Pith. sign in

REVIEW 3 major objections 5 minor 36 references

Computation-resource-efficient Task-oriented Communications

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Task-oriented communication can be made computation-efficient by a multi-exit encoder that spends more compute only on hard images, while a static variant already transmits 16 features and beats stronger baselines on accuracy, transmitted…

desk verdict Static model is solid; dynamic budget claim needs a fix—Algorithm 2 calibrates thresholds marginally but the greedy exit rule changes realized FLOPs. read the letter →

arxiv 2507.07422 v1 pith:NLRDWX3L submitted 2025-07-10 eess.IV

Jingwen Fu , Ming Xiao , Chao Ren , Mikael Skoglund This is my paper · ORCID
classification eess.IV
keywords task-orientedcommunicationssemanticdynamicneuralnetworksmulti-exitcomputationbudgetimageclassificationwirelesstransmissionFLOPs
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

Task-oriented communication (TOC) normally spends heavy compute to extract and transmit only task-relevant features, which is hard for phones, UAVs, and sensors. This paper proposes two TOC modes: a static encoder for devices with no computation cap, and a dynamic multi-exit encoder for devices with a computation budget. The static mode compresses transmitted features to 16 dimensions and cuts FLOPs while improving classification accuracy relative to DeepJSCC, VFE, and GatedJSCC on image tasks. The dynamic mode estimates each image's difficulty from a confidence score and exits early for easy images, so average compute stays within a budget while accuracy rises as the budget grows. The paper also proves an $O(1/\sqrt{T})$ convergence rate for SGD training of the end-to-end system.

What carries the argument

The load-bearing mechanism is a multi-exit dynamic neural network used as the task-oriented encoder. Each input image gets a confidence score at each exit, defined as the softmax value of the predicted-class logit; easy images pass a validation-calibrated threshold at an early exit, while hard images proceed to deeper exits. The network is structured with horizontal depth and vertical feature scales, dense connectivity to stabilize propagation, and exits assumed evenly spaced with $\varphi$ FLOPs between consecutive exits. A simplified channel encoder/decoder with linear layers compresses features to 16 dimensions before wireless transmission, and exit thresholds are chosen on a validation set so that expected FLOPs $\sum_k \Pr_k C_k$ respect the device budget $B$.

What would settle it

Run the trained dynamic model on a test set deliberately shifted from validation, e.g., blurred images or a different class distribution, while keeping the validation-calibrated thresholds fixed, and measure the realized average FLOPs per batch; if it exceeds the budget $B$, the budget-guarantee mechanism fails. Also count FLOPs between consecutive exits directly; if they are not roughly equal, the $[\varphi, K\varphi]$ range used in Proposition 2 is not accurate.

Watch

Extended reading notes

Core claim

The paper's central claim is that computation efficiency and task accuracy do not have to be traded off in TOC. With ResNet-20 as the static task-oriented encoder on CIFAR-100, the proposed system transmits 16-dimensional features, uses about 40.81M FLOPs, and reaches 55.58% accuracy at PSNR 0 dB in AWGN, against DeepJSCC's 2048 dimensions, 50.25M FLOPs, and 39.22% accuracy; VFE uses more FLOPs and GatedJSCC uses 256 dimensions. The dynamic model operates under a device budget $B$ and, at $B \ge 60$M FLOPs, reaches 62.26% accuracy, exceeding both static baselines, because simple images exit early and complex images go deeper. Accuracy generally grows with budget up to a point, after which deeper exits can hurt.

Load-bearing premise

The load-bearing premise, stated in Section V, is that training, validation, and test data share a similar distribution, so confidence thresholds calibrated on the validation set keep the test-time average computation spend within the device budget.

Editorial extensions

If this is right

  • If the static claim is right, task-oriented systems can transmit 16 features instead of 2048 with higher accuracy and fewer FLOPs on small-image classification.
  • The dynamic model gives a single trained network that can serve devices with different computation budgets by only changing exit thresholds.
  • Accuracy-budget curves improve with PSNR, meaning the same computation budget buys higher accuracy when the wireless channel is cleaner.
  • The convergence proof implies that end-to-end training of the encoder, channel codec, and inference block stabilizes after enough SGD iterations.

Reading between the lines

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

  • Editorial inference: the paper's Section V assumption that training, validation, and test sets share a similar distribution is what makes Algorithm 2's validation-calibrated thresholds enforce the budget at test time; under distribution shift the realized FLOPs could exceed $B$, so deployed systems should monitor exit-rate statistics.
  • Editorial inference: the same multi-exit encoder could double as a channel-adaptive transmission mechanism, e.g., skipping late exits under poor SNR, but the paper does not test this.
  • Editorial inference: the even-exit-spacing assumption could be relaxed by measuring per-exit FLOPs directly, and a version with uneven exits would need a modified budget formula.
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 proposes computation-resource-efficient task-oriented communication (TOC) systems with static and dynamic operation modes. The static mode uses a ResNet as the task-oriented encoder and a lightweight channel encoder/decoder to reduce transmitted feature dimensions (e.g., 16 dimensions on CIFAR) and FLOPs, while the dynamic mode uses a multi-exit dynamic neural network with confidence-score thresholds calibrated on a validation set to allocate computation within a device budget B. The authors provide a convergence analysis for SGD training and evaluate on CIFAR-10/100 and ImageNet, reporting accuracy gains over DeepJSCC, VFE, and GatedJSCC baselines. The central claims are that the static model reduces transmitted dimensions, FLOPs, and accuracy loss simultaneously, and that the dynamic model meets device computation budgets while further improving accuracy as the budget grows.

Significance. The paper addresses a timely and practically important problem: making task-oriented communication feasible on resource-constrained devices. The static model's reductions in transmitted feature dimension (to 16 on CIFAR) and FLOPs are clearly demonstrated and are valuable contributions. The dynamic model's idea of early exiting based on task complexity is natural for TOC and could be impactful if the budget mechanism is sound. The convergence analysis is standard nonconvex SGD theory and is correctly applied, though it does not use TOC-specific structure. The main weaknesses are a gap between the threshold calibration and the realized exit distribution in the dynamic model, an inconsistency in the expected-FLOPs formula and its proof, and an overstated claim of simultaneous accuracy improvement. These issues are fixable, but they affect the load-bearing budget-feasibility claim and the accuracy comparison.

major comments (3)
  1. [Section V-C-2, Algorithms 2 and 3, Eq. (17)-(18), Proposition 2] The threshold calibration in Algorithm 2 sets each θk to match the marginal per-exit probability Prk, but Algorithm 3 uses a sequential greedy rule: a sample exits at k only if it fails all earlier thresholds and then passes θk. The realized exit probability at exit k is therefore the joint probability P(φ1 < θ1, ..., φk−1 < θk−1, φk ≥ θk), which equals the marginal Prk only under strong assumptions about correlation among confidence scores across exits. Consequently, the expected-FLOPs formula in Eq. (18) and the budget bound in Proposition 2 hold for the nominal distribution, not for the distribution actually realized by Algorithm 3, and the realized computation cost can exceed B even when the nominal bound is satisfied. Since Figs. 6–9 plot the nominal budget B on the x-axis and the paper reports no measured FLOPs for the dynamic model on the test set, the central claim that the dynamic model meets a device computation budget is not verified. Please either revise the threshold calibration to be sequential (e.g., set θk on the subset of validation samples that failed all earlier exits) or report measured average FLOPs on the test set alongside the nominal budget curves.
  2. [Eq. (18) and Appendix B] The denominator in Eq. (18) is printed as ∑_{j=1}^k r^j, but the derivation in Appendix B (Eq. (39) and the following simplification) replaces it with ∑_{j=1}^K r^j, which equals r(1−r^K)/(1−r) and is independent of k. This substitution is unjustified as written: for k < K, ∑_{j=1}^k r^j ≠ r(1−r^K)/(1−r). The final closed form and Lemmas 1–2 correspond to the denominator ∑_{j=1}^K r^j, which is consistent with the definition of Prk in Eq. (17). The printed Eq. (18) should be corrected to the total sum, and the proof in Appendix B should be rewritten to avoid the invalid substitution.
  3. [Section VII-B, Table III] The abstract and Section VII.B claim that the proposed static model outperforms baseline models in transmitted dimensions, FLOPs, and accuracy simultaneously, but Table III shows counterexamples. In Rayleigh fading at PSNR = 0 dB, the proposed static ResNet-20 achieves 34.98% accuracy and ResNet-32 achieves 34.57%, both below DeepJSCC (36.20%) and well below GatedJSCC (53.28%). The claim should be qualified to the operating points where the accuracy advantage actually holds (e.g., higher PSNR levels or AWGN channels), or the comparison should be aggregated in a way that acknowledges these exceptions.
minor comments (5)
  1. [Algorithm 2] The pseudocode in Algorithm 2 places the 'Sort ϕj,k in descending order' step inside the per-sample loop, but threshold setting requires sorting the confidence scores for each exit across all validation samples. Please clarify the loop structure so that the sorting and threshold selection are performed per exit over the entire validation set.
  2. [Section VII-C-2, Table IV] The text accompanying Table IV states that 'neither the baseline models nor the proposed static TOC model show an increase in accuracy' as the budget grows, but the table shows the static model accuracy increasing from 55.58% (ResNet-20 at 50M FLOPs) to 64.94% (ResNet-32 at 70M FLOPs) when a larger network becomes affordable. The intended point is that accuracy is constant for a fixed architecture once its FLOPs requirement is met; please reword to avoid the contradiction.
  3. [Section VII-A-3] All results are reported as averages over 5 trials without standard deviations or confidence intervals. Given the non-monotonic accuracy across ResNet depths, including error bars or variance information would strengthen the comparisons.
  4. [Section VI] The convergence analysis is a standard nonconvex SGD bound that does not use any TOC-specific structure. The authors should state explicitly that the channel and task modules are subsumed in the objective function, so the analysis applies to the whole end-to-end system, not just the encoder.
  5. [Eq. (29)] The PSNR definition PSNR = 10 log10(P/σ²) uses signal power P and noise power σ²; please clarify whether P denotes peak signal power or average power, and specify the normalization used in the simulations.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: accuracy and FLOPs results are measured out-of-sample against external baselines, the budget-feasibility chain is derived analytically from stated assumptions, and the paper contains no load-bearing self-citations.

full rationale

The paper's central claims are empirical measurements, not fitted predictions. Static-model dimension/FLOPs/accuracy gains (Section VII-B, Table III) are out-of-sample results compared against external baselines (DeepJSCC [3], VFE [13], GatedJSCC [10]); no fitted parameter is renamed as a prediction. Dynamic-model accuracy at each budget B (Table IV, Figs. 6-9) is a genuine measurement of test-set accuracy after thresholds are set on the validation set, which is standard model selection. The budget-feasibility chain (Eqs. 17-18, Lemmas 1-2, Proposition 2, Appendices B-D) is derived analytically from the definition Pr_k = r_k / sum r_j and monotonicity of the expected FLOPs R(r); it does not assume the accuracy results it supports. The convergence analysis (Theorem 1) is the standard nonconvex SGD bound under stated smoothness and bounded-variance assumptions (Assumptions 1-3, [30], [31]) and is applied generically; it neither assumes nor derives the system's specific output. The skeptical concern that Algorithm 2 sets per-exit marginal pass rates while Algorithm 3's sequential rule realizes the joint exit distribution (so realized FLOPs can deviate from the nominal R used in Proposition 2, and measured FLOPs are not reported to close the loop) is a legitimately flagged correctness/validation gap for the budget-satisfaction claim, but it is not a circular reduction: accuracy and realized computation are not forced by construction. No self-citations by the present authors appear anywhere in the reference list, so self-citation and imported-uniqueness patterns are absent. The complexity-sorting claim (Section VII-C.1) is partly definitional since complexity is operationalized via the confidence score, but that illustrative claim is not load-bearing for any quantitative result. No circular step can be exhibited, so the honest finding is no significant circularity.

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

The central claims rest on two calibrated components: validation-set exit thresholds and the exit-probability hyperparameter r, both fitted to data. The theoretical results use standard smooth nonconvex SGD assumptions plus a domain assumption that validation and test distributions match. No new physical or conceptual entities are postulated.

free parameters (6)
  • Exit thresholds theta_k = Calibrated on validation set
    Set in Algorithm 2 so that measured validation exit frequencies approximate target probabilities P rk; they transfer to the test set only under distribution similarity.
  • Exit probability hyperparameter r = Chosen per budget B, no values reported
    Determines P rk via Eq. (17); the paper proves a feasible upper bound rup exists but does not report chosen values for the experiments.
  • Transmitted feature dimension = 16 for CIFAR-10/100, 64 for ImageNet
    Selected via ablation in Appendix I; increasing beyond 16 gave marginal accuracy gains that did not justify added transmission cost.
  • Per-exit loss weight omega_k = 1
    Set to an empirical value taken from reference [28]; not tuned within this paper.
  • Step-size constant d in alpha = d/sqrt(T) = Unspecified constant
    Appears in the O(1/sqrt(T)) convergence bound; the bound holds for any positive d, so it is a free constant in the analysis.
  • Number of exits K = 5
    Pre-set based on device capability and processing needs; no sensitivity analysis is provided.
assumptions (6)
  • standard math Assumptions 1-3: twice-differentiable loss, Lipschitz-continuous gradients with constant L, and bounded stochastic gradient noise with variance sigma^2
    Standard smooth nonconvex SGD conditions, stated in Section VI as approximately satisfied by smooth-activation NNs.
  • standard math Stochastic gradients are unbiased, E[grad f_tilde_t | w_t] = grad f(w_t)
    Implicitly used in the expectation step of Lemma 3 (Eq. 23) but never stated; mini-batch SGD normally satisfies this.
  • domain assumption Training, validation, and test sets have similar distributions
    Stated in Section V; load-bearing because validation-calibrated exit thresholds are applied to the test set.
  • ad hoc to paper Exits are evenly distributed in computation, so FLOPs between consecutive exits is a constant phi
    Section V.C.2 claims this can be implemented by choosing exits in the NN, but the actual multi-scale network is not verified to satisfy it.
  • domain assumption The softmax max-logit confidence phi_j,k is a valid proxy for input complexity and classifier correctness
    Used to route simple images to early exits; supported only indirectly through example sortings in Fig. 5.
  • domain assumption Channel model Y = hX + epsilon with known PSNR, treated as a non-trainable NN layer
    Standard in the semantic communication literature; required for joint training over AWGN and Rayleigh channels.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Computation-resource-efficient Task-oriented Communications." pith.science (2026). https://pith.science/paper/NLRDWX3L

@misc{pith2026250707422,
  author       = {Pith},
  title        = {Pith review of: Computation-resource-efficient Task-oriented Communications},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NLRDWX3L}},
  note         = {Machine review of arXiv:2507.07422}
}
abstract

The rapid development of deep-learning enabled task-oriented communications (TOC) significantly shifts the paradigm of wireless communications. However, the high computation demands, particularly in resource-constrained systems e.g., mobile phones and UAVs, make TOC challenging for many tasks. To address the problem, we propose a novel TOC method with two models: a static and a dynamic model. In the static model, we apply a neural network (NN) as a task-oriented encoder (TOE) when there is no computation budget constraint. The dynamic model is used when device computation resources are limited, and it uses dynamic NNs with multiple exits as the TOE. The dynamic model sorts input data by complexity with thresholds, allowing the efficient allocation of computation resources. Furthermore, we analyze the convergence of the proposed TOC methods and show that the model converges at rate $O\left(\frac{1}{\sqrt{T}}\right)$ with an epoch of length $T$. Experimental results demonstrate that the static model outperforms baseline models in terms of transmitted dimensions, floating-point operations (FLOPs), and accuracy simultaneously. The dynamic model can further improve accuracy and computational demand, providing an improved solution for resource-constrained systems.

Figures

Figures reproduced from arXiv: 2507.07422 by the authors.

Figure 1
Figure 1. Example of classifying a bird [7] with one simple [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. System schematics of the proposed TOC models. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. (a) The NN structure of Dynamic TOE, which takes image samples as inputs, and extracts task-related features for [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: The training and validation loss of the proposed static model in AWGN and Rayleigh channels in CIFAR-100 dataset. [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: Dynamic model outputs at PSNR = 6 dB in Rayleigh [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 8
Figure 8. Figure 8: Accuracy as a function of budget B in Rayleigh fading channel on ImageNet dataset [PITH_FULL_IMAGE:figures/full_fig_p011_8.png]
Figure 6
Figure 6. Figure 6: Accuracy as a function of budget B in AWGN and Rayleigh fading channels in CIFAR-100 dataset [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: Accuracy as a function of budget B in AWGN channel on CIFAR-10 dataset. useful when devices with different computation budgets are used in TOC. Additionally, it is observed that the highest accuracy of the proposed dynamic model also outperforms the two baseline models…
Figure 10
Figure 10. Figure 10: The training and validation loss for AWGN and Rayleigh fading channels in the dynamic TOC model on the [PITH_FULL_IMAGE:figures/full_fig_p012_10.png]
Figure 11
Figure 11. Figure 11: Model accuracy under different channel dimensions [PITH_FULL_IMAGE:figures/full_fig_p015_11.png]
Figure 12
Figure 12. Figure 12: Ablation study on learning rate in Rayleigh channel [PITH_FULL_IMAGE:figures/full_fig_p015_12.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

36 extracted references · 32 canonical work pages

  1. [1]

    Weaver, The mathematical theory of communication

    W. Weaver, The mathematical theory of communication . Uni- versity of Illinois Press, 1963

    work page 1963

  2. [2]

    Deep learning- based end-to-end wireless communication systems with con- ditional gans as unknown channels,

    H. Ye, L. Liang, G. Y . Li, and B.-H. Juang, “Deep learning- based end-to-end wireless communication systems with con- ditional gans as unknown channels,” IEEE Trans. Wirel. Commun., vol. 19, no. 5, pp. 3133–3143, 2020

    work page 2020

  3. [3]

    Deep joint source-channel coding for wireless image transmission,

    E. Bourtsoulatze, D. B. Kurka, and D. G ¨und¨uz, “Deep joint source-channel coding for wireless image transmission,” IEEE Trans. Cogn. Commun. Netw. , vol. 5, no. 3, pp. 567–579, 2019

    work page 2019

  4. [4]

    Deep learning en- abled semantic communication systems,

    H. Xie, Z. Qin, G. Y . Li, and B.-H. Juang, “Deep learning en- abled semantic communication systems,” IEEE Trans. Signal Process., vol. 69, pp. 2663–2675, 2021

    work page 2021

  5. [5]

    6g networks: Beyond shan- non towards semantic and goal-oriented communications,

    E. C. Strinati and S. Barbarossa, “6g networks: Beyond shan- non towards semantic and goal-oriented communications,” Comput. Netw., vol. 190, p. 107 930, 2021. (a) Training loss at different learning rates. (b) Model performance at different learning rates. Fig. 12: Ablation study on learning rate in Rayleigh channel at PSNR = 12dB on CIFAR-100 dataset

    work page 2021

  6. [6]

    A lite distributed semantic communication system for internet of things,

    H. Xie and Z. Qin, “A lite distributed semantic communication system for internet of things,” IEEE J. Sel. Areas Commun. , vol. 39, no. 1, pp. 142–153, 2020

    work page 2020

  7. [7]

    Imagenet: A large-scale hierarchical image database,

    J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei- Fei, “Imagenet: A large-scale hierarchical image database,” in IEEE Conf. Comput. Vis. Pattern Recognit. , 2009, pp. 248–

    work page 2009

  8. [8]

    Multi-scale dense networks for resource efficient image classification,

    G. Huang, D. Chen, T. Li, F. Wu, L. Van Der Maaten, and K. Q. Weinberger, “Multi-scale dense networks for resource efficient image classification,” arXiv preprint arXiv:1703.09844, 2017

    arXiv 2017

Show all 36 references
  1. [9]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit., 2016, pp. 770–778

    2016

  2. [10]

    Semantic commu- nications for image recovery and classification via deep joint source and channel coding,

    Z. Lyu, G. Zhu, J. Xu, B. Ai, and S. Cui, “Semantic commu- nications for image recovery and classification via deep joint source and channel coding,” IEEE Transactions on Wireless Communications, vol. 23, no. 8, pp. 8388–8404, 2024

    2024

  3. [11]

    Task-oriented multi-user semantic communications for vqa,

    H. Xie, Z. Qin, and G. Y . Li, “Task-oriented multi-user semantic communications for vqa,” IEEE Wirel. Commun. Lett., vol. 11, no. 3, pp. 553–557, 2021

    2021

  4. [12]

    Performance optimiza- tion for semantic communications: An attention-based rein- forcement learning approach,

    Y . Wang, M. Chen, T. Luo, et al. , “Performance optimiza- tion for semantic communications: An attention-based rein- forcement learning approach,” IEEE J. Sel. Areas Commun. , vol. 40, no. 9, pp. 2598–2613, 2022

    2022

  5. [13]

    Learning task-oriented communication for edge inference: An information bottleneck approach,

    J. Shao, Y . Mao, and J. Zhang, “Learning task-oriented communication for edge inference: An information bottleneck approach,” IEEE J. Sel. Areas Commun. , vol. 40, no. 1, pp. 197–211, 2021

    2021

  6. [14]

    Robust information bottleneck for task-oriented communication with digital modulation,

    S. Xie, S. Ma, M. Ding, Y . Shi, M. Tang, and Y . Wu, “Robust information bottleneck for task-oriented communication with digital modulation,” IEEE J. Sel. Areas Commun. , 2023

    2023

  7. [15]

    Multi-task-oriented broadcast for edge ai inference via information bottleneck,

    Y . Yang, Y . Wu, S. Ma, and Y . Shi, “Multi-task-oriented broadcast for edge ai inference via information bottleneck,” JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 16 in Proc. IEEE Global Commun. Conf. (GLOBECOM) , IEEE, 2023, pp. 3142–3147

    2021

  8. [16]

    Task-oriented communication for multidevice cooperative edge inference,

    J. Shao, Y . Mao, and J. Zhang, “Task-oriented communication for multidevice cooperative edge inference,” IEEE Trans. Wirel. Commun., vol. 22, no. 1, pp. 73–87, 2022

    2022

  9. [17]

    Distributed foundation models for multi-modal learning in 6g wireless networks,

    J. Du, T. Lin, C. Jiang, Q. Yang, C. F. Bader, and Z. Han, “Distributed foundation models for multi-modal learning in 6g wireless networks,” IEEE Wireless Communications , vol. 31, no. 3, pp. 20–30, 2024

    2024

  10. [18]

    Task-oriented image transmission for scene classification in unmanned aerial systems,

    X. Kang, B. Song, J. Guo, Z. Qin, and F. R. Yu, “Task-oriented image transmission for scene classification in unmanned aerial systems,” IEEE Trans. Commun. , vol. 70, no. 8, pp. 5181– 5192, 2022

    2022

  11. [19]

    Genet: A graph neural network-based anti-noise task-oriented semantic communication paradigm,

    C. Zheng and K. Cai, “Genet: A graph neural network-based anti-noise task-oriented semantic communication paradigm,” arXiv preprint arXiv:2403.18296 , 2024

    2024 arXiv

  12. [20]

    Resource allocation for text semantic communications,

    L. Yan, Z. Qin, R. Zhang, Y . Li, and G. Y . Li, “Resource allocation for text semantic communications,” IEEE Wireless Communications Letters, vol. 11, no. 7, pp. 1394–1398, 2022

    2022

  13. [21]

    Adaptable semantic compression and resource allocation for task-oriented commu- nications,

    C. Liu, C. Guo, Y . Yang, and N. Jiang, “Adaptable semantic compression and resource allocation for task-oriented commu- nications,” IEEE Transactions on Cognitive Communications and Networking, 2023

    2023

  14. [22]

    Adaptive resource allocation for semantic communication networks,

    L. Wang, W. Wu, F. Zhou, Z. Yang, Z. Qin, and Q. Wu, “Adaptive resource allocation for semantic communication networks,” IEEE Transactions on Communications , 2024

    2024

  15. [23]

    Re- thinking resource management in edge learning: A joint pre- training and fine-tuning design paradigm,

    Z. Lyu, Y . Li, G. Zhu, J. Xu, H. V . Poor, and S. Cui, “Re- thinking resource management in edge learning: A joint pre- training and fine-tuning design paradigm,” IEEE Transactions on Wireless Communications, 2024

    2024

  16. [24]

    Scan: Semantic commu- nication with adaptive channel feedback,

    G. Zhang, Q. Hu, Y . Cai, and G. Yu, “Scan: Semantic commu- nication with adaptive channel feedback,” IEEE Transactions on Cognitive Communications and Networking , 2024

    2024

  17. [25]

    Dynamic neural networks: A survey,

    Y . Han, G. Huang, S. Song, L. Yang, H. Wang, and Y . Wang, “Dynamic neural networks: A survey,” IEEE Trans. Pattern Anal. Mach. Intell. , vol. 44, no. 11, pp. 7436–7456, 2021

    2021

  18. [26]

    Semantic communication with adaptive universal transformer,

    Q. Zhou, R. Li, Z. Zhao, C. Peng, and H. Zhang, “Semantic communication with adaptive universal transformer,” IEEE Wirel. Commun. Lett., vol. 11, no. 3, pp. 453–457, 2021

    2021

  19. [27]

    Feature pyramid networks for object detection,

    T.-Y . Lin, P. Doll ´ar, R. Girshick, K. He, B. Hariharan, and S. Belongie, “Feature pyramid networks for object detection,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. , 2017, pp. 2117–2125

    2017

  20. [28]

    Densely connected convolutional networks,

    G. Huang, Z. Liu, L. van der Maaten, and K. Q. Weinberger, “Densely connected convolutional networks,” in Proceedings of the IEEE Conf. Comput. Vis. Pattern Recognit. , Jul. 2017

    2017

  21. [29]

    University, Lecture 5: Adaptive methods and non-convex optimization, https : / / www

    C. University, Lecture 5: Adaptive methods and non-convex optimization, https : / / www. cs . cornell . edu / courses / cs6787 / 2024sp/lectures/Lecture5.pdf, Accessed: 2024-07-14, 2024

    2024

  22. [30]

    Optimization methods for large-scale machine learning,

    L. Bottou, F. E. Curtis, and J. Nocedal, “Optimization methods for large-scale machine learning,” SIAM review, vol. 60, no. 2, pp. 223–311, 2018

    2018

  23. [31]

    Stochastic first-and zeroth-order methods for nonconvex stochastic programming,

    S. Ghadimi and G. Lan, “Stochastic first-and zeroth-order methods for nonconvex stochastic programming,” SIAM jour- nal on optimization , vol. 23, no. 4, pp. 2341–2368, 2013

    2013

  24. [32]

    Krizhevsky et al., Learning multiple layers of features from tiny images, 2009

    A. Krizhevsky et al., Learning multiple layers of features from tiny images, 2009. [Online]. Available: https://www.cs.toronto. edu/∼kriz/learning-features-2009-TR.pdf

    2009

  25. [33]

    An image is worth 16x16 words: Transformers for image recognition at scale,

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, et al. , “An image is worth 16x16 words: Transformers for image recognition at scale,” ICLR, 2020

    2020

  26. [34]

    Overfitting mechanism and avoidance in deep neural networks,

    S. Salman and X. Liu, “Overfitting mechanism and avoidance in deep neural networks,” arXiv preprint arXiv:1901.06566 , 2019

    1901 arXiv

  27. [35]

    Training data-efficient image transformers & distillation through attention,

    H. Touvron, M. Cord, M. Douze, F. Massa, A. Sablayrolles, and H. J ´egou, “Training data-efficient image transformers & distillation through attention,” in International conference on machine learning, PMLR, 2021, pp. 10 347–10 357. Jingwen Fu received the B.S. degree from Bei...

    2021

  28. [255]

    DOI: 10.1109/CVPR.2009.5206848

    2009

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.