Pith. sign in

REVIEW 4 major objections 5 minor 31 references

Generative AI Empowered Semantic Feature Multiple Access (SFMA) Over Wireless Networks

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read A power-dependent semantic interference factor in the SINR equation lets one superimposed signal serve two video users, with simulated rate gains up to 66.1% over OFDMA.

desk verdict The SFMA concept is genuinely new and worth knowing, but the headline rate gains are not supported because the semantic interference factor rho that powers every rate calculation is never specified. read the letter →

arxiv 2412.20937 v1 pith:2GZQCVPY submitted 2024-12-30 eess.SP

Jiaxiang Wang , Yinchao Yang , Zhaohui Yang , Chongwen Huang , Mingzhe Chen , Zhaoyang Zhang , Mohammad Shikh-Bahaei This is my paper · ORCID
classification eess.SP
keywords semanticcommunicationgenerativeAImultipleaccessnon-orthogonalvideoframeinterpolationpowerallocationinterferenceuserpairing
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 base station can serve two paired video users at once by superimposing their semantic features into a single signal, provided interference between users is measured in semantic space rather than by raw power overlap. To make that measurement tractable, the authors insert a semantic interference factor into the standard SINR formula, so that each user's achievable rate depends on both users' transmit powers through that factor. They then decompose the joint user-pairing and power-allocation problem into a stable matching among users, an inter-group power allocation solved through KKT conditions, and an intra-group concavity-based power split. Simulations on CIFAR-10 with a generative video frame interpolator show the scheme outperforming fixed-power NOMA by 24.8%, orthogonal joint source-channel coding by 45.8%, and OFDMA by 66.1% in sum rate while preserving interpolation quality.

What carries the argument

The load-bearing mechanism is the modified SINR equation, in which a semantic interference factor converts cross-user semantic confusion into a power-dependent weight in the denominator. Because this factor is assumed to depend on the two users' transmit powers, the rate expression becomes intrinsically coupled across users, which is what forces the three-stage solution: a stable matching algorithm for user pairing, a KKT-based computation of extreme power points for inter-group allocation, and a concave one-dimensional search for the intra-group split. The same semantic interference factor appears in the KKT derivative condition, so the entire optimization is carried by this single semantic quantity.

What would settle it

Run the trained semantic encoder-decoder over a grid of power pairs at fixed channels, measure the actual MSE-based SINR, and check whether a single choice of the semantic interference factor inserted into the modified SINR equation predicts the rates used in the optimization; if no such function fits the data, the reported gains and allocation policies do not follow.

Watch

Extended reading notes

Core claim

The paper's central discovery is that the physical-layer SINR formula, which treats a superimposed user's signal purely as power interference, is the wrong performance model for semantic multiple access. Using the MSE between the original and reconstructed frames as the true SINR, the authors show a gap between measured semantic performance and the standard formula, and close that gap by introducing a semantic interference factor that scales the interfering user's power in the denominator of the SINR. With this modified SINR, the sum rate of a pair becomes a function of the group power budget through that factor, which justifies a two-level power allocation: first across groups, then within each group. The optimized SFMA system achieves the reported gains and, when paired users' frames are temporally close, the GAI interpolation model produces intermediate frames with MS-SSIM around 0.82 and LPIPS around 0.05, indicating that the spectral-efficiency gain does not come at the cost of video quality.

Load-bearing premise

The semantic interference factor is treated as a known, deterministic, differentiable function of transmit power in the SINR equation and in the KKT derivative, but the paper provides no closed-form expression, no estimation algorithm, and no measurement procedure for it, so the rate gains and power allocations cannot be reproduced without that missing function.

Editorial extensions

If this is right

  • If SFMA works as claimed, a base station can double the number of video users served per resource block without splitting bandwidth, because the semantic decoder and the GAI interpolator jointly suppress the superimposed user's interference.
  • The power-dependent semantic interference factor turns user pairing and power control into one coupled design problem, so systems that fix power allocation (F-NOMA) or use orthogonal bandwidth (OFDMA, O-JSCC) are leaving throughput on the table.
  • The measured MS-SSIM and LPIPS results imply that video quality stays high when the temporal gap between paired users' frames is small, making the temporal gap a first-class resource to schedule rather than just a transmission artifact.
  • The three-user extension indicates that adding more users per group increases semantic interference and reduces fairness, so the two-user pairing design is a deliberate compromise between spectral efficiency and decoding complexity.

Reading between the lines

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

  • A natural next step, not pursued in the paper, is to estimate the semantic interference factor from the trained encoder-decoder by measuring MSE under many power pairs and fitting a function; the same optimization machinery would then apply at deployment time.
  • Because the semantic interference factor is architecture- and content-dependent, the reported rate gains are likely to shift for different video datasets or interpolation models; the method's general claim is the SINR structure, not the specific percentages.
  • The SINR modification suggests a general recipe for other semantic multiple-access schemes: replace physical interference weights with data-derived semantic weights, then reuse standard resource-allocation tools that assume a power-dependent SINR.
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

4 major / 5 minor

Summary. The paper proposes SFMA, a downlink semantic multiple access scheme in which a base station pairs users into groups of two, superimposes their semantic features, and transmits a combined signal; receivers reconstruct their frames and use a GAI-based video frame interpolation model to generate intermediate frames. To capture semantic-level interference, the authors modify the standard SINR by introducing a semantic interference factor rho_21(pk,1,pk,2), then formulate a sum-rate maximization problem with a temporal-gap penalty, which they decompose into user pairing (Gale-Shapley), inter-group power allocation (KKT-based), and intra-group power allocation (gradient descent). Simulations claim rate gains of up to 24.8%, 45.8%, and 66.1% over F-NOMA, O-JSCC, and OFDMA, together with favorable MS-SSIM and LPIPS interpolation quality.

Significance. If the modified SINR in Eq. (9) were a validated achievable-rate expression, the SFMA concept would be of interest to the semantic communication and multiple access communities: treating semantic interference as a power-dependent scaling factor and pairing users via stable matching is a plausible design direction, and the use of GAI interpolation to exploit temporal correlation is timely. The paper provides a clear system architecture, a structured three-step solution, and a concrete comparison setup. However, the central metric is not specified, and the current evidence does not establish that the reported gains are real, reproducible, or attributable to the proposed system rather than to the fitted interference factor.

major comments (4)
  1. [Section III-A, Eq. (9)] The semantic interference factor rho_21(pk,1,pk,2) is never defined. The paper gives no closed form, no estimation algorithm, and no measurement protocol; Fig. 5 shows a simulated surface but without axis labels, numeric values, or the underlying formula. Since Eq. (9) defines the SINR used in the rate expression (10), the objective (12), the constraints (17a), and the KKT conditions (27), every rate claim in Section V-A depends on an unspecified quantity. The optimization problem (17) is therefore not well-posed as stated, and the reported gains cannot be reproduced by a reader.
  2. [Section IV-B, Eq. (27a) and Lemma 1] The KKT stationarity condition (27a) contains the derivative rho'(pk) (written as rho_21'(pk) and rho_12'(pk)), and Lemma 1's extreme points (19)-(21) require evaluating rho at specific power values. Because no functional form or differentiability assumptions for rho are provided, these conditions cannot be checked or computed. Moreover, the paper does not establish the convexity or regularity conditions needed for the KKT conditions to characterize a global optimum; the derivation simply asserts the KKT system as the proof of Lemma 1.
  3. [Section IV-C, paragraph after Eq. (23)] The claim that the objective in (23) is 'a sum of two concave logarithmic functions with respect to pk,1' is not established and is, in general, false. For fixed pk, rk,1(pk,1) has the form log(1 + A p1/(rho(pk - p1) + sigma^2)), which is not generally concave; for example, in the interference-dominated regime where the +1 is negligible, it behaves like log(c/(b - p1)), which is convex. Consequently, the gradient-descent step in Algorithm 2 is not guaranteed to converge to a maximum, and the decomposition of (22) into independent per-group problems requires a proof that is not supplied.
  4. [Section V-A, Figs. 6-7] The headline gains (24.8%, 45.8%, 66.1%) are computed from the rate expression (10) that incorporates the fitted rho. Since rho is calibrated from the very system it describes (Fig. 5), comparing these rates against standard Shannon-rate baselines is circular unless the modified SINR is independently validated as an achievable rate for the actual semantic codec, for example, by relating rho to measured end-to-end distortion. The paper provides no such validation, and the comparisons in Figs. 6 and 7 report no error bars or multiple trials, so the claimed margins are not shown to be significant.
minor comments (5)
  1. [Section VI, conclusion paragraph] The final paragraph of the conclusion states that the proposed method yields 'significant improvements in terms of positioning accuracy,' which is unrelated to the transmission-rate results reported in the paper; this appears to be a leftover from another manuscript and should be corrected.
  2. [Section V-D, paragraph on three-user case] The sentence 'there will be random 2 users whose performances are the same' is unclear; it should say 'two of the users' or 'a randomly chosen pair of users,' and the comparison should be quantified.
  3. [Section IV-B, Algorithm 1, line 4] Algorithm 1 sets pk,1 = pk,2 in each iteration, which is inconsistent with the earlier notation in which pk,1 and pk,2 follow a fixed power allocation factor eta; the algorithm should state how eta is chosen and how the extreme points (19)-(20) are evaluated under that constraint.
  4. [Fig. 5] The axes of Fig. 5 are not fully labeled and no colorbar or numerical scale is provided, so the reader cannot infer the range or behavior of rho from the figure.
  5. [Eq. (18) and throughout Section IV-B] The subscripts of rho are used inconsistently: Eq. (18) writes rho_21(pk) and rho_12(pk), while the earlier definition in Eq. (9) uses rho_21(pk,1, pk,2); the relationship between the two-argument and one-argument forms should be stated explicitly.

Circularity Check

2 steps flagged · score 6.0 of 10

The reported rate gains are driven by a semantic interference factor rho that is calibrated to the system's own simulation and then inserted into the rate formula, so the 'predicted' transmission-rate advantage is partly constructed by the metric.

  1. fitted input called prediction [Section III-A, Eqs. (8)-(11), Fig. 5; Section V-A, Figs. 6-7]
    "To accurately represent the SINR of semantic communications, we introduce a semantic interference factor ρ into (7) such that the SINR of the signal xk,1 is denoted by (9) ... Fig. 5 shows how ρk21 varies with (pk, SNRk,1) when pk,1 = pk,2. This relationship can be represented as a function of pk,1 and pk,2, i.e., ρk21 = ρk21 (pk,1, pk,2). Given (9), the achievable rate of user i in group k is rk,i (pk,1, pk,2) = log2 (1 + γk,i (pk,1, pk,2))."

    The interference factor rho is obtained from simulation of the same semantic system that the rate formula then evaluates: the paper first uses MSE-based SINR to show that ordinary SINR understates semantic performance, then calibrates rho so that Eq. (9) represents that performance. Consequently Eq. (10) is not an independent achievable-rate expression; it is log(1 + a SINR whose interference term has been adjusted to match the system being tested. The sum-rate comparisons in Figs. 6-7 therefore compare F-NOMA/O-JSCC/OFDMA under ordinary SINR with SFMA under a metric pre-fitted to favor semantic interference suppression. No closed form or estimation procedure for rho is given, so the rate gain is not reproducible outside the calibrating simulation.

  2. self definitional [Eqs. (8)-(9) in Section III-A]
    "we can use the mean square error (MSE) between the original frame xt1 k,1 and the recovered frame ˆxt1 k,1 to represent the SINR ... To accurately represent the SINR of semantic communications, we introduce a semantic interference factor ρ into (7) ..."

    Eq. (8) defines the semantic SINR using the system's own reconstruction error, and Eq. (9) is then declared to represent this SINR by inserting rho. The rate metric is thus defined in terms of the system's reconstruction performance, while the paper's headline claim is that the system, evaluated with this same metric, achieves higher transmission rates. This is a self-definitional loop: the performance measure and the measured system are mutually calibrated, so the 'prediction' of superior rate is an artifact of how rho was chosen rather than an independent derivation.

full rationale

The central circularity is in the semantic SINR model. The paper motivates rho by simulation (Fig. 4) and then plots rho as a function of transmit power (Fig. 5), but never specifies a closed form, fitting rule, or measurement protocol. Because Eq. (10) defines the achievable rate through this calibrated rho, the optimized SFMA rates in Figs. 6-7 are not grounded in an externally defined channel capacity; they are logarithms of a SINR expression whose interference term was adjusted to match the semantic system's simulated behavior. The optimization itself is a genuine mathematical exercise, and the GAI interpolation results (MS-SSIM, LPIPS) are independent, so this is not a pure identity or a fully fabricated result. However, the central rate-improvement claim is partially forced by the construction of the rate metric. I found no load-bearing self-citation: [15] and [18] are cited as background on prior semantic MA work, not as the justification for rho or the rate model, and the user-pairing and interpolation components rely on external algorithms (Gale-Shapley, TAIN) and external baselines. The paper also contains a non-circular but serious correctness risk: the concavity assertion for Eq. (23) is unjustified because rho(pk,1) may destroy log-concavity, and the unspecified rho'(pk) in Eq. (27a) makes the KKT system unevaluable.

Assumptions & free parameters 3 free parameters · 4 assumptions · 1 invented entities

The central rate engine of the paper is the modified SINR containing rho, which is fitted rather than derived. All optimization and all reported gains inherit this fitted quantity, so the ledger is dominated by rho plus hand-set objective weights.

free parameters (3)
  • semantic interference factor rho_21(pk,1,pk,2) = not given, shown graphically in Fig. 5
    Introduced in Eq. (9) to force the SINR equation to match simulated semantic communication performance; all rates in Eq. (10) depend on it.
  • rate-gap trade-off weight alpha = not specified
    Chosen by hand in objective (12) to balance transmission rate versus temporal gap; directly shapes user pairing.
  • maximum temporal gap Delta = 4 and 16 in experiments
    Design threshold in constraint (12c); determines which user pairs are allowed and is a hyperparameter, not derived.
assumptions (4)
  • domain assumption The achievable rate of a semantic link equals log2(1 + modified SINR) from Eq. (10)
    Standard Shannon rate is applied to a semantic system after inserting the fitted rho; no information-theoretic derivation is provided.
  • ad hoc to paper The semantic interference factor rho is a deterministic and differentiable function of transmit power pk
    Needed for the KKT condition (27a) which contains rho'(pk); the paper gives no closed form, only a simulation curve.
  • domain assumption MSE between original and reconstructed frames can be used as inverse SINR in Eq. (8)
    Used to justify the modified SINR; assumes pixel-domain distortion fully captures semantic communication quality.
  • domain assumption Perfect channel state information is available at the BS
    User pairing and power allocation use perfect CSI in Sections IV-A and IV-B.
invented entities (1)
  • Semantic interference factor rho
    purpose: To absorb the gap between standard SINR and simulated semantic communication performance in Eq. (9).
    No theoretical definition, measurement protocol, or external falsifiable prediction is provided; it is calibrated from the system's own simulation in Fig. 5.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Generative AI Empowered Semantic Feature Multiple Access (SFMA) Over Wireless Networks." pith.science (2026). https://pith.science/paper/2GZQCVPY

@misc{pith2026241220937,
  author       = {Pith},
  title        = {Pith review of: Generative AI Empowered Semantic Feature Multiple Access (SFMA) Over Wireless Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2GZQCVPY}},
  note         = {Machine review of arXiv:2412.20937}
}
read the original abstract

This paper investigates a novel generative artificial intelligence (GAI) empowered multi-user semantic communication system called semantic feature multiple access (SFMA) for video transmission, which comprises a base station (BS) and paired users. The BS generates and combines semantic information of several frames simultaneously requested by paired users into a single signal. Users recover their frames from this combined signal and input the recovered frames into a GAI-based video frame interpolation model to generate the intermediate frame. To optimize transmission rates and temporal gaps between simultaneously transmitted frames, we formulate an optimization problem to maximize the system sum rate while minimizing temporal gaps. Since the standard signal-to-interference-plus-noise ratio (SINR) equation does not accurately capture the performance of our semantic communication system, we introduce a weight parameter into the SINR equation to better represent the system's performance. Due to its dependence on transmit power, we propose a three-step solution. First, we develop a user pairing algorithm that pairs two users with the highest preference value, a weighted combination of semantic transmission rate and temporal gap. Second, we optimize inter-group power allocation by formulating an optimization problem that allocates proper transmit power across all user groups to maximize system sum rates while satisfying each user's minimum rate requirement. Third, we address intra-group power allocation to enhance each user's performance. Simulation results demonstrate that our method improves transmission rates by up to 24.8%, 45.8%, and 66.1% compared to fixed-power non-orthogonal multiple access (F-NOMA), orthogonal joint source-channel coding (O-JSCC), and orthogonal frequency division multiple access (OFDMA), respectively.

Figures

Figures reproduced from arXiv: 2412.20937 by the authors.

Figure 1
Figure 1. System model of a downlink SFMA single cell network. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The framework of GAI-enabled semantic video transmission in the SFMA system. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. The architecture of the Attention-based JSCC network. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (5 more)
Figure 5
Figure 5. Figure 5: The impact of the semantic interference factor [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Total sum rate of the system versus different number of users. [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: Total sum rate of the system versus different power of BS [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 8
Figure 8. Figure 8: PSNR of the two reconstructed images in our Attention-based [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]
Figure 10
Figure 10. Figure 10: Performance comparison of LPIPS versus different SNR [PITH_FULL_IMAGE:figures/full_fig_p011_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

31 extracted references · 18 canonical work pages

  1. [1]

    A vision of 6g wireless systems: Applications, trends, technologies, and open research problems,

    W. Saad, M. Bennis, and M. Chen, “A vision of 6g wireless systems: Applications, trends, technologies, and open research problems,” IEEE Network, vol. 34, no. 3, pp. 134–142, 2020

    2020

  2. [2]

    Distributed learning in wireless networks: Recent progress and future challenges,

    M. Chen, D. G ¨und¨uz, K. Huang, W. Saad, M. Bennis, A. V . Feljan, and H. V . Poor, “Distributed learning in wireless networks: Recent progress and future challenges,” IEEE Journal on Selected Areas in Communications, vol. 39, no. 12, pp. 3579–3605, 2021

    2021

  3. [3]

    Deep learning enabled semantic communication systems,

    H. Xie, Z. Qin, G. Y . Li, and B.-H. Juang, “Deep learning enabled semantic communication systems,” IEEE Transactions on Signal Pro- cessing, vol. 69, pp. 2663–2675, 2021

    2021

  4. [4]

    Task-oriented multi- user semantic communications,

    H. Xie, Z. Qin, X. Tao, and K. B. Letaief, “Task-oriented multi- user semantic communications,” IEEE Journal on Selected Areas in Communications, vol. 40, no. 9, pp. 2584–2597, 2022

    work page 2022

  5. [5]

    Performance optimization for semantic communications: An attention- based reinforcement learning approach,

    Y . Wang, M. Chen, T. Luo, W. Saad, D. Niyato, H. V . Poor, and S. Cui, “Performance optimization for semantic communications: An attention- based reinforcement learning approach,”IEEE Journal on Selected Areas in Communications, vol. 40, no. 9, pp. 2598–2613, 2022

    2022

  6. [6]

    Semantic communication systems for speech transmission,

    Z. Weng and Z. Qin, “Semantic communication systems for speech transmission,” IEEE Journal on Selected Areas in Communications , vol. 39, no. 8, pp. 2434–2444, 2021

    2021

  7. [7]

    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 Transactions on Cognitive Communications and Networking , vol. 5, no. 3, pp. 567–579, 2019

    2019

  8. [8]

    Deepwive: Deep-learning-aided wireless video transmission,

    T.-Y . Tung and D. G ¨und¨uz, “Deepwive: Deep-learning-aided wireless video transmission,” IEEE Journal on Selected Areas in Communica- tions, vol. 40, no. 9, pp. 2570–2583, 2022

    2022

Show all 31 references
  1. [9]

    Cddm: Channel denoising diffusion models for wireless communications,

    T. Wu, Z. Chen, D. He, L. Qian, Y . Xu, M. Tao, and W. Zhang, “Cddm: Channel denoising diffusion models for wireless communications,” in GLOBECOM 2023-2023 IEEE Global Communications Conference , pp. 7429–7434, IEEE, 2023

    2023

  2. [10]

    Dmce: Diffusion model channel enhancer for multi-user semantic communication systems,

    Y . Zeng, X. He, X. Chen, H. Tong, Z. Yang, Y . Guo, and J. Hao, “Dmce: Diffusion model channel enhancer for multi-user semantic communication systems,” arXiv preprint arXiv:2401.16017 , 2024

    2024 arXiv

  3. [11]

    A hybrid wireless image transmission scheme with diffusion,

    X. Niu, X. Wang, D. G ¨und¨uz, B. Bai, W. Chen, and G. Zhou, “A hybrid wireless image transmission scheme with diffusion,” in 2023 IEEE 24th International Workshop on Signal Processing Advances in Wireless Communications (SPAWC), pp. 86–90, IEEE, 2023

    2023

  4. [12]

    Generative model based highly efficient semantic communication approach for im- age transmission,

    T. Han, J. Tang, Q. Yang, Y . Duan, Z. Zhang, and Z. Shi, “Generative model based highly efficient semantic communication approach for im- age transmission,” in ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pp. 1–5, IEEE, 2023

    2023

  5. [13]

    Generative joint source-channel coding for semantic image transmission,

    E. Erdemir, T.-Y . Tung, P. L. Dragotti, and D. G¨und¨uz, “Generative joint source-channel coding for semantic image transmission,” IEEE Journal on Selected Areas in Communications , vol. 41, no. 8, pp. 2645–2657, 2023

    2023

  6. [14]

    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

  7. [15]

    Secure semantic communications: Fundamentals and challenges,

    Z. Yang, M. Chen, G. Li, Y . Yang, and Z. Zhang, “Secure semantic communications: Fundamentals and challenges,” IEEE Network, pp. 1– 1, 2024

    2024

  8. [16]

    Edge learning for b5g networks with distributed signal processing: Semantic communication, edge computing, and wireless sensing,

    W. Xu, Z. Yang, D. W. K. Ng, M. Levorato, Y . C. Eldar, and M. Debbah, “Edge learning for b5g networks with distributed signal processing: Semantic communication, edge computing, and wireless sensing,” IEEE Journal of Selected Topics in Signal Processing, vol. 17, no. 1, pp. 9...

    2023

  9. [17]

    Generative ai-driven semantic communication networks: Architecture, technologies and applications,

    C. Liang, H. Du, Y . Sun, D. Niyato, J. Kang, D. Zhao, and M. A. Imran, “Generative ai-driven semantic communication networks: Architecture, technologies and applications,” IEEE Transactions on Cognitive Com- munications and Networking , pp. 1–1, 2024

    2024

  10. [18]

    Energy efficient semantic communication over wireless networks with rate splitting,

    Z. Yang, M. Chen, Z. Zhang, and C. Huang, “Energy efficient semantic communication over wireless networks with rate splitting,”IEEE Journal 13 on Selected Areas in Communications , vol. 41, no. 5, pp. 1484–1495, 2023

    2023

  11. [19]

    Model division multiple access for semantic communications,

    P. Zhang, X. Xu, C. Dong, K. Niu, H. Liang, Z. Liang, X. Qin, M. Sun, H. Chen, N. Ma, et al. , “Model division multiple access for semantic communications,” Frontiers of Information Technology & Electronic Engineering, vol. 24, no. 6, pp. 801–812, 2023

    2023

  12. [20]

    Orthogonal model division multiple access,

    H. Liang, K. Liu, X. Liu, H. Jiang, C. Dong, X. Xu, K. Niu, and P. Zhang, “Orthogonal model division multiple access,” IEEE Trans- actions on Wireless Communications , vol. 23, no. 9, pp. 11693–11707, 2024

    2024

  13. [21]

    Non-orthogonal multiple access enhanced multi-user semantic communication,

    W. Li, H. Liang, C. Dong, X. Xu, P. Zhang, and K. Liu, “Non-orthogonal multiple access enhanced multi-user semantic communication,” IEEE Transactions on Cognitive Communications and Networking , 2023

    2023

  14. [22]

    Squeeze-and-excitation networks,

    J. Hu, L. Shen, and G. Sun, “Squeeze-and-excitation networks,” in Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 7132–7141, 2018

    2018

  15. [23]

    Efficient multi-scale attention module with cross-spatial learning,

    D. Ouyang, S. He, G. Zhang, M. Luo, H. Guo, J. Zhan, and Z. Huang, “Efficient multi-scale attention module with cross-spatial learning,” in ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pp. 1–5, IEEE, 2023

    2023

  16. [24]

    Cross-attention transformer for video interpolation,

    H. H. Kim, S. Yu, S. Yuan, and C. Tomasi, “Cross-attention transformer for video interpolation,” in Proceedings of the Asian Conference on Computer Vision, pp. 320–337, 2022

    2022

  17. [25]

    Weaver, The mathematical theory of communication

    W. Weaver, The mathematical theory of communication . University of Illinois Press, 1963

    1963

  18. [26]

    Machiavelli and the gale-shapley al- gorithm,

    L. E. Dubins and D. A. Freedman, “Machiavelli and the gale-shapley al- gorithm,” The American Mathematical Monthly, vol. 88, no. 7, pp. 485– 494, 1981

    1981

  19. [27]

    Learning to learn by gradient descent by gradient descent,

    M. Andrychowicz, M. Denil, S. Gomez, M. W. Hoffman, D. Pfau, T. Schaul, B. Shillingford, and N. De Freitas, “Learning to learn by gradient descent by gradient descent,” Advances in neural information processing systems, vol. 29, 2016

    2016

  20. [28]

    Channel attention is all you need for video frame interpolation,

    M. Choi, H. Kim, B. Han, N. Xu, and K. M. Lee, “Channel attention is all you need for video frame interpolation,” in Proceedings of the AAAI Conference on Artificial Intelligence , vol. 34, pp. 10663–10671, 2020

    2020

  21. [29]

    Impact of user pairing on 5g nonorthog- onal multiple-access downlink transmissions,

    Z. Ding, P. Fan, and H. V . Poor, “Impact of user pairing on 5g nonorthog- onal multiple-access downlink transmissions,” IEEE Transactions on Vehicular Technology, vol. 65, no. 8, pp. 6010–6023, 2016

    2016

  22. [30]

    Multiscale structural similarity for image quality assessment,

    Z. Wang, E. P. Simoncelli, and A. C. Bovik, “Multiscale structural similarity for image quality assessment,” in The Thrity-Seventh Asilomar Conference on Signals, Systems & Computers, 2003 , vol. 2, pp. 1398– 1402, Ieee, 2003

    2003

  23. [31]

    The unreasonable effectiveness of deep features as a perceptual metric,

    R. Zhang, P. Isola, A. A. Efros, E. Shechtman, and O. Wang, “The unreasonable effectiveness of deep features as a perceptual metric,” in 2018 IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion, pp. 586–595, 2018

    2018

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.