Pith. sign in

REVIEW 3 major objections 3 minor 35 references

Black-Box Edge AI Model Selection with Conformal Latency and Accuracy Guarantees

T0 review · 3 major / 3 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 Black-box edge AI model selection can be certified to meet both expected-loss and deadline-violation requirements.

desk verdict A nice integration of conformal risk control and wireless delay modeling, but the latency guarantee rests on a conditional/marginal confusion in the order-statistic lemma, and the model selection makes it worse. read the letter →

arxiv 2506.11391 v1 pith:A2W5HGBR submitted 2025-06-13 cs.IT cs.NImath.IT

Anders E. Kal{o}r , Tomoaki Ohtsuki This is my paper
classification cs.ITcs.NImath.IT
keywords edgeAIconformalriskcontrolmodelselectiondeadlineguaranteewirelesschannel6Gorderstatisticsimageclassification
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 proposes a way to choose, from a collection of black-box compression and inference models, the combination that meets two hard requirements for wireless edge AI: the expected loss of the returned prediction set, conditioned on meeting the deadline, is at most $\alpha$, and the probability of missing the deadline is at most $\beta$. The selection is made without knowing the data distribution or model internals, using a labeled calibration set for the loss guarantee and an unlabeled calibration set for the delay guarantee. The paper presents a fixed scheme that picks one encoder/decoder and one edge model from channel statistics, and a dynamic scheme that re-selects the edge model after observing the instantaneous uplink rate. If correct, this gives deadline-constrained applications such as automated driving and robotics a principled way to trade compression quality against model complexity while keeping formal end-to-end guarantees.

What carries the argument

The central objects are composite black-box models $g_{l,k}$ that chain one encoder/decoder pair to one inference model; each combination is calibrated independently because the intermediate representations differ across encoder/decoder pairs. The loss guarantee comes from the conformal risk control threshold of Lemma 1, which makes the empirical loss on the labeled calibration set small enough to compensate for finite calibration size. The deadline guarantee comes from Proposition 1, which combines a Rayleigh-fading conditional success probability with a Boole's-inequality lower bound $\frac{n+m}{N_U+1}-1$ on the joint probability that a new sample's uplink and downlink message sizes fall below order statistics of the unlabeled calibration set. The dynamic scheme uses Proposition 2, the same bound conditioned on the observed uplink rate, to re-pick the edge model at inference time.

What would settle it

Take a fixed unlabeled calibration set, compute Proposition 1's bound for a chosen model, then measure the empirical deadline violation probability on a large fresh set of independent samples; if the measured rate systematically exceeds the bound, the fixed-set application of the rank-uniform argument fails.

Watch

Extended reading notes

Core claim

For each composite black-box model $g_{l,k}=f_k(d_l(e_l(X)))$ that chains an encoder/decoder pair to an inference model, the paper calibrates a threshold $\lambda_{l,k}$ so that the expected loss satisfies $\mathbb{E}[\ell(\Gamma(X),Y)]/(1-\beta)\le\alpha$, which implies the conditional requirement $\mathbb{E}[\ell(\Gamma(X),Y)\mid T_{\mathrm{tot}}\le T]\le\alpha$. It then proves an upper bound on the deadline violation probability from the order statistics of the uplink and downlink message sizes on an unlabeled calibration set, giving $\Pr(T_{\mathrm{tot}}>T)\le\min_{n,m}(1-e^{\bar\beta_{\mathrm{cal}}(l,k,n,m)}(\frac{n+m}{N_U+1}-1))$. The fixed scheme selects the model combination with the smallest estimated prediction-set size among those satisfying the bound, and the dynamic scheme conditions the same bound on the observed uplink rate and re-selects the edge model at inference time. Numerical results on an ImageNet classification task with WebP compression and EfficientNetV2 classifiers show the guarantees met where feasible, with prediction-set size decreasing as SNR increases.

Load-bearing premise

The deadline guarantee depends on the unlabeled calibration set being a random draw at the moment the bound is used; the algorithms, however, use a single fixed set, so the numerical certificate is only as reliable as that random-draw assumption.

Editorial extensions

If this is right

  • An operator can certify before deployment, without any white-box knowledge, whether a given model combination meets both the loss and deadline requirements; the certificate is computed offline from calibration data and channel statistics.
  • The dynamic scheme keeps the conformal thresholds precomputed offline, so the only online work is evaluating the delay bound and comparing the $K$ edge models against the observed uplink rate.
  • Because the loss guarantee applies to any loss that cannot increase when the prediction set is enlarged, the same framework covers misclassification probability and other risk measures such as false negative rate.
  • When no available combination meets the deadline requirement, the procedures still return the combination with the smallest provable violation probability, providing graceful degradation at low SNR.
  • At higher SNR the schemes select larger models and produce smaller prediction sets, turning favorable channel conditions directly into more informative predictions.

Reading between the lines

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

  • I read Lemma 3 as a joint statement over the random calibration set and the new sample; the algorithms fix $U$ in advance, so strictly the deadline bound is an approximation for any particular calibration set. A resampling or jackknife correction over $U$ would turn the guarantee into a conditional one, but that step is not in the paper.
  • The paper's own numerical results note that the delay bounds are conservative because they treat uplink and downlink delays as correlated; an independence-aware bound could select larger models at high SNR, where the reported violation probability sits well below $\beta$.
  • Because the method treats each composite model as a black box, the same selection procedure extends to any encoder or feature extractor and any inference model that emits confidence scores, including models with variable-length outputs beyond the ImageNet classification setting.
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

3 major / 3 minor

Summary. The paper considers wireless edge inference where a sensor transmits an encoded observation to an edge server, which runs a black-box classifier and sends a prediction set back within a frame deadline. It proposes to select, from L encoder/decoder pairs and K inference models, the combination that minimizes the prediction-set size subject to an expected-loss constraint, handled by conformal risk control, and a deadline-violation constraint, handled by order-statistic bounds on uplink and downlink message sizes. Two schemes are presented: a fixed offline selection and a channel-adaptive dynamic selection, the latter with a variant allowing prediction-set truncation. The numerical evaluation uses ImageNet with WebP compression and EfficientNetV2 classifier models.

Significance. If valid, the framework would be a useful contribution to 6G edge-AI literature: it is model-agnostic, handles variable compression rates and random channel conditions, and the numerical setup is realistic. The conformal risk-control step for a single pre-specified composite model is standard and correctly stated, and the use of separate labeled and unlabeled calibration sets is a sensible design. However, the paper's main novelty is the deadline-violation bound, and that bound is not mathematically valid as stated. The flaws are load-bearing: Proposition 1 and Proposition 2 fail, and the selection algorithms do not preserve the conformal guarantee. The contribution therefore reduces to an empirical study accompanied by a guarantee that is not established.

major comments (3)
  1. [Appendix C, Lemma 3; Proposition 1] The rank-uniform identity used in Lemma 3 is false conditional on the realized calibration set. The statement Pr(Dul,t ≤ \bar D_ul(n) | gl,k) = n/(NU+1) holds only when the probability is taken jointly over the new sample X and the random calibration set U; for the fixed U used in Proposition 1, the conditional probability is F_{gl,k}(\bar D_ul(n)), which is unknown and need not equal n/(NU+1). Because Algorithm 1 computes the Proposition 1 bound for a particular U and then selects the model minimizing that U-dependent bound, the returned \bar P_{l,k} is not a valid upper bound on Pr(Ttot,t > T | gl,k). Proposition 2 uses the same Lemma 3, so the dynamic scheme inherits the error. Consequently, the deadline constraint in Eq. (6c) is unproven for both schemes.
  2. [Algorithms 1 and 2; Section III] Even setting Lemma 3 aside, the selection rules in Algorithms 1 and 2 do not preserve the conformal guarantee. The thresholds λ_{l,k} are computed from the labeled set D, and the same thresholds enter the U-based delay bounds and prediction-set-size estimates used for selection; the selected model is therefore a function of D as well as U. Lemma 1 is a marginal guarantee for a pre-specified model and does not automatically hold for a model chosen by a D-dependent rule. Section III explicitly defers learn-then-test for joint model/threshold search, but that is precisely the correction needed here. Without a selection-adjusted argument, the expected-loss constraint in Eq. (6b) is also not guaranteed for the returned model.
  3. [Appendix D, Lemma 4; Proposition 2] Lemma 4 is not valid when the uplink transmission consumes the entire available time. The chosen maximizer φ = T − τul,l − Dul,t/Rul,t lies outside the feasible interval [τfk, T − τul,l] whenever Dul,t/Rul,t ≥ T − τul,l − τfk; in that regime the conditional probability of meeting the deadline is 0, while the exponential lower bound in Lemma 4 exceeds 1. Proposition 2 then reports 1 − e^{\hat β}(...) which can be negative, and a negative number is not a valid upper bound on a probability. This affects the dynamic scheme at low SNR.
minor comments (3)
  1. [Appendix B, Eq. (22)] The displayed derivative of 2^{D/(B(T−φ−τ))} is not correct: it should contain a factor ln(2), not log2(Dul,t) or log2(Ddl,t). Since the derivative is used only as motivation and not in the final bound, this appears to be a typographical error, but it should be corrected.
  2. [Appendix B, proof of Lemma 2] In the paragraph after Eq. (21), the sentence stating that 'Tdl,t is independent of Ddl,t' should presumably say that Tdl,t is independent of Dul,t; as written it is inconsistent with the conditioning used in the expansion.
  3. [Section VI-B, Fig. 2(b)] The text says the proposed schemes achieve the smallest possible deadline violation probability at low SNRs, but this is only an empirical observation. It does not compensate for the missing mathematical guarantee, and the caption/abstract should not imply a strict bound in the regime where no candidate satisfies the requirement.

Circularity Check

1 steps flagged · score 6.0 of 10

The conformal loss guarantee is self-contained, but the deadline-violation guarantee for the selected model reduces to an in-sample minimum of the same U-computed bounds, so the central latency claim is partially circular.

  1. fitted input called prediction [Section IV, Algorithm 1 lines 7–10; Proposition 1; Appendix C Lemma 3]
    "Pr(Dul,t ≤ ¯Dul,l(n), Ddl,t ≤ ¯Ddl,l,k(m) | gl,k) ≥ n+m/(NU+1)−1, where the probability is over X ∼ PX. ... Compute ¯Pl,k as the resulting delay violation probability upper bound in Proposition 1 using U. ... if ( ¯Pl,k ≤ β and ¯Γl,k < ¯Γ∗) or ( ¯P ∗ ≥ β and ¯Pl,k < ¯P ∗) then g∗ ← gl,k; λ∗ ← λl,k."

    The rank-uniform identity in Lemma 3 holds only marginally over a random calibration set U and the new sample; conditional on the realized U used in Algorithm 1, Pr(D ≤ D_(n) | U) = F(D_(n)) is unknown and can be far from n/(NU+1). Proposition 1 nevertheless treats the U-order statistics as known deterministic bounds, and Algorithm 1 selects g* by minimizing the very same U-computed ¯P over L×K candidates. The returned ¯P* is thus a minimum of in-sample, mutually correlated estimates, not an upper bound on the true Pr(Ttot > T) of the selected model. The claimed guarantee Pr(Ttot > T) ≤ β for the returned model is therefore the same data-dependent quantity used for selection, i.e., a fitted input presented as a prediction.

full rationale

The expected-loss half of the framework is not circular: Lemma 1 is the standard conformal risk control guarantee applied independently to each composite model gl,k, with the calibration dataset D and a finite-sample correction; the conditioning adjustment ε = α(1−β) is a valid sufficient condition via the law of total expectation. The numerical evaluation against ImageNet and external baselines gives independent empirical content to the loss claim. The circularity is confined to the deadline-violation half. Lemma 3 borrows the conformal rank identity (citing [7, Appendix D]) but applies it as if the order statistics of the fixed calibration set U were valid deterministic quantiles; the identity needs the probability to be taken jointly over U and the new sample. More importantly, Algorithm 1 uses the same U to compute every candidate's delay bound and then selects the model minimizing that bound, and Algorithm 2 does the same for the conditional bound. No learn-then-test or selection-adjusted (e.g., split-conformal or Bonferroni) correction is provided, even though Section III notes learn-then-test as the tool for exactly this joint search. Consequently, the central claim of a strict statistical guarantee on latency for the returned model reduces to an in-sample minimum of U-computed bounds and is not established. The self-citations ([2], [16], [17], [21]) are background and not load-bearing; there is no imported uniqueness theorem or ansatz-by-citation. Score 6 reflects a partially circular central claim: the loss guarantee is independent, but the deadline guarantee is effectively fitted on the selection data.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The paper adds two algorithm components, conformal risk control and an order-statistic delay bound, over a standard system model. The main unearned input is the conditional rank-uniform assumption, which is invalid for fixed calibration sets.

free parameters (2)
  • Conformal thresholds λl,k = data-dependent
    Chosen via Lemma 1 on the labeled calibration set D for each composite model; not externally specified. These thresholds set the prediction set sizes and affect the downlink data size distribution.
  • Encoder/decoder computation times τul,l = 10.0, 12.5, 15.0, 17.5 ms
    Table I; described as selected for illustrative purposes based on rough estimates from ImageNet, not measured. These times directly affect which model combinations can meet the deadline.
assumptions (4)
  • standard math Conformal risk control theorem (Lemma 1) from Angelopoulos et al. [8] is valid.
    Invoked in Section III to set thresholds λl,k that control expected loss on new data; accepted as a proven theorem.
  • domain assumption Calibration and test samples are i.i.d. from unknown PXY; labeled set D and unlabeled set U are disjoint and independent.
    Stated in Section II-C; required for both conformal calibration and order-statistic bounds.
  • domain assumption Uplink and downlink channels are Rayleigh fading with known average SNR and independent across frames.
    Used in Lemma 2 and Lemma 4 to express transmission success probability as an exponential function of message sizes.
  • ad hoc to paper Rank-uniform property of message sizes holds conditionally on the fixed unlabeled set U.
    Appendix C, Lemma 3: 'data sizes of a new sample are equally likely to fall anywhere between the calibration samples'. This is only true marginally over random U, not for the particular fixed U used in model selection.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Black-Box Edge AI Model Selection with Conformal Latency and Accuracy Guarantees." pith.science (2026). https://pith.science/paper/A2W5HGBR

@misc{pith2026250611391,
  author       = {Pith},
  title        = {Pith review of: Black-Box Edge AI Model Selection with Conformal Latency and Accuracy Guarantees},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/A2W5HGBR}},
  note         = {Machine review of arXiv:2506.11391}
}
read the original abstract

Edge artificial intelligence (AI) will be a central part of 6G, with powerful edge servers supporting devices in performing machine learning (ML) inference. However, it is challenging to deliver the latency and accuracy guarantees required by 6G applications, such as automated driving and robotics. This stems from the black-box nature of ML models, the complexities of the tasks, and the interplay between transmitted data quality, chosen inference model, and the random wireless channel. This paper proposes a novel black-box model selection framework for reliable real-time wireless edge AI designed to meet predefined requirements on both deadline violation probability and expected loss. Leveraging conformal risk control and non-parametric statistics, our framework intelligently selects the optimal model combination from a collection of black-box feature-extraction and inference models of varying complexities and computation times. We present both a fixed (relying on channel statistics) and a dynamic (channel-adaptive) model selection scheme. Numerical results validate the framework on a deadline-constrained image classification task while satisfying a maximum misclassification probability requirement. These results indicate that the proposed framework has the potential to provide reliable real-time edge AI services in 6G.

Figures

Figures reproduced from arXiv: 2506.11391 by the authors.

Figure 1
Figure 1. The considered edge inference scenario. The sensor encodes its input [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Model performance of the proposed schemes and baselines vs. SNR [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Distribution of the models selected by the proposed model selection [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

35 extracted references · 30 canonical work pages

  1. [1]

    Edge artificial intelligence for 6G: Vision, enabling technologies, and applications,

    K. B. Letaief, Y . Shi, J. Lu, and J. Lu, “Edge artificial intelligence for 6G: Vision, enabling technologies, and applications,” IEEE J. Sel. Areas Commun., vol. 40, no. 1, pp. 5–36, 2022

    work page 2022

  2. [2]

    Wireless 6G connectivity for massive number of devices and critical services,

    A. E. Kalør et al. , “Wireless 6G connectivity for massive number of devices and critical services,” Proc. IEEE, 2024, early access

    work page 2024

  3. [3]

    Mobile edge intelligence and computing for the Internet of vehicles,

    J. Zhang and K. B. Letaief, “Mobile edge intelligence and computing for the Internet of vehicles,” Proc. IEEE, vol. 108, no. 2, pp. 246–261, 2020

    work page 2020

  4. [4]

    Knowledge-based ultra-low-latency semantic communications for robotic edge intelligence,

    Q. Zeng, Z. Wang, Y . Zhou, H. Wu, L. Yang, and K. Huang, “Knowledge-based ultra-low-latency semantic communications for robotic edge intelligence,” IEEE Trans. Commun. , 2024, early access

    work page 2024

  5. [5]

    Service requirements for cyber-physical control applications in vertical domains,

    3GPP, “Service requirements for cyber-physical control applications in vertical domains,” 3rd Generation Partnership Project (3GPP), TS 22.104, June 2024, version 19.2.0

    work page 2024

  6. [6]

    EfficientNet: Rethinking model scaling for convolu- tional neural networks,

    M. Tan and Q. Le, “EfficientNet: Rethinking model scaling for convolu- tional neural networks,” in Proc. Int. Conf. Mach. Learn. (ICML) , 2019, pp. 6105–6114

    work page 2019

  7. [7]

    Conformal prediction: A gentle introduction,

    A. N. Angelopoulos and S. Bates, “Conformal prediction: A gentle introduction,” F ound. Trends Mach. Learn, vol. 16, no. 4, pp. 494–591, 2023

    work page 2023

  8. [8]

    Conformal risk control,

    A. N. Angelopoulos, S. Bates, A. Fisch, L. Lei, and T. Schuster, “Conformal risk control,” in Proc. Int. Conf. Learn. Represent. (ICLR) , 2024

    work page 2024

Show all 35 references
  1. [9]

    Prediction-powered inference,

    A. N. Angelopoulos, S. Bates, C. Fannjiang, M. I. Jordan, and T. Zrnic, “Prediction-powered inference,” Science, vol. 382, no. 6671, pp. 669– 674, 2023

    2023

  2. [10]

    Communication-computation trade-off in resource-constrained edge inference,

    J. Shao and J. Zhang, “Communication-computation trade-off in resource-constrained edge inference,” IEEE Commun. Mag. , vol. 58, no. 12, pp. 20–26, 2020

    2020

  3. [11]

    Dynamic compression ratio selection for edge inference systems with hard deadlines,

    X. Huang and S. Zhou, “Dynamic compression ratio selection for edge inference systems with hard deadlines,” IEEE Internet Things J. , vol. 7, no. 9, pp. 8800–8810, 2020

    2020

  4. [12]

    Edge AI: On-demand acceler- ating deep neural network inference via edge computing,

    E. Li, L. Zeng, Z. Zhou, and X. Chen, “Edge AI: On-demand acceler- ating deep neural network inference via edge computing,” IEEE Trans. Wireless Commun., vol. 19, no. 1, pp. 447–457, 2019

    2019

  5. [13]

    JALAD: Joint accuracy-and latency-aware deep structure decoupling for edge-cloud execution,

    H. Li, C. Hu, J. Jiang, Z. Wang, Y . Wen, and W. Zhu, “JALAD: Joint accuracy-and latency-aware deep structure decoupling for edge-cloud execution,” in Proc. IEEE Int. Conf. Parallel Dist. Syst. (ICPADS) , 2018, pp. 671–678

    2018

  6. [14]

    Joint device-edge inference over wireless links with pruning,

    M. Jankowski, D. G ¨und¨uz, and K. Mikolajczyk, “Joint device-edge inference over wireless links with pruning,” in Proc. IEEE Int. Workshp. Signal Process. Adv. Wireless Commun. (SPA WC) , 2020, pp. 1–5

    2020

  7. [15]

    Wireless image retrieval at the edge,

    ——, “Wireless image retrieval at the edge,” IEEE J. Sel. Areas Commun., vol. 39, no. 1, pp. 89–100, 2021

    2021

  8. [16]

    Ultra- low-latency edge inference for distributed sensing,

    Z. Wang, A. E. Kalør, Y . Zhou, P. Popovski, and K. Huang, “Ultra- low-latency edge inference for distributed sensing,” arXiv:2407.13360, 2024

    2024 arXiv

  9. [17]

    Ultra- low-latency edge intelligent sensing: A source-channel tradeoff and its application to coding rate adaptation,

    Q. Zeng, J. Huang, Z. Wang, K. Huang, and K. K. Leung, “Ultra- low-latency edge intelligent sensing: A source-channel tradeoff and its application to coding rate adaptation,” arXiv:2503.04645, 2025

    2025 arXiv

  10. [18]

    Resource allocation for multiuser edge inference with batching and early exiting,

    Z. Liu, Q. Lan, and K. Huang, “Resource allocation for multiuser edge inference with batching and early exiting,” IEEE J. Sel. Areas Commun. , vol. 41, no. 4, pp. 1186–1200, 2023

    2023

  11. [19]

    On-demand edge inference scheduling with accuracy and deadline guarantee,

    Y . She, M. Li, Y . Jin, M. Xu, J. Wang, and B. Liu, “On-demand edge inference scheduling with accuracy and deadline guarantee,” in Proc. IEEE/ACM Int. Symp. Qual. Service (IWQoS) , 2023, pp. 1–10

    2023

  12. [20]

    Adaptive early exiting for collaborative inference over noisy wireless channels,

    M. Jankowski, D. G ¨und¨uz, and K. Mikolajczyk, “Adaptive early exiting for collaborative inference over noisy wireless channels,” in Proc. IEEE Int. Conf. Mach. Learn. Commun. Netw. (ICMLCN) , 2024, pp. 126–131

    2024

  13. [21]

    Over-the- air multi-view pooling for distributed sensing,

    Z. Liu, Q. Lan, A. E. Kalør, P. Popovski, and K. Huang, “Over-the- air multi-view pooling for distributed sensing,” IEEE Trans. Wireless Commun., vol. 23, no. 7, pp. 7652–7667, 2024

    2024

  14. [22]

    Progressive feature transmission for split classification at the wireless edge,

    Q. Lan, Q. Zeng, P. Popovski, D. G ¨und¨uz, and K. Huang, “Progressive feature transmission for split classification at the wireless edge,” IEEE Trans. Wireless Commun. , vol. 22, no. 6, pp. 3837–3852, 2023

    2023

  15. [23]

    Beyond the cloud: Edge inference for generative large language models in wireless networks,

    X. Zhang et al., “Beyond the cloud: Edge inference for generative large language models in wireless networks,” IEEE Trans. Wireless Commun., vol. 24, no. 1, pp. 643–658, 2025

    2025

  16. [24]

    Guaranteed dynamic scheduling of ultra-reliable low-latency traffic via conformal prediction,

    K. M. Cohen, S. Park, O. Simeone, P. Popovski, and S. Shamai, “Guaranteed dynamic scheduling of ultra-reliable low-latency traffic via conformal prediction,” IEEE Signal Process. Lett. , vol. 30, pp. 473–477, 2023

    2023

  17. [25]

    Calibrating AI models for wireless communications via conformal prediction,

    K. M. Cohen, S. Park, O. Simeone, and S. Shamai Shitz, “Calibrating AI models for wireless communications via conformal prediction,” IEEE Trans. Mach. Learn. Commun. Netw. , vol. 1, pp. 296–312, 2023

    2023

  18. [26]

    Fed- erated inference with reliable uncertainty quantification over wireless channels via conformal prediction,

    M. Zhu, M. Zecchin, S. Park, C. Guo, C. Feng, and O. Simeone, “Fed- erated inference with reliable uncertainty quantification over wireless channels via conformal prediction,” IEEE Trans. Signal Process., vol. 72, pp. 1235–1250, 2024

    2024

  19. [27]

    Branchynet: Fast inference via early exiting from deep neural networks,

    S. Teerapittayanon, B. McDanel, and H. Kung, “Branchynet: Fast inference via early exiting from deep neural networks,” in proc. Int. Conf. Pattern Recognit. (ICPR) , 2016, pp. 2464–2469

    2016

  20. [28]

    YOLOv10: Real-time end-to-end object detection,

    A. Wang et al. , “YOLOv10: Real-time end-to-end object detection,” arXiv:2405.14458, 2024

    2024 arXiv

  21. [29]

    Learn then test: Calibrating predictive algorithms to achieve risk control,

    A. N. Angelopoulos, S. Bates, E. J. Cand `es, M. I. Jordan, and L. Lei, “Learn then test: Calibrating predictive algorithms to achieve risk control,” arXiv:2110.01052, 2021

    2021 arXiv

  22. [30]

    Adaptive learn-then-test: Statis- tically valid and efficient hyperparameter selection,

    M. Zecchin, S. Park, and O. Simeone, “Adaptive learn-then-test: Statis- tically valid and efficient hyperparameter selection,” arXiv:2409.15844, 2025

    2025 arXiv

  23. [31]

    ImageNet large scale visual recognition chal- lenge,

    O. Russakovsky et al. , “ImageNet large scale visual recognition chal- lenge,” Int. J. Comput. Vision (IJCV) , vol. 115, no. 3, pp. 211–252, 2015

    2015

  24. [32]

    WebP image format,

    J. Zern, P. Massimino, and J. Alakuijala, “WebP image format,” Nov

  25. [33]

    EfficientNetV2: Smaller models and faster training,

    M. Tan and Q. Le, “EfficientNetV2: Smaller models and faster training,” in Proc. Int. Conf. Mach. Learn. (ICML) , 2021, pp. 10 096–10 106

    2021

  26. [34]

    PyTorch 2: Faster machine learning through dynamic Python bytecode transformation and graph compilation,

    J. Ansel et al. , “PyTorch 2: Faster machine learning through dynamic Python bytecode transformation and graph compilation,” in Proc. ACM Int. Conf. Architectural Support Program. Lang. Operating Syst., V ol. 2 , 2024, p. 929–947

    2024

  27. [2024]

    Available: https://www.rfc-editor.org/info/rfc9649

    [Online]. Available: https://www.rfc-editor.org/info/rfc9649

Pith tools

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