Pith. sign in

REVIEW 3 major objections 5 minor 35 references

Distributed Integrated Sensing and Edge AI Exploiting Prior Information

T0 review · 3 major / 5 minor · reviewed 2026-08-03 · deepseek-v4-flash

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

Pith's one-line read Prior-aware sensing and power allocation can improve distributed edge-AI inference at low SNR.

desk verdict Solid, useful engineering paper with a real gap in its headline optimality proof: Theorem 4 silently assumes equal per-device variances, and the stress-test counterexample is correct. read the letter →

arxiv 2512.00309 v5 pith:HT3LOKWB submitted 2025-11-29 eess.SP

Biao Dong , Bin Cao , Guan Gui , Qinyu Zhang This is my paper · ORCID
classification eess.SP
keywords integratedsensingandedgeAIGaussianmixturepriorresponsibility-weightedBayesianestimationover-the-aircomputationdecision-optimalpowerallocationMahalanobisdistanceTDMFDMschemeslow-SNRinference
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 claims that a distributed integrated sensing and edge-AI system can get better inference accuracy if prior information is exploited at both levels of the pipeline: when devices estimate features from noisy radar measurements, and when the access point aggregates those estimates over the air. At the sensing level, the authors propose a responsibility-weighted Bayesian estimator that shrinks noisy features toward class means, which they show outperforms maximum-likelihood estimation at low signal-to-noise ratios. At the communication level, they introduce two performance proxies and derive closed-form power-allocation rules for time- and frequency-division multiplexing, showing that allocating power according to how discriminative each feature is yields an additional accuracy gain. The paper's message is that classical ML-based feature estimation and communication-agnostic power control leave accuracy on the table when the task has known class structure.

What carries the argument

The core object is the responsibility-weighted Bayesian (RWB) estimator, which uses the posterior probability that an observation belongs to each class, θℓ, to weight class-conditional MMSE estimates. This shrinks estimates toward class means depending on the sensing noise level. At the communication level, the key machinery is the minimum inter-class Mahalanobis distance Gmin, used as a discriminative prior in the AirComp power-allocation problem; solving that problem yields a threshold-based solution in TDM and a dual-decomposition solution in FDM.

What would settle it

Compare the proposed threshold power allocation c_k* = min{u_k, τ*} against a numerically optimized solution for a scenario where each device has a different estimated-feature variance, especially with a wide variance spread; if the threshold solution is consistently beaten, the central optimality claim fails in the general case.

Watch

Extended reading notes

Core claim

The central claim is that task-relevant priors should be used twice: once in feature estimation and once in over-the-air aggregation. The proposed RWB estimator computes class responsibilities from a Gaussian-mixture prior and weights class-conditional posterior means by them, effectively denoising features at low SNR while converging to ML at high SNR. On the communication side, the paper argues that the right optimization objective is not minimum aggregation error alone but the separability of received features, and proves that the decision-optimal power allocation has a threshold structure in TDM and a dual-decomposition structure in FDM. Numerical results on a human-posture classificatio

Load-bearing premise

The decision-optimal TDM proof assumes the estimated-feature variance is effectively the same across devices, even though the system model defines it per device; if those variances differ materially, the threshold solution is not proven optimal.

Editorial extensions

If this is right

  • If the RWB estimator works as claimed, devices can produce more reliable features without changing the sensing hardware, simply by using offline class statistics.
  • If the decision-optimal power allocation is correct, communication resources can be skewed toward discriminative features, yielding accuracy gains over MSE-optimal and even channel-inversion schemes.
  • The asymptotic connection between classification error and minimum inter-class distance implies that aggregating more devices or subcarriers shrinks class uncertainty and pushes accuracy toward an upper bound.
  • The threshold structure suggests a simple on-device policy: transmit at full power on discriminative dimensions and cap power on weak links, which is practical for real-time deployment.

Reading between the lines

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

  • The threshold form of the power allocation is likely generalizable to other federated or distributed inference settings where features have known class-conditional statistics, beyond FMCW radar.
  • The RWB estimator's shrinkage behavior hints that a soft-thresholding variant could be derived for non-Gaussian or sparse priors, which may improve robustness at very low SNR.
  • The paper's proxies could be combined: a weighted objective that balances computation MSE and inter-class MD might reveal a trade-off frontier that neither proxy alone captures.
  • A testable extension is to let the threshold τ* adapt per device based on local sensing noise, which may recover the optimality lost when per-device variances differ.
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 / 5 minor

Summary. This paper studies a distributed integrated sensing and edge AI (ISEA) system in which K devices sense a common target, extract features, and transmit them to an access point via over-the-air computation for collaborative classification. The authors propose a responsibility-weighted Bayesian (RWB) estimator that uses a Gaussian mixture prior, and they demonstrate experimentally that it outperforms maximum-likelihood estimation at low SNR. At the communication level, they introduce two proxies—computation-optimal (MSE-based) and decision-optimal (Mahalanobis-distance-based)—and derive power-allocation solutions for TDM and FDM transmission. The TDM decision-optimal solution is claimed to have a threshold-based closed form; the FDM solution is obtained by dual decomposition. Numerical experiments on FMCW human-pose data with MLP and SVM classifiers illustrate gains for both the Bayesian estimator and the discriminant-aware transceiver design.

Significance. If the claims are correct, the paper offers a useful Bayesian perspective for ISEA by exploiting priors at both the sensing and communication levels. The RWB estimator is a reasonable moment-matched GM posterior mean, and the conditional-entropy comparisons in Section III-D are analytically sound. The reproducible code and data link, the use of a real FMCW spectrogram dataset, and the separate training of classifiers give the experimental section credibility. The main concern is a gap in the proof of Theorem 4, which is load-bearing for the claimed threshold-based TDM power allocation. Because that result is central to the abstract and to the paper's communication-level contribution, the manuscript requires major revision.

major comments (3)
  1. [IV-B2 (Theorem 4)] The proof of Theorem 4 silently replaces the device-dependent estimated-feature variances σ̂²_{k,m} (defined in Section II-A3 and used in Eq. (39)) with a single σ̂²_m. The reduction to (P4) therefore has denominator Σ c_k² + σ_eq², whereas the correct denominator from (39) is Σ a_k c_k² + σ_w² with a_k = σ̂²_{k,m}. With heterogeneous a_k, the claimed solution c_k* = min{u_k, τ} is not optimal. For a concrete counterexample with K=2, a=(1,10), u=(1,1), σ_w=0: c=(1,1) gives F=4/11≈0.364, while c=(1,0.1) gives F=1.21/1.1=1.1. Thus Theorem 4, and the abstract's 'threshold-based' TDM closed-form allocation, are unproven as stated. Please either prove the result under the true per-device variances or explicitly state and justify an additional assumption (e.g., identical σ̂²_{k,m}).
  2. [IV-C, Eqs. (47)-(48)] The derivation of the decision-optimal FDM solution is too compressed to verify optimality. Problem (P7) is a sum of M fractional functions, and the text says 'Following the same procedure as in (44) and (45)' without showing the KKT stationarity conditions or the fixed-point equation for the auxiliary variable z_n. Please provide the complete derivation, including why the per-subcarrier dual decomposition is valid for this non-concave ratio-sum objective and what time-sharing conditions are required.
  3. [III-D, Eq. (32)] The text says 'if each device adopts the MMSE estimator in (24)', but Eq. (24) is the RWB estimator, whose shrinkage coefficient depends on the responsibilities θ_ℓ. The conditional-entropy formula (32) uses the fixed coefficient ρ_k = σ²/(σ²+σ_k²), which corresponds to the class-known MMSE estimator (17), not to the RWB estimator. Thus the theoretical entropy comparison does not directly support the RWB estimator as claimed. Please correct the attribution or extend the analysis to the RWB posterior in (22).
minor comments (5)
  1. [Abstract / Conclusion] The phrase 'closed-form power allocation' is accurate for TDM, but the FDM solution is an iterative dual-decomposition procedure. Please use a more nuanced expression, e.g., 'closed-form per-subcarrier structure with dual-decomposition', to avoid overstating the result.
  2. [Theorem 2] The exponent κ K G_min is stated without specifying the asymptotic regime: is K the number of devices, or the number of i.i.d. observations? The connection between the exponential-order formula and the aggregated MD in (39) should be clarified.
  3. [IV-B2 / System model] Problem (P3) uses a per-slot power constraint |b_k|²ν_k² ≤ P_k, while the system model in (7) states a total power constraint over M subcarriers for FDM. The analogous TDM power budget should be defined explicitly to avoid ambiguity about per-slot versus total power.
  4. [Theorem 4 proof] The variables Δ' and σ_eq² are introduced without definitions; Δ_m is not squared in (39), so the notation Δ' = |Δ_m|²/σ̂²_m needs to be reconciled with the original definition of Δ_m. Please align the notation.
  5. [Miscellaneous typos] In (42), 'Pk' should be 'P_k'; in (46), 'Pc' should be 'P_k'; and 'Base on (45)' should be 'Based on (45)'. Please proofread the equations and surrounding text.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the sensing and communication derivations are self-contained and the numerical benchmarks use separately trained classifiers; the only flagged weakness (Theorem 4's per-device variance simplification) is a proof gap, not a circular dependency.

full rationale

The RWB estimator is the standard posterior mean under the GM prior (Eqs. 20-24): it is defined directly from Bayes' rule, and its MSE comparison with ML (Eqs. 12, 18, 25) is a derived inequality, not a restatement of an assumption. The communication proxies are not fitted predictions: the computation-optimal proxy is a Markov-based lower bound (Theorem 1) and the decision-optimal proxy follows from the Chernoff/Mahalanobis bound (Theorem 2). Optimal power allocations for TDM/FDM are obtained by KKT and Lagrange duality with the stated objectives. The paper's self-citations ([1], [4], [10]) appear only in related-work lists and a submission note; they do not carry the derivations. One non-circular correctness concern is flagged: in the proof of Theorem 4 (Section IV-B2), the equivalent reformulation (P4) replaces the device-dependent variances \hat\sigma^2_{k,m} of Eq. (39) and Section II-A3 by a single \hat\sigma^2_m, so the threshold solution is not guaranteed optimal under heterogeneous variances. This is a mathematical gap rather than an input-output equivalence, and it does not by itself make the derivation circular.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

The central claims rest on a Gaussian-mixture feature model, an asymptotic error-exponent link between Mahalanobis distance and classification error, and standard AirComp assumptions (synchronized, perfectly-compensated channels). Several parameter sets are estimated offline from training data. The optimality proofs also rely on a time-sharing/zero-duality-gap argument and, in the TDM decision-optimal case, on an equal-variance assumption not stated in the model.

free parameters (4)
  • GM prior statistics {π_l, μ_l, Σ} = estimated from a 12,000-sample training set
    The RWB estimator and the decision-optimal Δ_m both depend on these offline estimates; their accuracy directly controls the claimed low-SNR gain.
  • Sensing noise variance σ_k^2 = treated as a tunable hyper-parameter (Remark 2)
    Controls the shrinkage strength in RWB and appears in every MSE/MD expression; the paper tunes it rather than deriving it from a physical model.
  • Discriminative prior Δ_m = computed from training class means
    Defines the decision-optimal objective in (39) and (P7); it is estimated offline from the same data used to train the classifiers.
  • Feature power ν^2_{k,n} and estimated feature variance \hatσ^2_{k,n} = estimated from offline training data [7], [8]
    These enter the power constraints and the MSE/MD objectives; their estimates are taken from prior work practice.
assumptions (6)
  • domain assumption Ground-truth features follow a Gaussian mixture with shared diagonal covariance: p(x)=Σπ_l N(x|μ_l,Σ).
    Invoked in Section II-A3 and used by both the RWB estimator and the MD proxy; real PCA features need not be Gaussian.
  • domain assumption PCA makes feature dimensions statistically independent with diagonal covariance.
    Used in Definition 2 to decompose Mahalanobis distance into per-dimension terms and to justify per-subcarrier optimization.
  • domain assumption Sensing observations satisfy r_k = g + z_k with AWGN, and the feature x is a random vector with known prior.
    This is the Bayesian sensing model in Section II-A; if the noise or prior is misspecified, RWB's MSE guarantee is not exact.
  • domain assumption Asymptotic error-exponent relation P_e \doteq exp(-κ K G_min) holds for finite K via Fano/Chernoff bounds.
    Used in Theorem 2 to justify the decision-optimal proxy; simulations use K=3, which is far from the asymptotic regime.
  • standard math The nonconvex FDM problems (P6) and (P7) have zero duality gap by the time-sharing condition.
    Invoked in Section IV-C following [33]; standard but a condition the paper does not verify numerically for its channel realizations.
  • domain assumption AirComp assumptions: perfect synchronization, perfect CSI via TDD reciprocity, slow fading, and linear receiver combining.
    Stated in Section II-B1; any violation changes the MSE and MD expressions and the derived power allocations.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Distributed Integrated Sensing and Edge AI Exploiting Prior Information." pith.science (2026). https://pith.science/paper/HT3LOKWB

@misc{pith2026251200309,
  author       = {Pith},
  title        = {Pith review of: Distributed Integrated Sensing and Edge AI Exploiting Prior Information},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HT3LOKWB}},
  note         = {Machine review of arXiv:2512.00309}
}
read the original abstract

This paper investigates a distributed ISEA system under a Bayesian framework, focusing on incorporating task-relevant priors to maximize inference performance. At the sensing level, an RWB estimator with a GM prior is designed. By weighting class-conditional posterior means with responsibilities, RWB effectively denoises features and outperforms ML at low SNR. At the communication level, two theoretical proxies are introduced: the computation-optimal and decision-optimal proxies. Optimal transceiver designs in terms of closed-form power allocation are derived for both TDM and FDM settings, revealing threshold-based and dual-decomposition structures. Results show that the discriminant-aware allocation yields additional inference gains.

Figures

Figures reproduced from arXiv: 2512.00309 by the authors.

Figure 1
Figure 1. The considered ISEA system consists of a single common [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Two graphical models, where white nodes denote latent [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. MSE versus sensing SNR under ML estimation and the [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: MSE and MD versus the communication SNR under MSE [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: MSE and MD versus communication SNR under MSE-optimal [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 7
Figure 7. Figure 7: Classification performance comparison between ML and RWB [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]
Figure 6
Figure 6. Figure 6: Two wireless sensing samples of human motion: (a) standing [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 8
Figure 8. Figure 8: Classification performance comparison under varying communication SNR: (a) MLP and (b) SVM. [PITH_FULL_IMAGE:figures/full_fig_p011_8.png]
Figure 9
Figure 9. Figure 9: MLP classification performance comparison under varying number of users [PITH_FULL_IMAGE:figures/full_fig_p011_9.png]
Figure 10
Figure 10. Figure 10: The feature visualization under different communication schemes: (a) raw features, (b) decision-optimal, (c) computation-optimal, [PITH_FULL_IMAGE:figures/full_fig_p012_10.png]
Figure 11
Figure 11. Figure 11: MLP confusion matrices with SNRc = 10 dB under differ￾ent communication schemes: (a) decision-optimal, (b) computation￾optimal, (c) equal allocation, and (d) channel inversion. In summary, the communication-level results demon￾strate that the decision-optimal scheme e…

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

35 extracted references · 1 linked inside Pith

  1. [1]

    Optimized power control for multi-user integrated sensing and edge AI,

    B. Dong and B. Cao, “Optimized power control for multi-user integrated sensing and edge AI,” inProc. IEEE Int. Conf. Commun. (ICC), Glasgow, U.K., May 2026, submitted

    2026

  2. [2]

    Integrated sensing and communications: Toward dual-functional wire- less networks for 6G and beyond,

    F. Liu, Y . Cui, C. Masouros, J. Xu, T. X. Han, Y . C. Eldar, and S. Buzzi, “Integrated sensing and communications: Toward dual-functional wire- less networks for 6G and beyond,”IEEE J. Sel. Areas Commun., vol. 40, no. 6, pp. 1728–1767, 2022

    2022

  3. [3]

    Flocoff: Data heterogeneity resilient federated learning with communication-efficient edge offloading,

    M. Ma, C. Gong, L. Zeng, Y . Yang, and L. Wu, “Flocoff: Data heterogeneity resilient federated learning with communication-efficient edge offloading,”IEEE J. Sel. Areas Commun., 2024

    2024

  4. [4]

    Robust deep joint source- channel coding enabled distributed image transmission with imperfect channel state information,

    B. Dong, B. Cao, G. Gui, and Q. Zhang, “Robust deep joint source- channel coding enabled distributed image transmission with imperfect channel state information,”IEEE Trans. Wireless Commun., early access, Sep. 2025. 13

    2025

  5. [5]

    Task- oriented sensing, computation, and communication integration for multi- device edge AI,

    D. Wen, P. Liu, G. Zhu, Y . Shi, J. Xu, Y . C. Eldar, and S. Cui, “Task- oriented sensing, computation, and communication integration for multi- device edge AI,”IEEE Trans. Wireless Commun., vol. 23, no. 3, pp. 2486–2502, 2023

    2023

  6. [6]

    Integrated sensing and edge AI: Realizing intelligent perception in 6G,

    Z. Liu, X. Chen, H. Wu, Z. Wang, X. Chen, D. Niyato, and K. Huang, “Integrated sensing and edge AI: Realizing intelligent perception in 6G,” IEEE Comm. Surv. Tutor., early access, May 2025

    2025

  7. [7]

    Task-oriented over-the-air computation for multi-device edge AI,

    D. Wen, X. Jiao, P. Liu, G. Zhu, Y . Shi, and K. Huang, “Task-oriented over-the-air computation for multi-device edge AI,”IEEE Trans. Wire- less Commun., vol. 23, no. 3, pp. 2039–2053, 2023

  8. [8]

    On the view-and-channel aggregation gain in integrated sensing and edge AI,

    X. Chen, K. B. Letaief, and K. Huang, “On the view-and-channel aggregation gain in integrated sensing and edge AI,”IEEE J. Sel. Areas Commun., vol. 42, no. 9, pp. 2292–2305, 2024

    2024

Show all 35 references
  1. [9]

    Exploiting semantic communication for non- orthogonal multiple access,

    X. Mu and Y . Liu, “Exploiting semantic communication for non- orthogonal multiple access,”IEEE J. Sel. Areas Commun., vol. 41, no. 8, pp. 2563–2576, 2023

    2023

  2. [10]

    Inference-optimal ISAC via task- oriented feature transmission and power allocation,

    B. Dong, B. Cao, and Q. Zhang, “Inference-optimal ISAC via task- oriented feature transmission and power allocation,”arXiv preprint arXiv:2510.20429, 2025

    2025 arXiv

  3. [11]

    A survey on over-the-air computation,

    A. S ¸ahin and R. Yang, “A survey on over-the-air computation,”IEEE Comm. Surv. Tutor., vol. 25, no. 3, pp. 1877–1908, 2023

    1908

  4. [12]

    Optimized power control for over-the-air computation in fading channels,

    X. Cao, G. Zhu, J. Xu, and K. Huang, “Optimized power control for over-the-air computation in fading channels,”IEEE Trans. Wireless Commun., vol. 19, no. 11, pp. 7498–7513, 2020

    2020

  5. [13]

    Over-the-air computation systems: Optimization, analysis and scaling laws,

    W. Liu, X. Zang, Y . Li, and B. Vucetic, “Over-the-air computation systems: Optimization, analysis and scaling laws,”IEEE Trans. Wireless Commun., vol. 19, no. 8, pp. 5488–5502, 2020

    2020

  6. [14]

    S. M. Kay,Fundamentals of statistical signal processing: Estimation theory. Prentice-Hall, Inc., 1993

    1993

  7. [15]

    End-to-end learning for task-oriented semantic communications over MIMO channels: An information-theoretic framework,

    C. Cai, X. Yuan, and Y .-J. A. Zhang, “End-to-end learning for task-oriented semantic communications over MIMO channels: An information-theoretic framework,”IEEE J. Sel. Areas Commun., vol. 43, no. 4, pp. 1292–1307, 2025

    2025

  8. [16]

    Information-theoretic asymptotics of Bayes methods,

    B. S. Clarke and A. R. Barron, “Information-theoretic asymptotics of Bayes methods,”IEEE Trans. Inf. Theory, vol. 36, no. 3, pp. 453–471, 2002

    2002

  9. [17]

    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, 2023

    2023

  10. [18]

    Energy-efficient edge inference in integrated sensing, communication, and computation networks,

    J. Yao, W. Xu, G. Zhu, K. Huang, and S. Cui, “Energy-efficient edge inference in integrated sensing, communication, and computation networks,”IEEE J. Sel. Areas Commun., early access, May 2025

    2025

  11. [19]

    C. M. Bishop and N. M. Nasrabadi,Pattern recognition and machine learning. New York, USA: Springer Science & Business Media, 2006

    2006

  12. [20]

    Lightweight semantic-aware commu- nication with packet transmission,

    B. Dong, B. Cao, and Q. Zhang, “Lightweight semantic-aware commu- nication with packet transmission,”IEEE Commun. Lett., vol. 29, no. 7, pp. 1569–1573, 2025

    2025

  13. [21]

    Wasserstein-distance-based Gaussian mixture reduction,

    A. Assa and K. N. Plataniotis, “Wasserstein-distance-based Gaussian mixture reduction,”IEEE Signal Process. Lett., vol. 25, no. 10, pp. 1465– 1469, 2018

    2018

  14. [22]

    Tse and P

    D. Tse and P. Viswanath,Fundamentals of wireless communication. Cambridge, U.K.: Cambridge Univ. Press, 2005

    2005

  15. [23]

    Interference channels,

    A. Carleial, “Interference channels,”IEEE Trans. Inf. Theory, vol. 24, no. 1, pp. 60–70, 2003

    2003

  16. [24]

    The expectation-maximization algorithm,

    T. K. Moon, “The expectation-maximization algorithm,”IEEE Signal Process. Mag., vol. 13, no. 6, pp. 47–60, 1996

    1996

  17. [25]

    T. M. Cover,Elements of information theory. John Wiley & Sons, 1999

    1999

  18. [26]

    G. H. Hardy, J. E. Littlewood, and G. P ´olya,Inequalities, 2nd ed. Cambridge, U.K.: Cambridge Univ. Press, 1952

    1952

  19. [27]

    Robust large margin deep neural networks,

    J. Sokoli ´c, R. Giryes, G. Sapiro, and M. R. Rodrigues, “Robust large margin deep neural networks,”IEEE Trans. Signal Process., vol. 65, no. 16, pp. 4265–4280, 2017

    2017

  20. [28]

    The asymptotics of posterior entropy and error probability for Bayesian estimation,

    F. Kanayaet al., “The asymptotics of posterior entropy and error probability for Bayesian estimation,”IEEE Trans. Inf. Theory, vol. 41, no. 6, pp. 1988–1992, 1995

    1988

  21. [29]

    Fukunaga,Introduction to statistical pattern recognition

    K. Fukunaga,Introduction to statistical pattern recognition. Amster- dam, The Netherlands: Elsevier, 2013

    2013

  22. [30]

    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, 2022

    2022

  23. [31]

    Kreyszig,Introductory functional analysis with applications

    E. Kreyszig,Introductory functional analysis with applications. John Wiley & Sons, 1991

    1991

  24. [32]

    S. P. Boyd and L. Vandenberghe,Convex optimization. Cambridge, U.K.: Cambridge Univ. Press, 2004

    2004

  25. [33]

    Dual methods for nonconvex spectrum optimization of multicarrier systems,

    W. Yu and R. Lui, “Dual methods for nonconvex spectrum optimization of multicarrier systems,”IEEE Trans. Wireless Commun., vol. 54, no. 7, pp. 1310–1322, 2006

    2006

  26. [34]

    Wireless sens- ing with deep spectrogram network and primitive based autoregressive hybrid channel model,

    G. Li, S. Wang, J. Li, R. Wang, X. Peng, and T. X. Han, “Wireless sens- ing with deep spectrogram network and primitive based autoregressive hybrid channel model,” inProc. IEEE Int. Workshop Signal Process. Adv. Wireless Commun. (SPAWC), Sep. 2021, pp. 481–485

    2021

  27. [35]

    Visualizing data using t-SNE,

    L. Van der Maaten and G. Hinton, “Visualizing data using t-SNE,”J. Mach. Learn. Res., vol. 9, no. Nov, pp. 2579–2605, 2008

    2008

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.