Pith. sign in

REVIEW 4 major objections 4 minor 1 cited by

LightCom: A Generative AI-Augmented Framework for QoE-Oriented Communications

T0 review · 4 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A transmitter that sends only low-pass filtered, weakly coded images, paired with a GenAI receiver, can meet QoE targets at SNRs up to 14 dB below JPEG-based baselines.

desk verdict A genuinely useful integration and a concrete perceived-coverage definition, but the 14 dB/9 dB headline numbers rest on a single image and an arbitrary QoE threshold, so treat the quantitative claims as conditional until broader evidence appears. read the letter →

arxiv 2507.17352 v1 pith:AZU2FLZX submitted 2025-07-23 eess.SP

Chunmei Xu , Siqi Zhang , Yi Ma , Rahim Tafazolli This is my paper · ORCID
classification eess.SP
keywords semanticcommunicationgenerativeAIqualityofexperienceperceivedcoveragelow-passfilteringweakchannelcodingimportance-awarepowerallocationimagerestoration
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

LightCom argues that a communication link does not need to deliver bits accurately to deliver experience: if the receiver has a powerful generative model, the transmitter can send a blurry, lightly protected version of the image and let the model fill in the missing detail. The authors construct an asymmetric system in which the transmitter only low-pass filters and applies weak channel coding, while a pre-trained generative image-restoration model reconstructs the content. They define perceived coverage with a hybrid QoE metric that combines NIQE naturalness and CLIP semantic similarity, and they show the scheme meets quality targets at SNRs up to 14 dB below JPEG-based baselines at $r=9\%$. This matters because it points toward simpler, more energy-efficient transmitters and wider effective coverage without sacrificing what users actually perceive.

What carries the argument

The machinery is an asymmetric split-inference pipeline. The source encoder is a low-pass mean filter over non-overlapping blocks that keeps a fraction $r$ of the pixels and discards high-frequency detail, reducing inter-symbol dependencies; a weak channel code (none, repetition, or Hamming) keeps residual errors independent and Gaussian-like rather than bursty. The receiver feeds the corrupted representation into a pre-trained generative image-restoration model, which uses learned natural-image priors to regenerate the discarded detail. Power allocation is importance-aware waterfilling: the compressed pixels are split into $K$ bit-level significance sequences with weights $\gamma_k = 2^{2(k-1)}$, and the transmitter distributes power across sub-channels to minimize the importance-weighted mean squared error (IMSE) between sent and decoded representations under a total power constraint.

What would settle it

Run the LightCom pipeline at $r=9\%$ with uncoded QPSK and $E_b/N_0=-2$ dB on a dense-text or fine-texture image, reconstruct with the same generative model, and check whether the QoE thresholds $D_{\mathrm{NIQE}}\leq5$ and $D_{\mathrm{CLIP}}\leq0.1$ are met; failure on any content class would falsify the claimed 14 dB robustness and 8 dB coverage gains.

Watch

Extended reading notes

Core claim

The central claim is that generative inference can substitute for bit-level reliability in low-SNR image delivery. When the transmitter sends only the low-frequency, low-dependency content of an image (low-pass filtered blocks) protected by no or weak channel coding, a generative-AI receiver using a pre-trained image-restoration model can reconstruct content that satisfies a QoE threshold even at negative $E_b/N_0$. The residual errors from weak coding, being unstructured and Gaussian-like, are more tolerable to the generative model than the bursty errors of strong codes, and importance-aware waterfilling allocates power according to bit significance to further push down the image-domain error. The result is a claimed 14 dB SNR gain over JPEG uncoded transmission and a 4.5 dB gain over JPEG with LDPC at $r=9\%$, with perceived coverage extended by roughly 7.5\textendash 8 dB at $r=33\%$\textendash$50\%$.

Load-bearing premise

The load-bearing premise is that the high-frequency parts of an image can be thrown away at the transmitter and reliably recreated at the receiver from a blurry, error-corrupted version, and this premise is tested on only three example images.

Editorial extensions

If this is right

  • At $r=9\%$, the uncoded LightCom link meets the QoE threshold at SNRs 14 dB lower than JPEG with no channel coding and 4.5 dB lower than JPEG with LDPC.
  • Perceived coverage extends by about 7.5\textendash 8 dB at near-lossless compression rates of $33\%$\textendash$50\%$, where the conventional JPEG-LDPC link already fails.
  • Increasing the compression rate can lower the SNR required to meet the QoE target, because the extra redundancy gives the generative receiver more to work with.
  • Importance-aware waterfilling power allocation outperforms equal-power allocation, particularly in semantic similarity measured by CLIP distance, even though it is optimized for image-domain MSE rather than QoE directly.
  • The transmitter's processing complexity and energy drop sharply because source coding is a mean filter and channel coding is weak or absent.

Reading between the lines

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

  • Beyond the paper: the same asymmetric design could apply to video or speech if a generative prior strong enough for those modalities exists, but the claimed gains would need to be re-established because temporal consistency and small moving objects are not covered by natural-image priors.
  • Beyond the paper: the argument that weak-code errors are Gaussian-like and independent while strong-code errors are bursty is testable directly; comparing the empirical residual-error distributions of Hamming, convolutional, and LDPC decoding on the same channel would confirm or weaken the mechanism.
  • Beyond the paper: a deployment could use the perceived-coverage map as a link-adaptation signal, skipping retransmissions when the generative receiver already meets the QoE threshold and reserving retransmissions for content the prior cannot reconstruct.
  • Beyond the paper: fine-tuning the generative receiver on error-contaminated, low-quality inputs would likely smooth the low-SNR fluctuations the paper observes; the authors themselves point to this as future work.
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 / 4 minor

Summary. The paper proposes LightCom, an asymmetric image communication framework in which a lightweight transmitter performs low-pass filtering for source coding and minimal channel coding, while a receiver augmented with a generative image-restoration model (SUPIR) reconstructs the content. The authors analyze the sufficiency and error-resilience of the resulting representation, introduce a hybrid NIQE/CLIP QoE metric and a perceived-coverage notion, and develop importance-aware waterfilling power allocation for uncoded and coded transmission. Simulations on a single representative image claim up to 14 dB robustness improvement over JPEG uncoded transmission and 4.5 dB over JPEG-LDPC at a compression rate of 9%, together with coverage gains of 7.5–8 dB at rates of 33–50%.

Significance. If the reported gains hold across content classes and with a validated QoE threshold, the paper would make a significant conceptual contribution: it demonstrates that generative inference at the receiver can substitute for bit-level reliability, enabling very simple transmitters and extending perceived coverage. The paper also provides a useful formalization of perceived coverage and a concrete power-allocation algorithm. However, the quantitative evidence is currently limited to one test image and an unvalidated QoE threshold, so the significance of the headline numbers is conditional on further substantiation.

major comments (4)
  1. [Sec. VI (first paragraph) and Sec. II-A] The simulation section states that results are obtained by averaging over 100 independent realizations with a single representative image used, even though the text earlier says three RGB images with diverse visual characteristics were selected. The central claim that high-frequency content is semantically expendable (Sec. II-A) is content-class dependent; the 14 dB, 4.5 dB, and coverage gains rest entirely on one image. Please report per-image results and statistics across multiple images, or explicitly restrict the claims to the demonstrated content class.
  2. [Sec. IV-A and Def. 3] The QoE requirement QoEth = [DNIQE, DCLIP] = [5, 0.1] is used to define perceived coverage and to compute every dB gain in Sec. VI. No evidence is provided that this threshold corresponds to user-perceived QoE or that NIQE/CLIP correlate with subjective quality for images reconstructed by a generative model. Please add a sensitivity analysis over QoEth and a validation against subjective or task-based QoE; without this, the quantitative headline is conditional on an arbitrary operating point.
  3. [Sec. VI-C and Abstract] The abstract claims 'a 9 dB gain in perceived coverage,' but Sec. VI-C reports an overall coverage gain of G = 0.73 dB relative to the conventional limit and gains of approximately 7.5–8 dB for rates of 33–50%. The 9 dB figure appears to be only the power-domain gain at the SNR-limit point (from -8.5 dB vs. 0.5 dB), not the coverage gain defined in Eq. (23). This numerical inconsistency should be corrected, and the abstract should specify which gain is being reported.
  4. [Sec. III-A, Definitions 1 and 2] The sufficiency and error-resilience analysis is qualitative. Definition 1 requires that any distortion of s* makes QoE unacceptable, but the existence of such an s* for the LPF representation is not established. Definition 2 introduces IMSEmax without quantifying it. The statement in Sec. VI-A that Figs. 6 and 7 'confirm the analysis' refers to single-image simulation curves, not a derivation. Please either provide a formal sufficiency/error-resilience characterization or explicitly reposition these definitions as qualitative design principles.
minor comments (4)
  1. [Sec. VI-A, Sec. III-B2, Sec. VI-C] There are several typographical errors: 'analysis pf' should be 'analysis of', 'Tubor' should be 'Turbo', and 'high lighting' should be 'highlighting'.
  2. [Eq. (10)] The displayed formula for IMSE is missing the summation over k in the final approximation; the last term should read approximately the sum over k of gamma_k times the per-stream BER contribution.
  3. [General] The paper does not state whether code or data will be released. Given the single-image evaluation and the use of fitted BER parameters alpha_c and beta_c from prior work [39,41], releasing evaluation scripts and fitted parameter values would improve reproducibility.
  4. [Sec. VI-C] The phrase 'corresponding to lossless PNG compression' for rates 33–50% is unclear, since the LPF source encoder is lossy at these rates; please clarify the intended meaning.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: LightCom's headline robustness and coverage gains are empirical comparisons, not derivations that assume their own conclusions.

full rationale

The central claims—14 dB robustness gain at r=9% and 7.5–8 dB perceived-coverage extension—are obtained by measuring QoE curves (NIQE/CLIP) for LightCom and JPEG/LDPC baselines under a fixed threshold QoEth=[5,0.1], then reading the SNR differences where thresholds are crossed. These are benchmark comparisons against external baselines, not predictions derived from a model that already encodes the outcome. The importance-aware waterfilling power allocation is derived from standard BER expressions and an explicit IMSE surrogate; the paper then validates its QoE effect empirically with the SUPIR generative receiver. Citations to the authors' prior work supply BER fit parameters (αc, βc) and the channel-coded waterfilling solution, but these do not force the QoE gains—the gains are measured with an external GenAI model and compared with traditional systems, and the paper itself acknowledges MSE minimization is only a surrogate. The paper's use of a single representative image and its chosen QoE threshold are validity limitations, not circularity: they condition the quantitative claims but do not make the evaluation equal to its inputs by construction.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The framework's quantitative gains are downstream of a few hand-chosen quantities: the bit-importance weights, the BER fitting constants, and the QoE thresholds. The sufficiency argument also relies on a domain assumption about high-frequency image content that is not independently validated.

free parameters (3)
  • BER fitting parameters alpha_c, beta_c = Not reported
    Used in Eq. (9) to model coded BER; obtained by data fitting per [39,41]. The values are not given in the paper, so the coded transmission results cannot be independently reproduced.
  • QoE thresholds QoEth = [D_NIQE, D_CLIP] = [5, 0.1] = 5 and 0.1
    Chosen in Sec. VI without user-study calibration. Perceived coverage and the 14 dB and 8 dB gains are all defined relative to these thresholds.
  • Importance weights gamma_k = 2^(2(k-1)) = gamma_k = 2^(2(k-1)), k=1..8
    Adopted from prior work [39] to weight bit-level importance in the IMSE objective; drives the waterfilling power split and thus the WF-versus-EP gains.
assumptions (5)
  • domain assumption High-frequency image details contribute less to semantic understanding and can be discarded at the transmitter.
    Sec. II.A: 'fine-grained details... contribute less to semantic understanding but perceptual quality' and can be 'regenerated at the receiver'. This is load-bearing but not proven.
  • domain assumption Residual errors from weak channel codes are Gaussian and unstructured, while strong channel code errors are bursty and correlated.
    Sec. III.B.2 models weak-code errors as N(0, sigma_w^2 I) and strong-code errors as Burst(sigma_s^2, L_s). This distinction is asserted, not empirically validated.
  • domain assumption A pre-trained generative model (SUPIR) can infer high-fidelity content from heavily degraded low-pass-filtered inputs.
    Sec. II.B and VI rely on the published SUPIR model [30]; the paper does not fine-tune or adapt it to the communication channel.
  • standard math Standard M-QAM bit error rate formula in AWGN, ber = alpha Q(beta sqrt(snr)).
    Eq. (8) from Goldsmith [40]; used as the basis for the uncoded power allocation and IMSE analysis.
  • standard math The BER functions are convex in allocated power, making the waterfilling problem convex.
    Sec. V states convexity and invokes the Lagrange multiplier technique from Boyd and Vandenberghe [45].

how reviews work

0 comments
Cite this review

Pith. "Pith review of LightCom: A Generative AI-Augmented Framework for QoE-Oriented Communications." pith.science (2026). https://pith.science/paper/AZU2FLZX

@misc{pith2026250717352,
  author       = {Pith},
  title        = {Pith review of: LightCom: A Generative AI-Augmented Framework for QoE-Oriented Communications},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AZU2FLZX}},
  note         = {Machine review of arXiv:2507.17352}
}
abstract

Data-intensive and immersive applications, such as virtual reality, impose stringent quality of experience (QoE) requirements that challenge traditional quality of service (QoS)-driven communication systems. This paper presents LightCom, a lightweight encoding and generative AI (GenAI)-augmented decoding framework, designed for QoE-oriented communications under low signal-to-noise ratio (SNR) conditions. LightCom simplifies transmitter design by applying basic low-pass filtering for source coding and minimal channel coding, significantly reducing processing complexity and energy consumption. At the receiver, GenAI models reconstruct high-fidelity content from highly compressed and degraded signals by leveraging generative priors to infer semantic and structural information beyond traditional decoding capabilities. The key design principles are analyzed, along with the sufficiency and error-resilience of the source representation. We also develop importance-aware power allocation strategies to enhance QoE and extend perceived coverage. Simulation results demonstrate that LightCom achieves up to a $14$ dB improvement in robustness and a $9$ dB gain in perceived coverage, outperforming traditional QoS-driven systems relying on sophisticated source and channel coding. This paradigm shift moves communication systems towards human-centric QoE metrics rather than bit-level fidelity, paving the way for more efficient and resilient wireless networks.

Figures

Figures reproduced from arXiv: 2507.17352 by the authors.

Figure 1
Figure 1. Paradigm shift from QoS-oriented to QoE-oriented communications driven by QoE demands and generative capabilities of GenAI models. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Schematic of LightCom transmitter and receiver architecture. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Diagram of sufficient and insufficient representations, where orange [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: Perceived coverage and conventional converge under fixed trans [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Visual results of the original source, compressed representation obtained by [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: The QoE performance versus compression rates under error-free [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: The QoE performance versus the normalized MSE at a compression [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 8
Figure 8. Figure 8: Visual results of the reconstructed sources across the proposed LightCom framework and traditional communication systems, where [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]
Figure 9
Figure 9. Figure 9: QoE performance comparison under different compression rates. [PITH_FULL_IMAGE:figures/full_fig_p011_9.png]
Figure 10
Figure 10. Figure 10: The normalized IMSE comparison under different compression rates: (a). Compression rate of [PITH_FULL_IMAGE:figures/full_fig_p011_10.png]
Figure 11
Figure 11. Figure 11: Perceived coverage of LightCom and conventional communication [PITH_FULL_IMAGE:figures/full_fig_p012_11.png]
Figure 12
Figure 12. Figure 12: Visualization of the reconstructed images at low SNRs using the [PITH_FULL_IMAGE:figures/full_fig_p012_12.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Inference-Driven Uplink for 6G: Architecture, Principles, and Challenges

    eess.SP 2025-08 unverdicted novelty 4.0 of 10

    InferCom is an inference-driven 6G uplink architecture: lightweight transmitters compress, generative receivers reconstruct, and task-agnostic compression plus QoE-aware retransmission outperform conventional 5G NR an...

Reference graph

Works this paper leans on

46 extracted references · 41 canonical work pages · cited by 1 Pith paper

  1. [1]

    Toward 6G networks: Use cases and technologies,

    M. Giordani, M. Polese, M. Mezzavilla, S. Rangan, and M. Zorzi, “Toward 6G networks: Use cases and technologies,” IEEE Commun. Mag., vol. 58, no. 3, pp. 55–61, 2020

    work page 2020

  2. [2]

    On the road to 6G: Visions, requirements, key technologies, and testbeds,

    C.-X. Wang, X. You, X. Gao, X. Zhu, Z. Li, C. Zhang, H. Wang, Y . Huang, Y . Chen, H. Haas et al. , “On the road to 6G: Visions, requirements, key technologies, and testbeds,” IEEE Commun. Surv. Tutor., vol. 25, no. 2, pp. 905–974, 2023

    work page 2023

  3. [3]

    Tech- niques for measuring quality of experience,

    F. Kuipers, R. Kooij, D. De Vleeschauwer, and K. Brunnström, “Tech- niques for measuring quality of experience,” in Wired/Wireless Internet Communications. Berlin, Heidelberg: Springer Berlin Heidelberg, 2010, pp. 216–227

    work page 2010

  4. [4]

    Towards a QoE model to evaluate holographic augmented reality devices,

    L. Zhang, H. Dong, and A. El Saddik, “Towards a QoE model to evaluate holographic augmented reality devices,” IEEE MultiMedia , vol. 26, no. 2, pp. 21–32, 2018

    work page 2018

  5. [5]

    QoE-driven data communication framework for consumer electronics in tele-healthcare system,

    T. Zhang, X. Zhou, J. Liu, B. Cheng, X. Xu, L. Qi, Q. Tian, and Z. Wan, “QoE-driven data communication framework for consumer electronics in tele-healthcare system,” IEEE Trans. Consumer Electronics , vol. 69, no. 4, pp. 719–733, 2023

    work page 2023

  6. [6]

    QoE in video transmission: A user experience-driven strategy,

    T. Zhao, Q. Liu, and C. W. Chen, “QoE in video transmission: A user experience-driven strategy,” IEEE Commun. Surv. Tutor., vol. 19, no. 1, pp. 285–302, 2016

    work page 2016

  7. [7]

    QoE-oriented mobile virtual reality game in distributed edge networks,

    Y . Zhang, L. Pu, T. Lin, and J. Yan, “QoE-oriented mobile virtual reality game in distributed edge networks,” IEEE Trans. Multimedia , vol. 25, pp. 9132–9146, 2023

    work page 2023

  8. [8]

    From semantic communication to 13 semantic-aware networking: Model, architecture, and open problems,

    G. Shi, Y . Xiao, Y . Li, and X. Xie, “From semantic communication to 13 semantic-aware networking: Model, architecture, and open problems,” IEEE Commun. Mag. , vol. 59, no. 8, pp. 44–50, 2021

    work page 2021

Show all 46 references
  1. [9]

    Semantic communications for future internet: Fundamentals, applications, and challenges,

    W. Yang, H. Du, Z. Q. Liew, W. Y . B. Lim, Z. Xiong, D. Niyato, X. Chi, X. Shen, and C. Miao, “Semantic communications for future internet: Fundamentals, applications, and challenges,” IEEE Commun. Sur. Tutor., vol. 25, no. 1, pp. 213–250, 2022

    2022

  2. [10]

    The jpeg still picture compression standard,

    G. K. Wallace, “The jpeg still picture compression standard,” IEEE Trans. Consumer Electronics, vol. 38, no. 1, pp. xviii–xxxiv, 1992

    1992

  3. [11]

    Overview of the h. 264/avc video coding standard,

    T. Wiegand, G. J. Sullivan, G. Bjontegaard, and A. Luthra, “Overview of the h. 264/avc video coding standard,” IEEE Trans. Circuits Syst. Video Technol., vol. 13, no. 7, pp. 560–576, 2003

    2003

  4. [12]

    Error detecting and error correcting codes,

    R. W. Hamming, “Error detecting and error correcting codes,” The Bell System Tech. J., vol. 29, no. 2, pp. 147–160, 1950

    1950

  5. [13]

    Convolutional codes and their performance in communica- tion systems,

    A. Viterbi, “Convolutional codes and their performance in communica- tion systems,” IEEE Trans. Commun. Tech., vol. 19, no. 5, pp. 751–772, 1971

    1971

  6. [14]

    Near shannon limit error- correcting coding and decoding: Turbo-codes. 1,

    C. Berrou, A. Glavieux, and P. Thitimajshima, “Near shannon limit error- correcting coding and decoding: Turbo-codes. 1,” in IEEE Int. Conf. Commun., vol. 2. IEEE, 1993, pp. 1064–1070

    1993

  7. [15]

    Design of capacity-approaching irregular low-density parity-check codes,

    T. J. Richardson, M. A. Shokrollahi, and R. L. Urbanke, “Design of capacity-approaching irregular low-density parity-check codes,” IEEE Trans. Inf. theory, vol. 47, no. 2, pp. 619–637, 2001

    2001

  8. [16]

    Rescue: Links- on-the-fly technology for robust, efficient and smart communication in unpredictable environments,

    K. Anwar, R. Datta, A. Festag, G. Fettweis, G. Del Galdo, S. Gurgul, M. Juntti, H. Khalifé, P. Komulainen, Y . Ma et al. , “Rescue: Links- on-the-fly technology for robust, efficient and smart communication in unpredictable environments,” in Proc. Eur. Conf. Net. Commun. , 2014

    2014

  9. [17]

    Link budget and radio coverage design for various multipath urban communication links,

    N. Yarkoni, N. Blaunstein, and D. Katz, “Link budget and radio coverage design for various multipath urban communication links,”Radio Science, vol. 42, no. 02, pp. 1–15, 2007

    2007

  10. [18]

    Understanding link budget in wireless com- munication,

    Simpulse, “Understanding link budget in wireless com- munication,” 2025. [Online]. Available: https://www.simpulse- sdr.com/articles/understanding-link-budget-in-wireless-communication

    2025

  11. [19]

    Wireless communication technologies for iot in 5g: Vision, applications, and challenges,

    Q. V . Khanh, N. V . Hoai, L. D. Manh, A. N. Le, and G. Jeon, “Wireless communication technologies for iot in 5g: Vision, applications, and challenges,” Wireless Communi. Mobile Computing , vol. 2022, no. 1, p. 3229294, 2022

    2022

  12. [20]

    An overview of emergency communication networks,

    Q. Wang, W. Li, Z. Yu, Q. Abbasi, M. Imran, S. Ansari, Y . Sambo, L. Wu, Q. Li, and T. Zhu, “An overview of emergency communication networks,” Remote Sensing, vol. 15, no. 6, p. 1595, 2023

    2023

  13. [21]

    Beyond transmitting bits: Context, seman- tics, and task-oriented communications,

    D. Gündüz, Z. Qin, I. E. Aguerri, H. S. Dhillon, Z. Yang, A. Yener, K. K. Wong, and C.-B. Chae, “Beyond transmitting bits: Context, seman- tics, and task-oriented communications,” IEEE J. Sel. Areas Commun. , vol. 41, no. 1, pp. 5–41, 2022

    2022

  14. [22]

    Gen- erative semantic communications with foundation models: Perception- error analysis and semantic-aware power allocation,

    C. Xu, M. B. Mashhadi, Y . Ma, R. Tafazolli, and J. Wang, “Gen- erative semantic communications with foundation models: Perception- error analysis and semantic-aware power allocation,” IEEE J. S. Areas Commun., 2025

    2025

  15. [23]

    Resitok: A resilient tokenization- enabled framework for ultra-low-rate and robust image transmission,

    Z. Liu, Y . Ma, and R. Tafazolli, “Resitok: A resilient tokenization- enabled framework for ultra-low-rate and robust image transmission,” arXiv preprint arXiv:2505.01870 , 2025

    2025 arXiv

  16. [24]

    Big AI models for 6G wireless networks: Opportunities, challenges, and research directions,

    Z. Chen, Z. Zhang, and Z. Yang, “Big AI models for 6G wireless networks: Opportunities, challenges, and research directions,” IEEE Wireless Commun., 2024

    2024

  17. [25]

    Large generative AI models for telecom: The next big thing?

    L. Bariah, Q. Zhao, H. Zou, Y . Tian, F. Bader, and M. Debbah, “Large generative AI models for telecom: The next big thing?” IEEE Commun. Mag., vol. 62, no. 11, pp. 84–90, 2024

    2024

  18. [26]

    Learning transferable visual models from natural language supervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark et al., “Learning transferable visual models from natural language supervision,” in Proc. Int. conf. machine learning (ICML) , Virtual, Jul. 2021, pp. 8748–8763

    2021

  19. [27]

    Language mod- els are few-shot learners,

    T. Brown, B. Mann, N. Ryder, M. Subbiah, J. D. Kaplan, P. Dhariwal, A. Neelakantan, P. Shyam, G. Sastry, A. Askell et al., “Language mod- els are few-shot learners,” Advances in neural information processing systems, vol. 33, pp. 1877–1901, 2020

    1901

  20. [28]

    High-resolution image synthesis with latent diffusion models,

    R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer, “High-resolution image synthesis with latent diffusion models,” in Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. , New Orleans, Louisiana, Jun. 2022, pp. 10 684–10 695

    2022

  21. [29]

    Adding conditional control to text-to-image diffusion models,

    L. Zhang, A. Rao, and M. Agrawala, “Adding conditional control to text-to-image diffusion models,” inProc. IEEE/CVF Int. Conf. Computer Vision (ICCV), Paris, France, Oct. 2023, pp. 3836–3847

    2023

  22. [30]

    Scaling up to excellence: Practicing model scaling for photo- realistic image restoration in the wild,

    F. Yu, J. Gu, Z. Li, J. Hu, X. Kong, X. Wang, J. He, Y . Qiao, and C. Dong, “Scaling up to excellence: Practicing model scaling for photo- realistic image restoration in the wild,” 2024

    2024

  23. [31]

    Deep joint source- channel coding for wireless image transmission,

    E. Bourtsoulatze, D. B. Kurka, and D. Gündüz, “Deep joint source- channel coding for wireless image transmission,” IEEE Trans. Cognitive Commun. Networking, vol. 5, no. 3, pp. 567–579, 2019

    2019

  24. [32]

    Semantic communication systems for speech transmission,

    Z. Weng and Z. Qin, “Semantic communication systems for speech transmission,” IEEE J. Sel. Areas Commun. , vol. 39, no. 8, pp. 2434– 2444, 2021

    2021

  25. [33]

    Generative joint source-channel coding for semantic image transmission,

    E. Erdemir, T.-Y . Tung, P. L. Dragotti, and D. Gündüz, “Generative joint source-channel coding for semantic image transmission,” IEEE J. Sel. Areas Commun., 2023

    2023

  26. [34]

    Multimodal se- mantic communication for generative audio-driven video conferencing,

    H. Tong, H. Li, H. Du, Z. Yang, C. Yin, and D. Niyato, “Multimodal se- mantic communication for generative audio-driven video conferencing,” IEEE Wireless Commun. Lett. , 2024

    2024

  27. [35]

    Semantic-aware power allocation for generative semantic communications with foundation models,

    C. Xu, M. B. Mashhadi, Y . Ma, and R. Tafazolli, “Semantic-aware power allocation for generative semantic communications with foundation models,” 2024. [Online]. Available: https://arxiv.org/abs/2407.03050

    2024 arXiv

  28. [36]

    Latency-aware generative semantic communications with pre-trained diffusion models,

    L. Qiao, M. B. Mashhadi, Z. Gao, C. H. Foh, P. Xiao, and M. Bennis, “Latency-aware generative semantic communications with pre-trained diffusion models,” IEEE Wireless Commun. Lett. , 2024

    2024

  29. [37]

    Separate source channel coding is still what you need: An llm-based rethinking,

    T. Ren, R. Li, M.-m. Zhao, X. Chen, G. Liu, Y . Yang, Z. Zhao, and H. Zhang, “Separate source channel coding is still what you need: An llm-based rethinking,” arXiv preprint arXiv:2501.04285 , 2025

    2025 arXiv

  30. [38]

    Auto-split: A general framework of collaborative edge-cloud AI,

    A. Banitalebi-Dehkordi, N. Vedula, J. Pei, F. Xia, L. Wang, and Y . Zhang, “Auto-split: A general framework of collaborative edge-cloud AI,” in Proc. ACM SIGKDD Int. Conf. Knowl. Discov. Data Min., 2021, pp. 2543–2553

    2021

  31. [39]

    Data-importance-aware power allocation for adaptive real-time communication in computer vision applications,

    C. Xu, Y . Ma, R. Tafazolli, and J. Wang, “Data-importance-aware power allocation for adaptive real-time communication in computer vision applications,” IEEE J. S. Areas Commun. (Accepted with Minor Revision), 2025. [Online]. Available: https://arxiv.org/abs/2504.08922

    2025 arXiv

  32. [40]

    Goldsmith, Wireless Communications

    A. Goldsmith, Wireless Communications. Cambridge University Press, 2005

    2005

  33. [41]

    Data-importance-aware waterfilling for adaptive real-time communication in computer vision applications,

    C. Xu, Y . Ma, and R. Tafazolli, “Data-importance-aware waterfilling for adaptive real-time communication in computer vision applications,”

  34. [42]

    Diffbir: Toward blind image restoration with generative diffusion prior,

    X. Lin, J. He, Z. Chen, Z. Lyu, B. Dai, F. Yu, Y . Qiao, W. Ouyang, and C. Dong, “Diffbir: Toward blind image restoration with generative diffusion prior,” in Proc. Eur. Conf. Comput. Vis. Springer, 2024, pp. 430–448

    2024

  35. [43]

    Making a “completely blind

    A. Mittal, R. Soundararajan, and A. C. Bovik, “Making a “completely blind” image quality analyzer,”IEEE Signal Process. Lett., vol. 20, no. 3, pp. 209–212, 2012

    2012

  36. [44]

    Recent contributions to the mathematical theory of com- munication,

    W. Weaver, “Recent contributions to the mathematical theory of com- munication,” ETC: a review of general semantics , pp. 261–281, 1953

    1953

  37. [45]

    Boyd and L

    S. Boyd and L. Vandenberghe, Convex Optimization . Cambridge University Press, 2004

    2004

  38. [2025]

    Available: https://arxiv.org/abs/2502.20926

    [Online]. Available: https://arxiv.org/abs/2502.20926

    arXiv

Pith tools

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