Pith. sign in

REVIEW 3 major objections 4 minor 30 references

Toward Efficient Uncertainty in LLMs through Evidential Knowledge Distillation

T0 review · 3 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read A Dirichlet-headed student LLM can inherit a sampling-based teacher's uncertainty and run a single forward pass, matching accuracy and calibration on classification benchmarks.

desk verdict A practical distillation recipe with honest limitations; the core speed/calibration result holds, but the OOD attribution is under-determined by the current experiments. read the letter →

arxiv 2507.18366 v2 pith:2LT26V3Z submitted 2025-07-24 cs.LG stat.ML

Lakshmana Sri Harsha Nemani , P.K. Srijith , Tomasz Kuśmierczyk This is my paper · ORCID
classification cs.LGstat.ML
keywords knowledgedistillationevidentialdeeplearninguncertaintyquantificationlargelanguagemodelsDirichletdistributionLoRABayesianpromptensemblesout-of-distributiondetection
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

This paper tries to show that the expensive uncertainty estimates of sampling-based LLM teachers can be compressed into a single forward pass without losing predictive performance or calibration. It compares two student designs: a standard softmax head that learns only the teacher's mean probabilities, and a Dirichlet (evidential) head that also learns the shape of the teacher's predictive distribution. On four text-classification benchmarks, distilled students match or slightly beat their teachers on accuracy, calibration, NLL, and Brier score, while cutting inference time by 11x to 36x. The Dirichlet student also separates aleatoric from epistemic uncertainty and shows strong out-of-distribution detection, which the softmax student cannot do.

What carries the argument

The Dirichlet output head parameterized by concentration parameters $\alpha_c = 1 + \mathrm{softplus}(z_c)$, with total evidence $\alpha_0 = \sum_c \alpha_c$; the predictive mean is $\alpha_c/\alpha_0$ and the concentration measures epistemic confidence. A negative log-likelihood loss on teacher probability samples trains this head, so the student's Dirichlet shape, not just its mean, encodes the teacher's uncertainty. Entropy decomposition via the law of total entropy splits total, aleatoric ($\mathbb{E}_p[H[Y|p]]$), and epistemic (mutual information) components. LoRA adapters and NLL-based early stopping carry the distillation.

What would settle it

Train a Dirichlet student on data with known, controllable label noise and distribution shift, and compare its estimated epistemic uncertainty against the true shift magnitude; if the student's epistemic component does not track the true shift while the teacher's does, the claim that the single-pass student preserves the uncertainty decomposition is falsified.

Watch

Extended reading notes

Core claim

The central claim is that distillation based on evidential learning lets a single-pass LLM student recover both the predictions and the uncertainty decomposition of a multi-pass Bayesian or prompt-ensemble teacher. Teacher hypotheses $\{\theta_n\}$ with weights $w_n$ define a Monte Carlo predictive distribution; the softmax student fits the weighted mean $\bar{p}_{T,c}$, while the Dirichlet student fits the parameters $\alpha$ of a Dirichlet distribution by maximizing the log-likelihood of the sampled probability vectors. The paper finds that the Dirichlet student matches teacher accuracy (and exceeds it on some datasets), reduces expected calibration error substantially in several cases, and produces higher out-of-distribution AUROC than the teacher itself, while running a single forward pass.

Load-bearing premise

The teacher's Monte Carlo samples and prompt weights are assumed to fully capture its predictive uncertainty, so a student trained on only those samples, from a single prompt, can recover the same aleatoric and epistemic decomposition.

Editorial extensions

If this is right

  • Distilled students run inference in a single forward pass yet match or exceed multi-pass teachers on classification accuracy, ECE, NLL, and Brier score (Tables 2-3).
  • The Dirichlet student is the only variant that retains a usable epistemic/aleatoric split, enabling out-of-distribution detection with AUROC up to 1.00 on the tested shifts (Table 4).
  • Speed-ups of 11x to 36x over the BayesPE teacher are reported on the four datasets, with larger relative gains on larger datasets.
  • The distillation process itself can improve on the teacher: students beat their teacher on accuracy in some datasets, attributed to early stopping and the regularizing effect of the Dirichlet distribution.
  • Softmax students remain competitive when the teacher is Laplace-based, suggesting a simpler parameterization suffices when the teacher's predictive distribution is well-behaved.

Reading between the lines

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

  • Going beyond the paper, a testable extension is to compare the Dirichlet student's epistemic uncertainty against a gold-standard measure, such as disagreement among independently trained models, to check whether the transferred epistemic component is faithful rather than merely well-calibrated in aggregate.
  • Because the teacher's sampled prompts and weights define the distillation target, and because the student is initialized from the teacher's LoRA matrices, the student may inherit teacher-specific biases; a clean test would distill from a teacher with a known analytic predictive distribution and check exact recovery.
  • The authors restrict experiments to classification; one could extend the evidential head to regression or sequence outputs, for instance with an evidential Gaussian output layer, to see whether single-pass epistemic uncertainty transfers in those settings as well.
  • The regularization study shows the optimal fixed $\alpha_0$ varies by dataset, so a practical follow-up is a hyperprior over $\alpha_0$ that adapts per domain without manual tuning.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper proposes distilling sampling-based uncertainty-aware LLM teachers (BayesPE and Laplace LoRA) into single-pass students with either softmax or Dirichlet output heads, using LoRA fine-tuning and an evidential negative-log-likelihood distillation loss. Experiments on four text classification datasets compare accuracy, ECE, NLL, Brier score, OOD detection AUROC, and inference time. The central empirical claim is that Dirichlet students match or improve on the teacher's predictive accuracy and uncertainty quantification while requiring only a single forward pass, with 11x-36x speedups reported for the BayesPE teacher.

Significance. If the result holds, the approach is a practical step toward making Bayesian and prompt-ensemble uncertainty quantification affordable at LLM inference time, and the Dirichlet head's explicit aleatoric/epistemic decomposition is an attractive design. The paper has concrete strengths: public code, two teacher families, four datasets, three student seeds, detailed ablations on regularization and prompt weights, and an honest acknowledgment of the YouTube calibration exception and of the 'side effect' explanation for OOD gains. The main weakness is attribution: the current experiments do not cleanly separate the student's uncertainty behavior that comes from the teacher's predictive distribution from behavior induced by shared initialization, label-based early stopping, or the Dirichlet architecture. The requested control experiments are feasible and would substantially strengthen the central claim.

major comments (3)
  1. [Section 3.1 and Tables 2-3] The Laplace-teacher experiments are confounded by the student initialization described in Section 3.1: the student copies the teacher's LoRA adapters and shares the same Mistral backbone. As a result, the student starts at the teacher's fine-tuned solution, so the near-parity in accuracy, ECE, NLL, and Brier reported in Table 3 is expected even if the distillation losses contributed little. To make the uncertainty-transfer claim load-bearing, the authors should add a control with random LoRA initialization for the student and, ideally, a student distilled from a different LoRA seed of the teacher; without this, the Laplace results cannot be attributed to the distilled predictive distribution.
  2. [Section 4.3 and Table 4] The OOD detection evidence in Section 4.3 and Table 4 does not isolate the teacher's uncertainty. Inputs from SST2, YouTube, and Yahoo never appear in the distillation losses (Eqs. 2 and 3), so the high AUROC values of the Dirichlet student may reflect the Dirichlet head's inductive bias or early-stopping dynamics rather than transferred uncertainty; the paper itself says the student's better-than-teacher OOD behavior can be explained only as a side effect of the distillation process and model architecture. A control Dirichlet student trained on the same in-domain labels without any teacher signal, and preferably a student distilled from a teacher with permuted uncertainty estimates, is required before Table 4 can support the claim that uncertainty is inherited from the teacher.
  3. [Section 3.3 and Tables 2-4] The early-stopping rule in Section 3.3 uses ground-truth labels through the training-data NLL, so model selection is not purely teacher-driven. This is a second channel, in addition to the initialization issue, through which the student's improved ECE, NLL, and OOD AUROC can arise independently of the teacher's predictive distribution. The requested label-only and random-initialization controls should use the same early-stopping rule, and the paper should report the early-stopping metric on those controls, so the contribution of checkpoint selection is quantified.
minor comments (4)
  1. [Table 2, YouTube row] The Dirichlet student's ECE of 0.097 +/- 0.023 is about three times worse than the BayesPE teacher's 0.031, while the Softmax student's ECE is 0.015; the conclusion that Dirichlet students achieve calibration comparable to the teacher should explicitly except this dataset.
  2. [Tables 2 and 3] The teacher rows are single point estimates with no standard deviations, so the 'comparable' claims rest on deterministic teacher evaluations; a few repeated teacher runs or an explicit statement that the teacher metrics are deterministic under the fixed trained model would strengthen the comparison.
  3. [Section 4.1] The 11x-36x speed-up numbers are reported only for the BayesPE teacher; the Laplace teacher's timings are omitted because of the conservative 10k-sample setting. The abstract and conclusion should state that the speed-up claim is for the BayesPE teacher.
  4. [Supplement, Algorithm 1] The student is always queried with the highest-weight prompt Q_best, a design choice that is not highlighted in the main text; given the prompt-sensitivity results in Section 4.5, this choice and its effect on the headline results should be discussed.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: distillation targets are teacher-sampled distributions, and all headline metrics are external held-out evaluations.

full rationale

The paper's derivation chain is not circular. The student objectives in Eqs. (2) and (3) are standard cross-entropy and Dirichlet negative log-likelihood losses defined on teacher Monte Carlo samples; the quantities reported as results (test-set accuracy, ECE, NLL, Brier, OOD AUROC) are computed from student outputs on held-out data that never enter the training objective. No fitted parameter is renamed as a prediction: early stopping uses training-data NLL, and the alpha0=5K variant is presented as a hyperparameter comparison in Table 5, not as a predicted output. The teachers' own scores are taken from external evaluation or reproduced independently, and no load-bearing argument rests on a self-citation or on an author-imported uniqueness theorem. The paper explicitly concedes that student improvements over the teacher 'cannot arise from knowledge transfer alone' and attributes them to early stopping and architectural inductive bias, which removes any claim that those gains are distilled predictions. Copying the teacher's LoRA adapters as an initialization is a warm-start confound, but the student is still optimized under a distinct distillation loss and is assessed on external benchmarks, so it does not make the comparison true by construction. Accordingly, no circular step can be exhibited under the required standard.

Assumptions & free parameters 1 free parameters · 5 assumptions · 0 invented entities

The central claim depends on standard uncertainty-decomposition math and on domain assumptions about teacher-sample fidelity, the adequacy of the Dirichlet family, and the validity of teacher-based initialization and early stopping. No invented entities and no free parameters beyond standard training hyperparameters are introduced; the ablation value (α0=5K) is test-set-selected and tracked in red flags.

free parameters (1)
  • Fixed global precision α0 = 5K = 5K (test-best on SST2 and YouTube)
    In Section 4.4 and Table 5, the concentrated regularizer α0=5K yields the best ECE/NLL on SST2 and YouTube, but the value is chosen post hoc from test-set comparison rather than tuned on a separate validation set.
assumptions (5)
  • standard math The law of total entropy decomposition, H[Y|x,D] = E_θ[H[Y|x,θ]] + I[Y;θ|x,D], gives a valid separation of aleatoric and epistemic uncertainty for the teacher and student models.
    Invoked in Section 2.1 and used to compute epistemic uncertainty in Section 3.2 and in Tables 4 and Figures 2.
  • domain assumption Monte Carlo samples from the teacher (Eq. 1) accurately represent the full predictive distribution of the sampling-based model.
    The distillation losses in Eqs. (2) and (3) are trained on these samples; if samples are biased or too few, the student's uncertainty estimates inherit that bias.
  • domain assumption The Dirichlet distribution with α = 1 + softplus(z) is an adequate second-order representation of the teacher's predictive distribution, so minimizing the negative log-likelihood of teacher samples recovers both mean probabilities and their dispersion.
    Used in Section 3.2 to define the evidential student; the claim that epistemic uncertainty can be recovered from α0 depends on the Dirichlet being a good variational family for the teacher's distribution.
  • domain assumption Initializing the student with the teacher's LoRA matrices still counts as distillation and does not trivially force the reported match in performance.
    Section 3.1 states the student is initialized by copying the teacher's low-rank matrices; the evaluation treats the resulting model as a distilled student rather than as the teacher with a retrained head.
  • domain assumption Early stopping based on training-set NLL (Section 3.3) selects checkpoints with good generalization, not merely good fit to the training data.
    The paper uses per-epoch NLL on training data for early stopping; the validity of the reported test results assumes this does not overfit to teacher noise.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Toward Efficient Uncertainty in LLMs through Evidential Knowledge Distillation." pith.science (2026). https://pith.science/paper/2LT26V3Z

@misc{pith2026250718366,
  author       = {Pith},
  title        = {Pith review of: Toward Efficient Uncertainty in LLMs through Evidential Knowledge Distillation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2LT26V3Z}},
  note         = {Machine review of arXiv:2507.18366}
}
read the original abstract

Accurate uncertainty quantification remains a key challenge for standard LLMs, prompting the adoption of Bayesian and ensemble-based methods. However, such methods typically necessitate computationally expensive sampling, involving multiple forward passes to effectively estimate predictive uncertainty. In this paper, we introduce an approach enabling uncertainty estimation in LLMs without incurring the heavy inference latency typically associated with sampling methods. Specifically, we distill uncertainty-aware teachers - originally requiring multiple forward passes - into single-pass students, fine-tuned using LoRA. We compare two distinct distillation strategies: one in which the student employs traditional softmax-based outputs, and another in which the student leverages Dirichlet-distributed outputs to explicitly model epistemic uncertainty via evidential learning. Empirical evaluation on classification tasks demonstrate that such students can achieve comparable predictive and uncertainty quantification performance relative to their teachers, while requiring only a single forward pass.

Figures

Figures reproduced from arXiv: 2507.18366 by the authors.

Figure 1
Figure 1. Dirichlet distributions on the 2-simplex illustrating uncertainty quantification for categorical probabilities. Each panel illustrates different concentration parameters α. Blue heatmaps represent the probability densities, red circles indicate the expected probability vectors (means), and blue dots show samples from the distributions - each representing a possible "true" probability vector for the three categories.… view at source ↗
Figure 2
Figure 2. Predictive uncertainty distributions for the BayesPE teacher (blue), Dirichlet student (orange) and Softmax student (green). Each row shows the empirical distri￾butions of (left): total predictive entropy, (middle): mutual information (epistemic uncertainty), and (right): expected conditional entropy (aleatoric uncertainty) on in￾domain Amazon reviews (Row 1) and three OOD datasets: SST2 (Row 2), YouTube (Row 3) and… view at source ↗

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

30 extracted references · 20 canonical work pages

  1. [23]

    arXiv preprint arXiv:2505.11731 (2025)

    Vejendla, H., Shi, H., Wang, Y., Zhang, T., Zhang, H., Wang, H.: Efficient uncer- tainty estimation via distillation of bayesian large language models. arXiv preprint arXiv:2505.11731 (2025)

    work page arXiv 2025

  2. [25]

    In: Kiyavash, N., Mooij, J.M

    Wang, H., Ji, Q.: Beyond dirichlet-based models: When bayesian neural networks meet evidential deep learning. In: Kiyavash, N., Mooij, J.M. (eds.) Proceedings of the Fortieth Conference on Uncertainty in Artificial Intelligence. Proceedings of Machine Learning Research, vol. 244, pp. 3643–3665. PMLR (15–19 Jul 2024)

    work page 2024

  3. [1]

    In: 2015 IEEE 14th international conference on machine learning and applications (ICMLA)

    Alberto, T.C., Lochter, J.V., Almeida, T.A.: Tubespam: Comment spam filtering on youtube. In: 2015 IEEE 14th international conference on machine learning and applications (ICMLA). pp. 138–143. IEEE (2015)

    work page 2015

  4. [2]

    In: International conference on machine learning

    Blundell, C., Cornebise, J., Kavukcuoglu, K., Wierstra, D.: Weight uncertainty in neural network. In: International conference on machine learning. pp. 1613–1622. PMLR (2015)

    2015

  5. [3]

    In: Proceedings of the 41stInternationalConferenceonMachineLearning.ProceedingsofMachineLearn- ing Research, vol

    Fang, L., Chen, Y., Zhong, W., Ma, P.: Bayesian knowledge distillation: A Bayesian perspective of distillation with uncertainty quantification. In: Proceedings of the 41stInternationalConferenceonMachineLearning.ProceedingsofMachineLearn- ing Research, vol. 235, pp. 12935–12956. PMLR (21–27 Jul 2024)

    work page 2024

  6. [4]

    Pattern Recognition Letters 27(8), 861–874 (2006).https://doi.org/https://doi.org/10.1016/j.patrec

    Fawcett, T.: An introduction to roc analysis. Pattern Recognition Letters 27(8), 861–874 (2006).https://doi.org/https://doi.org/10.1016/j.patrec. 2005.10.010, rOC Analysis in Pattern Recognition

    doi:10.1016/j.patrec 2006

  7. [5]

    In: Graham, Y., Purver, M

    Gao, X., Zhang, J., Mouatadid, L., Das, K.: SPUQ: Perturbation-based uncer- tainty quantification for large language models. In: Graham, Y., Purver, M. (eds.) Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers). pp. 2336–2346. Association for Computational Linguistics (Mar 202...

    doi:10.18653/v1/2024 2024

  8. [6]

    In: The Twelfth International Conference on Learning Representations (2024)

    Harrison, J., Willes, J., Snoek, J.: Variational bayesian last layers. In: The Twelfth International Conference on Learning Representations (2024)

    work page 2024

Show all 30 references
  1. [7]

    In: Proceedings of the 25th In- ternational Conference on World Wide Web

    He, R., McAuley, J.: Ups and downs: Modeling the visual evolution of fash- ion trends with one-class collaborative filtering. In: Proceedings of the 25th In- ternational Conference on World Wide Web. WWW ’16 (Apr 2016).https: //doi.org/10.1145/2872427.2883037

    2016

  2. [8]

    Hinton, G., Vinyals, O., Dean, J.: Distilling the knowledge in a neural network (2015)

    2015

  3. [9]

    In: Proceedings of the 41st International Conference on Machine Learning

    Hou, B., Liu, Y., Qian, K., Andreas, J., Chang, S., Zhang, Y.: Decomposing un- certainty for large language models through input clarification ensembling. In: Proceedings of the 41st International Conference on Machine Learning. ICML’24, JMLR.org (2024)

    2024

  4. [10]

    Hu, E.J., Shen, Y., Wallis, P., Allen-Zhu, Z., Li, Y., Wang, S., Wang, L., Chen, W.: Lora: Low-rank adaptation of large language models (2021)

    2021

  5. [11]

    Jiang, A.Q., Sablayrolles, A., Mensch, A., Bamford, C., Chaplot, D.S., de las Casas, D., Bressand, F., Lengyel, G., Lample, G., Saulnier, L., Lavaud, L.R., Lachaux, M.A., Stock, P., Scao, T.L., Lavril, T., Wang, T., Lacroix, T., Sayed, W.E.: Mistral 7b (2023)

    2023

  6. [12]

    Li, Y., Rügamer, D., Bischl, B., Rezaei, M.: Calibrating llms with information- theoretic evidential deep learning (2025),https://arxiv.org/abs/2502.06351

    2025 arXiv

  7. [13]

    In: Bengio, S., Wallach, H., Larochelle, H., Grauman, K., Cesa-Bianchi, N., Garnett, R

    Malinin, A., Gales, M.: Predictive uncertainty estimation via prior networks. In: Bengio, S., Wallach, H., Larochelle, H., Grauman, K., Cesa-Bianchi, N., Garnett, R. (eds.) Advances in Neural Information Processing Systems. vol. 31. Curran Associates, Inc. (2018)

    2018

  8. [14]

    Springer (1996)

    Neal, R.M.: Bayesian Learning for Neural Networks. Springer (1996)

    1996

  9. [15]

    In: Sixth Symposium on Advances in Approximate Bayesian Inference - Non Archival Track (2024) 18 L.S.H

    Onal, E., Flöge, K., Caldwell, E., Sheverdin, A., Fortuin, V.: Gaussian stochastic weight averaging for bayesian low-rank adaptation of large language models. In: Sixth Symposium on Advances in Approximate Bayesian Inference - Non Archival Track (2024) 18 L.S.H. Nemani et al

    2024

  10. [16]

    In: 6th International Conference on Learning Representations (2018)

    Ritter, H., Botev, A., Barber, D.: A scalable laplace approximation for neural networks. In: 6th International Conference on Learning Representations (2018)

    2018

  11. [17]

    In: Proceedings of the 32nd International Conference on Neural Information Processing Systems

    Sensoy, M., Kaplan, L., Kandemir, M.: Evidential deep learning to quantify classifi- cation uncertainty. In: Proceedings of the 32nd International Conference on Neural Information Processing Systems. p. 3183–3193. NIPS’18 (2018)

    2018

  12. [18]

    In: The Thirty-ninth Annual Confer- ence on Neural Information Processing Systems (2025)

    Shi, H., Wang, Y., Han, L., Zhang, H., Wang, H.: Training-free bayesianization for low-rank adapters of large language models. In: The Thirty-ninth Annual Confer- ence on Neural Information Processing Systems (2025)

    2025

  13. [19]

    In: Yarowsky, D., Baldwin, T., Korhonen, A., Livescu, K., Bethard, S

    Socher, R., Perelygin, A., Wu, J., Chuang, J., Manning, C.D., Ng, A., Potts, C.: Recursive deep models for semantic compositionality over a sentiment treebank. In: Yarowsky, D., Baldwin, T., Korhonen, A., Livescu, K., Bethard, S. (eds.) Proceed- ings of the 2013 Conference on ...

    2013

  14. [20]

    arXiv preprint arXiv:2505.23854 (2025)

    Tao,L.,Yeh,Y.F.,Dong,M.,Huang,T.,Torr,P.,Xu,C.:Revisitinguncertaintyes- timation and calibration of large language models. arXiv preprint arXiv:2505.23854 (2025)

    2025 arXiv

  15. [21]

    In: Ku, L.W., Martins, A., Srikumar, V

    Tonolini,F.,Aletras,N.,Massiah,J.,Kazai,G.:Bayesianpromptensembles:Model uncertainty estimation for black-box large language models. In: Ku, L.W., Martins, A., Srikumar, V. (eds.) Findings of the Association for Computational Linguistics: ACL 2024. pp. 12229–12272. Association...

    2024 doi

  16. [22]

    arXiv preprint arXiv:2505.19060 (May 2025)

    Vashurin, R., Goloburda, M., Nakov, P., Panov, M.: Uncertainty-line: Length-invariant estimation of uncertainty for large language models. arXiv preprint arXiv:2505.19060 (May 2025)

    2025

  17. [24]

    Grundlehren der mathematischen Wissenschaften, Springer Berlin Heidelberg (2008)

    Villani, C.: Optimal Transport: Old and New. Grundlehren der mathematischen Wissenschaften, Springer Berlin Heidelberg (2008)

    2008

  18. [26]

    In: Globerson, A., Mackey, L., Belgrave, D., Fan, A., Paquet, U., Tomczak, J., Zhang, C

    Wang, Y., Shi, H., Han, L., Metaxas, D., Wang, H.: Blob: Bayesian low-rank adap- tation by backpropagation for large language models. In: Globerson, A., Mackey, L., Belgrave, D., Fan, A., Paquet, U., Tomczak, J., Zhang, C. (eds.) Advances in Neural Information Processing Syste...

    2024 doi

  19. [27]

    In: Proceedings of the 28th International Conference on International Confer- ence on Machine Learning

    Welling, M., Teh, Y.W.: Bayesian learning via stochastic gradient langevin dynam- ics. In: Proceedings of the 28th International Conference on International Confer- ence on Machine Learning. p. 681–688. ICML’11, Omnipress, Madison, WI, USA (2011)

    2011

  20. [28]

    arXiv preprint arXiv:2503.00172 (2025)

    Xia, Z., Xu, J., Zhang, Y., Liu, H.: A survey of uncertainty estimation methods on large language models. arXiv preprint arXiv:2503.00172 (2025)

    2025 arXiv

  21. [29]

    In: Kim, B., Yue, Y., Chaudhuri, S., Fragkiadaki, K., Khan, M., Sun, Y

    Yang, A., Robeyns, M., Wang, X., Aitchison, L.: Bayesian low-rank adaptation for large language models. In: Kim, B., Yue, Y., Chaudhuri, S., Fragkiadaki, K., Khan, M., Sun, Y. (eds.) International Conference on Representation Learning. vol. 2024, pp. 1812–1842 (2024)

    2024

  22. [30]

    Zhang, X., Zhao, J., LeCun, Y.: Character-level convolutional networks for text classification. Advances in neural information processing systems28(2015) Uncertainty in LLMs through Evidential Knowledge Distillation 19 Supplementary Material Toward Efficient Uncertainty in LLM...

    2015

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.