Tuning , author=. 2025 , eprint=

    2025

  26. [26]

    2018 , eprint=

    The Dataset Nutrition Label: A Framework To Drive Higher Data Quality Standards , author=. 2018 , eprint=

    2018

  27. [27]

    2021 , eprint=

    MAUVE: Measuring the Gap Between Neural Text and Human Text using Divergence Frontiers , author=. 2021 , eprint=

    2021

  28. [28]

    2023 , eprint=

    GENTLE: A Genre-Diverse Multilayer Challenge Set for English NLP and Linguistic Evaluation , author=. 2023 , eprint=

    2023

  29. [29]

    2021 , eprint=

    On the Importance of Gradients for Detecting Distributional Shifts in the Wild , author=. 2021 , eprint=

    2021

  30. [30]

    2018 , eprint=

    GradNorm: Gradient Normalization for Adaptive Loss Balancing in Deep Multitask Networks , author=. 2018 , eprint=

    2018

  31. [31]

    Louis , year=

    Shuofeng Zhang and Isaac Reid and Guillermo Valle-Perez and Ard A. Louis , year=

  32. [32]

    2024 , url=

    Gradient norm as a powerful proxy to out-of-distribution error estimation , author=. 2024 , url=

    2024

  33. [33]

    2023 , eprint=

    Gradient Norm Aware Minimization Seeks First-Order Flatness and Improves Generalization , author=. 2023 , eprint=

    2023

  34. [34]

    2021 , eprint=

    Self-Validation: Early Stopping for Single-Instance Deep Generative Priors , author=. 2021 , eprint=

    2021

  35. [35]

    2025 , eprint=

    Autoencoder-Based Framework to Capture Vocabulary Quality in NLP , author=. 2025 , eprint=

    2025

  36. [36]

    2025 , eprint=

    Massive Supervised Fine-tuning Experiments Reveal How Data, Layer, and Training Factors Shape LLM Alignment Quality , author=. 2025 , eprint=

    2025

  37. [37]

    2025 , eprint=

    Assessing the Impact of the Quality of Textual Data on Feature Representation and Machine Learning Models , author=. 2025 , eprint=

    2025

  38. [38]

    Model Explainability using SHAP Values for LightGBM Predictions , year=

    Bugaj, Michal and Wrobel, Krzysztof and Iwaniec, Joanna , booktitle=. Model Explainability using SHAP Values for LightGBM Predictions , year=

  39. [39]

    Proceedings of the 31st International Conference on Neural Information Processing Systems , pages =

    Ke, Guolin and Meng, Qi and Finley, Thomas and Wang, Taifeng and Chen, Wei and Ma, Weidong and Ye, Qiwei and Liu, Tie-Yan , title =. Proceedings of the 31st International Conference on Neural Information Processing Systems , pages =. 2017 , isbn =

    2017

  40. [40]

    2025 , issue_date =

    Wang, Ying and Wang, Pengxin and Tansey, Kevin and Liu, Junming and Delaney, Bethany and Quan, Wenting , title =. 2025 , issue_date =. doi:10.1016/j.compag.2024.109758 , journal =

    work page doi:10.1016/j.compag.2024.109758 2025

  41. [41]

    Garcia and Carlos Soares and Joaquin Vanschoren and André C.P.L.F

    Adriano Rivolli and Luís P.F. Garcia and Carlos Soares and Joaquin Vanschoren and André C.P.L.F. Meta-features for meta-learning , journal =. 2022 , issn =. doi:https://doi.org/10.1016/j.knosys.2021.108101 , url =

    work page doi:10.1016/j.knosys.2021.108101 2022

  42. [42]

    2022 , eprint=

    Predicting Fine-Tuning Performance with Probing , author=. 2022 , eprint=

    2022

  43. [43]

    2024 , eprint=

    Predictable Emergent Abilities of LLMs: Proxy Tasks Are All You Need , author=. 2024 , eprint=

    2024

  44. [44]

    2020 , eprint=

    The Break-Even Point on Optimization Trajectories of Deep Neural Networks , author=. 2020 , eprint=

    2020

  45. [45]

    2019 , eprint=

    Visualizing and Understanding the Effectiveness of BERT , author=. 2019 , eprint=

    2019

  46. [46]

    2023 , eprint=

    Efficient Bayesian Learning Curve Extrapolation using Prior-Data Fitted Networks , author=. 2023 , eprint=

    2023

  47. [47]

    International Conference on Learning Representations (ICLR) , year =

    Measuring Massive Multitask Language Understanding , author =. International Conference on Learning Representations (ICLR) , year =

  48. [48]

    Qwen2.5 Technical Report

    Qwen2.5 Technical Report , author =. arXiv preprint arXiv:2412.15115 , year =. doi:10.48550/arXiv.2412.15115 , url =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2412.15115

  49. [49]

    2025 , url =

    Qwen2.5-7B-Instruct , author =. 2025 , url =

    2025

  50. [50]

    2025 , eprint=

    A Multi-Power Law for Loss Curve Prediction Across Learning Rate Schedules , author=. 2025 , eprint=

    2025

  51. [51]

    2020 , eprint=

    Don't Stop Pretraining: Adapt Language Models to Domains and Tasks , author=. 2020 , eprint=

    2020

  52. [52]

    Enhancing Data Quality through Simple De-duplication: Navigating Responsible Computational Social Science Research

    Mu, Yida and Jin, Mali and Song, Xingyi and Aletras, Nikolaos. Enhancing Data Quality through Simple De-duplication: Navigating Responsible Computational Social Science Research. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024. doi:10.18653/v1/2024.emnlp-main.694

    work page doi:10.18653/v1/2024.emnlp-main.694 2024

  53. [53]

    2025 , eprint=

    Measuring Data Diversity for Instruction Tuning: A Systematic Analysis and A Reliable Metric , author=. 2025 , eprint=

    2025

  54. [54]

    2023 , eprint=

    The Vendi Score: A Diversity Evaluation Metric for Machine Learning , author=. 2023 , eprint=

    2023

  55. [55]

    2018 , eprint=

    Visualizing the Loss Landscape of Neural Nets , author=. 2018 , eprint=

    2018

  56. [56]

    2020 , eprint=

    Adversarial Weight Perturbation Helps Robust Generalization , author=. 2020 , eprint=

    2020

  57. [57]

    Mathematics , VOLUME =

    Liang, Hailun and Zheng, Haowen and Wang, Hao and He, Liu and Lin, Haoyi and Liang, Yanyan , TITLE =. Mathematics , VOLUME =. 2025 , NUMBER =

    2025

  58. [58]

    Technometrics , volume =

    Ridge Regression: Biased Estimation for Nonorthogonal Problems , author =. Technometrics , volume =. 1970 , publisher =

    1970

  59. [59]

    Advances in Neural Information Processing Systems , volume =

    Support Vector Regression Machines , author =. Advances in Neural Information Processing Systems , volume =. 1997 , publisher =

    1997

  60. [60]

    Statistics and Computing , volume =

    A Tutorial on Support Vector Regression , author =. Statistics and Computing , volume =. 2004 , doi =

    2004

  61. [61]

    Machine Learning , volume =

    Random Forests , author =. Machine Learning , volume =. 2001 , doi =

    2001

  62. [62]

    Nature , volume =

    Learning Representations by Back-Propagating Errors , author =. Nature , volume =. 1986 , doi =

    1986

  63. [63]

    Neural Networks , volume =

    Multilayer Feedforward Networks Are Universal Approximators , author =. Neural Networks , volume =. 1989 , doi =

    1989

  64. [64]

    Journal of Machine Learning Research , volume =

    Scikit-learn: Machine Learning in Python , author =. Journal of Machine Learning Research , volume =. 2011 , url =

    2011

  65. [65]

    arXiv preprint arXiv:1706.10239 , year=

    Towards Understanding Generalization of Deep Learning: Perspective of Loss Landscapes , author=. arXiv preprint arXiv:1706.10239 , year=

    Pith/arXiv arXiv

  66. [66]

    arXiv preprint arXiv:2412.13573 , year=

    Seeking Consistent Flat Minima for Better Domain Generalization via Refining Loss Landscapes , author=. arXiv preprint arXiv:2412.13573 , year=

    arXiv

  67. [67]

    2019 , eprint=

    Characterizing classification datasets: a study of meta-features for meta-learning , author=. 2019 , eprint=

    2019

  68. [68]

    2024 , eprint=

    Machine Translation Meta Evaluation through Translation Accuracy Challenge Sets , author=. 2024 , eprint=

    2024

  69. [69]

    2023 , eprint=

    Advances and Challenges in Meta-Learning: A Technical Review , author=. 2023 , eprint=

    2023

  70. [70]

    2018 , eprint=

    Don't Give Me the Details, Just the Summary! Topic-Aware Convolutional Neural Networks for Extreme Summarization , author=. 2018 , eprint=

    2018

  71. [71]

    ROUGE : A Package for Automatic Evaluation of Summaries

    Lin, Chin-Yew. ROUGE : A Package for Automatic Evaluation of Summaries. Text Summarization Branches Out. 2004

    2004

  72. [72]

    2018 , eprint=

    A Survey of Machine Learning for Big Code and Naturalness , author=. 2018 , eprint=

    2018

  73. [73]

    , title =

    Raffel, Colin and Shazeer, Noam and Roberts, Adam and Lee, Katherine and Narang, Sharan and Matena, Michael and Zhou, Yanqi and Li, Wei and Liu, Peter J. , title =. J. Mach. Learn. Res. , month = jan, articleno =. 2020 , issue_date =

    2020

  74. [74]

    2022 , eprint=

    Deduplicating Training Data Makes Language Models Better , author=. 2022 , eprint=

    2022

  75. [75]

    2023 , eprint=

    Extracting Training Data from Diffusion Models , author=. 2023 , eprint=

    2023

  76. [76]

    2022 , eprint=

    Training Compute-Optimal Large Language Models , author=. 2022 , eprint=

    2022

  77. [77]

    2019 , eprint=

    Deep Anomaly Detection with Outlier Exposure , author=. 2019 , eprint=

    2019

  78. [78]

    2022 , eprint=

    Confident Learning: Estimating Uncertainty in Dataset Labels , author=. 2022 , eprint=

    2022

  79. [79]

    2021 , eprint=

    Documenting Large Webtext Corpora: A Case Study on the Colossal Clean Crawled Corpus , author=. 2021 , eprint=

    2021

  80. [80]

    2020 , eprint=

    The Curious Case of Neural Text Degeneration , author=. 2020 , eprint=

    2020

Showing first 80 references.