Generalized User-Oriented Image Semantic Coding Empowered by Large Vision-Language Model

Sin-Yu Huang; Vincent W.S. Wong

REVIEW 3 major objections 5 minor 15 references

Generalized User-Oriented Image Semantic Coding Empowered by Large Vision-Language Model

T0 review · 3 major / 5 minor · reviewed 2026-08-04 · deepseek-v4-flash

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

Pith's one-line read User-intent image coding generalizes to unseen objects, lifting answer match by 34%

desk verdict A plausible query-aware image semantic coding system whose evaluation is circular (LLaVA defines both the loss and the metric); the architecture is worth a look, but the numbers don't yet support the claims. read the letter →

arxiv 2509.08913 v1 pith:JIIA3CSP submitted 2025-09-10 eess.IV

Sin-Yu Huang , Vincent W.S. Wong This is my paper

classification eess.IV

keywords semanticcommunicationuserintentimagecodingvision-languagemodelCLIPLLaVAzero-shotgeneralizationvectorquantization

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

The paper proposes a user-oriented image semantic coding (UO-ISC) framework that transmits only the features relevant to a user's text query, rather than the whole image. By anchoring the encoder with a pretrained vision-language model (CLIP) and training with a query-relevance loss from another vision-language model (LLaVA), the system learns to prioritize intent-relevant content and generalize to object categories never seen in training. On zero-shot tests with animal images excluded during training, the framework improves the exact-answer match rate by 34% over a non-pretrained variant, adds 5% through text alignment, and beats a query-aware semantic-coding baseline by 4.8% at 20 dB SNR. The paper's central claim is that pretrained vision-language backbones make user-intent semantic coding both effective and generalizable.

What carries the argument

The central mechanism is a segment-wise vector quantization pipeline applied to CLIP features that are text-conditioned through Feature-wise Linear Modulation (FiLM). The text query and image are projected into a shared embedding space; the text features generate scale and shift matrices that modulate the image features before quantization into discrete codewords. A user-intent relevance loss, computed as the cosine distance between LLaVA hidden states of the original and reconstructed images under the same query, is added to ℓ1, quantization, and adversarial losses in a two-phase training procedure. This machinery lets the transmitter send a compact sequence of integer indices that preserve

What would settle it

Run a human study in which raters are shown original and reconstructed images with the same query and asked whether the reconstruction preserves the queried content, then compare human agreement rates with the LLaVA-based answer match rates; if human gains are substantially smaller than the 34% and 5% improvements, the central claim of user-intent preservation is not supported.

Watch

Extended reading notes

Core claim

The paper discovers that integrating a frozen, pretrained VLM (CLIP) as the image/text feature extractor makes user-intent-driven semantic image coding generalize to unseen object categories, while a separately defined user-intent relevance loss (computed by LLaVA) aligns the reconstruction with the user's query. In zero-shot evaluations on animal images absent from the VQA training set, the proposed UO-ISC framework achieves the lowest user-intent relevance loss and the highest answer match rate among four baselines, including a state-of-the-art LLM-based query-aware codec. The improvement is attributed to the pretrained backbone's transferable semantic representations and to the FiLM-based

Load-bearing premise

The load-bearing assumption is that LLaVA's hidden-state cosine similarity and its exact-match answers are faithful, self-consistent proxies for what a human user actually intends; if LLaVA's judgments diverge from human intent, the reported gains overestimate true intent preservation.

Editorial extensions

If this is right

If the claim holds, wireless image transmission can become intent-adaptive: the same source image can be encoded differently for different user queries without retraining the backbone.
The zero-shot results suggest that semantic coding systems can be deployed on out-of-distribution imagery by leveraging pretrained VLMs, reducing the need for task-specific training data.
The framework transmits 82% fewer symbols per image than JPEG2000 at comparable segment settings, indicating potential bandwidth savings for semantic-aware communication.
The two-phase training recipe (semantic reconstruction first, adversarial refinement second) may be reusable for other discrete-feature generative transmission tasks.

Reading between the lines

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

Since the same LLaVA model is used for both the training loss and the evaluation metric, the reported answer-match gains may partly reflect the model learning to satisfy LLaVA's specific hidden-state geometry rather than genuine semantic preservation; a human-evaluation study or a different VLM as judge would clarify this.
The approach could be extended to multi-query or dynamically changing user intent, where the transmitter would need to prioritize content relevant to multiple simultaneous queries or update the transmitted features as intent shifts.
The segment-wise codebook design creates an explicit rate-quality trade-off (shorter segments improve relevance but increase symbol count), suggesting a rate-intent-relevance optimization problem that the paper leaves open.
If the user-intent relevance loss is robust across different VLM backbones, the methodology could also apply to video or 3D scene transmission, though the paper only demonstrates still images.

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.

Desk Editor's Note

read the letter

Quick take: this is a competent engineering paper with a coherent architecture, but its headline results are not trustworthy because the evaluation metric is defined by the same model used in the training loss. You should read it for the system design, not for the numbers.

What's new and good: the framework combines frozen CLIP features, FiLM-based text conditioning, segment-wise vector quantization, a U-Net-like decoder, and a LLaVA-based user-intent loss. It's a legitimate extension of query-aware semantic coding, and the zero-shot evaluation on unseen animal categories is a sensible step toward generalization. The two-phase training with adversarial loss is standard but properly implemented. The math is clean and internally consistent. The efficiency comparison (82% fewer symbols than JPEG2000) is useful, if not surprising from a semantic coder.

The soft spot is load-bearing. The user-intent loss in Eq. (11) minimizes the cosine distance between nanoLLaVA hidden states of the original and reconstructed images given the same query. The answer match rate in Section IV is then computed by asking the same nanoLLaVA to answer the query for both images and checking exact match. The model is literally trained to make LLaVA's internal representations closer, so a higher match rate on LLaVA-generated answers is expected even if no human would agree. Without an independent judge—human raters, a different VQA model, or at least a behavioral task—the +5% and +34% gains do not show that user intent is preserved. The paper does not acknowledge this circularity at all.

The ablation is also confounded: the "without text alignment and pretrained VLM" model differs from UO-ISC in both text alignment and CLIP initialization, so the 34% cannot be attributed to the VLM backbone alone. There are no error bars or multiple seeds, and the zero-shot test is a single within-dataset category split. These are fixable, but they need to be fixed before the claims can be accepted.

All that said, this is not a desk-reject. The idea of using frozen VLMs for generalized query-aware coding is worth exploring, and the system is a reasonable baseline for future work. The right audience is the semantic communication community, and the paper deserves a serious referee who can push for a proper evaluation.

My recommendation: send it to peer review, but expect a major revision. Insist on an independent metric, disentangled ablations, seeds, and code/data release. If those are addressed, the paper could become a solid contribution.

Referee Report

3 major / 5 minor

Summary. The paper proposes UO-ISC, a user-oriented image semantic coding framework. The transmitter extracts CLIP image features and aligns them with a user-provided text query via FiLM-conditioned projections, then quantizes the aligned features segment-wise and transmits discrete indices over a fading channel. The receiver reconstructs the image through a decoder with skip connections. Training uses an L1 loss, a quantization loss, a user-intent relevance loss (Eq. 11) defined as the cosine distance between LLaVA hidden states of the original and reconstructed images given the same query, and, in a second phase, an adversarial loss. The framework is evaluated on a zero-shot animal subset of VQA, reporting answer match rate and user-intent relevance loss versus SNR, plus ablations and a symbol-efficiency comparison with JPEG2000.

Significance. If the results hold, the framework would be a practical way to inject user intent into image semantic coding while leveraging pretrained vision-language models for zero-shot generalization. The architecture is plausible and the use of frozen CLIP with learnable projections and FiLM conditioning is well motivated. However, the evaluation has a load-bearing self-reference issue: the training objective (Eq. 11) and the main metric (LLaVA answer match rate) both rely on the same nanoLLaVA model. Without an independent human or cross-model validation of user-intent preservation, the reported gains may reflect optimization toward LLaVA's internal geometry rather than genuine semantic alignment. The single split and absence of error bars further weaken the empirical support. The contribution is potentially useful, but the evidence presented is not yet sufficient to support the central claim.

major comments (3)

[Section IV and Eq. (11)] The evaluation metric is self-referential. The user-intent loss Luser in Eq. (11) minimizes the cosine distance between nanoLLaVA hidden states of the original and reconstructed images, while the answer match rate is computed by asking the same nanoLLaVA (or its LLaVA variant) to generate answers from both images. Since the generated answers depend directly on those hidden states, optimizing Eq. (11) can inflate the answer match rate without actually preserving human-judged intent. The paper does not validate nanoLLaVA against human annotations or a different VLM, so the central claim of 'user-intent preservation' is not established. Please add an independent evaluation, e.g., human raters or a different pretrained VLM, and report agreement between nanoLLaVA and human judgments.
[Section IV, ablation and statistics] The experimental evidence rests on a single zero-shot split (train on non-animal VQA samples, test on animal samples) with no error bars, no multiple seeds, and no statistical significance tests. The curves in Fig. 4 show differences of a few percent, but without variance information we cannot assess whether the reported 5%, 34%, and 4.8% improvements are meaningful. Moreover, the text 'improves the answer match rate by 5% through text alignment and by 34% by using a large VLM backbone' is ambiguous. If the 34% comes from comparing 'UO-ISC without text alignment' with 'UO-ISC without text alignment and pretrained VLM', the isolation is valid but should be stated explicitly; if it instead compares the full UO-ISC with the random-initialized baseline, the contribution of the pretrained VLM is conflated with text alignment and Luser. Please clarify and provide multiple runs with confidence
[Section IV, baseline selection] The paper claims to outperform the 'state-of-the-art query-aware image semantic coding' but compares against a single method, reference [7], which is an arXiv preprint not established as state of the art. The JPEG2000 baseline is not query-aware. To support the SOTA claim, please either expand the comparison to include other query-aware semantic coding methods or temper the claim to 'the compared baseline'.

minor comments (5)

[Eq. (6)] The segment indexing uses '(l−1)L' in the subscript, but the segment length is N_L, not L. It should be '(l−1)N_L + 1' etc.
[Eq. (11) and Eq. (13)] The argument order of Luser is inconsistent: Eq. (11) defines Luser(Y, X_I; x_T) while Eq. (13) calls Luser(X_I, Y; x_T). Please unify.
[Section IV] The footnote says 'nanoLLaV A' with an extra space due to a line break. Please use the correct model name consistently.
[Section IV, Fig. 6] The number of transmitted symbols per image is not fully defined. Please specify whether it includes LDPC coding overhead, and how the symbol count for JPEG2000 is computed.
[Section II, channel model] The paper assumes LDPC coding and modulation but does not discuss residual channel errors after decoding. In a fading channel with noise, the received symbol vector may contain errors that affect the decoded indices. Please clarify whether the evaluation accounts for residual errors or assumes error-free decoding after LDPC.

Circularity Check

2 steps flagged · score 6.0 of 10

Evaluation is partially circular: Luser is both a training loss and a reported metric, and the answer match rate is scored by the same LLaVA model used in Luser.

fitted input called prediction [Section III Eq. (13)-(14); Section IV Fig. 4(b)]
"Results in Figs. 4 (a) and (b) show the answer match rate and the user-intent relevance loss Luser versus the SNR."

Eq. (13) includes Luser as a term in the training loss, so the model is explicitly optimized to minimize it. Fig. 4(b) then reports Luser as a performance metric. Thus the 'lowest user-intent relevance loss' claimed for UO-ISC is by construction the value of the training objective, not an independent measure of user intent.
other [Section III Eq. (11); Section IV answer match rate definition]
"The answer match rate measures the proportion of samples where the answer generated by LLaVA from the reconstructed image exactly matches the one from the original image under the same user query."

The training loss Luser (Eq. 11) minimizes the cosine distance between LLaVA hidden states of the original and reconstructed images, and the evaluation metric asks the same LLaVA model to generate answers from those same hidden states. The answer is a deterministic function of these hidden states, so the metric is not independent of the training signal. Improvements in answer match rate may reflect alignment of LLaVA's internal geometry rather than preservation of human-perceived user intent. This is a partial reduction, not an exact equality, but the same model is load-bearing on both sides.

full rationale

The paper's architectural derivation and quantization steps are self-contained and do not rely on load-bearing self-citations. However, the evaluation of the central user-intent claim is substantially circular. First, the user-intent relevance loss Luser is a training objective in Eqs. (13)-(14) and is then reported as an evaluation metric in Fig. 4(b), making that result true by construction. Second, the primary metric, answer match rate, is computed using the same LLaVA model whose hidden states define Luser, creating a self-referential loop: the model is optimized against LLaVA's internal representations and then judged by LLaVA-generated answers. The comparison with the query-aware baseline retains some independent content, and the answer match rate is not identical to Luser, so the paper is not fully circular, but the reported gains (+5% text alignment, +34% VLM backbone) are likely inflated by this same-model evaluation. The 34% ablation is also confounded because 'without text alignment and pretrained VLM' differs in two factors at once, though that is an experimental-design issue rather than circularity. Overall, the central claim is not independently established with respect to human user intent.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claims rely on three domain assumptions: LLaVA answers are a valid measure of user intent, a within-dataset animal split is a valid zero-shot test, and frozen CLIP features retain enough information for reconstruction. No new physical entities are introduced. The loss weights and codebook/segment hyperparameters are free choices not derived from theory, and their values are mostly unreported.

free parameters (3)

Regularization weights lambda_user, lambda_quant, beta, lambda_gen = Not reported
Appear in Eqs. (12)-(14); chosen by hand and not given, so the reported performance depends on unreported balancing.
Segment length N_L, number of segments L, codebook size N_cw = N_L=32, L=16, N_cw=64
Hand-chosen in Section IV; Fig. 5 shows the loss changes with N_L, so the choice affects the reported rate/quality tradeoff.
Selected CLIP layers M_S = {3,6,9,11}
Hand-chosen transmission subset in Section IV; it sets the number of transmitted feature matrices and thus rate and reconstruction quality.

assumptions (4)

domain assumption LLaVA's hidden-state similarity and its generated answers are valid, consistent proxies for human user intent.
The user-intent relevance loss is computed as cosine distance between LLaVA embeddings, and evaluation is exact-match of LLaVA answers; if LLaVA is not aligned with user intent, the reported gains do not measure intent preservation.
domain assumption Held-out animal samples from VQA constitute a meaningful zero-shot 'unseen object' generalization test.
Section IV trains on VQA excluding animal-related samples and evaluates on animal-related samples from the same dataset; this is a single category split within one dataset, not a cross-domain test.
domain assumption Frozen CLIP features retain enough spatial and semantic information for high-quality image reconstruction after quantization.
The decoder reconstructs from quantized CLIP embeddings only, with no source image pixels transmitted, so this assumption is load-bearing.
domain assumption The simulated flat fading channel with h~CN(0,1), LDPC coding, and 8-ary modulation adequately represents wireless transmission.
All performance claims are measured in this simulation, so the channel and physical-layer choices are load-bearing for the quantitative results.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Generalized User-Oriented Image Semantic Coding Empowered by Large Vision-Language Model." pith.science (2026). https://pith.science/paper/JIIA3CSP

@misc{pith2026250908913,
  author       = {Pith},
  title        = {Pith review of: Generalized User-Oriented Image Semantic Coding Empowered by Large Vision-Language Model},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JIIA3CSP}},
  note         = {Machine review of arXiv:2509.08913}
}

read the original abstract

Semantic communication has shown outstanding performance in preserving the overall source information in wireless transmission. For semantically rich content such as images, human users are often interested in specific regions depending on their intent. Moreover, recent semantic coding models are mostly trained on specific datasets. However, real-world applications may involve images out of the distribution of training dataset, which makes generalization a crucial but largely unexplored problem. To incorporate user's intent into semantic coding, in this paper, we propose a generalized user-oriented image semantic coding (UO-ISC) framework, where the user provides a text query indicating its intent. The transmitter extracts features from the source image which are relevant to the user's query. The receiver reconstructs an image based on those features. To enhance the generalization ability, we integrate contrastive language image pre-training (CLIP) model, which is a pretrained large vision-language model (VLM), into our proposed UO-ISC framework. To evaluate the relevance between the reconstructed image and the user's query, we introduce the user-intent relevance loss, which is computed by using a pretrained large VLM, large language-and-vision assistant (LLaVA) model. When performing zero-shot inference on unseen objects, simulation results show that the proposed UO-ISC framework outperforms the state-of-the-art query-aware image semantic coding in terms of the answer match rate.

Figures

Figures reproduced from arXiv: 2509.08913 by the authors.

**Figure 2.** The segment-wise vector quantization process in UO-ISC. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Visualization of the reconstructed images produced by the proposed UO-ISC and baseline methods with the SNR = 0 dB. The text [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Figures from the paper (2 more)

**Figure 4.** Figure 4: (a) Answer match rate and (b) user-intent relevance loss versus [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]

**Figure 5.** Figure 5: The user-intent relevance loss versus the SNR under different [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

15 extracted references · 1 linked inside Pith

[7]

Semantic communication based on large language model for underwa- ter image transmission,

W. Chen, W. Xu, H. Chen, X. Zhang, Z. Qin, Y . Zhang, and Z. Han, “Semantic communication based on large language model for underwa- ter image transmission,”arXiv preprint arXiv:2408.12616, 2024

arXiv 2024
[1]

Semantic communication: A survey on research landscape, challenges, and future directions,

T. M. Getu, G. Kaddoum, and M. Bennis, “Semantic communication: A survey on research landscape, challenges, and future directions,”Proc. of the IEEE, vol. 112, no. 11, pp. 1649–1685, Nov. 2024

2024
[2]

Beyond transmitting bits: Context, semantics, and task-oriented communications,

D. G ¨und¨uz, Z. Qin, I. E. Aguerri, H. S. Dhillon, Z. Yang, A. Yener, K. K. Wong, and C.-B. Chae, “Beyond transmitting bits: Context, semantics, and task-oriented communications,”IEEE J. Sel. Areas Commun., vol. 41, no. 1, pp. 5–41, Jan. 2023

2023
[3]

SCAN: Semantic communication with adaptive channel feedback,

G. Zhang, Q. Hu, Y . Cai, and G. Yu, “SCAN: Semantic communication with adaptive channel feedback,”IEEE Trans. Cogn. Commun. Netw., vol. 10, no. 5, pp. 1759–1773, Oct. 2024

2024
[4]

Toward semantic communications: Deep learning-based image semantic coding,

D. Huang, F. Gao, X. Tao, Q. Du, and J. Lu, “Toward semantic communications: Deep learning-based image semantic coding,”IEEE J. Sel. Areas Commun., vol. 41, no. 1, pp. 55–71, Jan. 2023

2023
[5]

Visual language model based cross-modal semantic communication systems,

F. Jiang, C. Tang, L. Dong, K. Wang, K. Yang, and C. Pan, “Visual language model based cross-modal semantic communication systems,” IEEE Trans. Wireless Commun., vol. 24, no. 5, pp. 3937–3948, May 2025

2025
[6]

LaMoSC: Large language model-driven semantic communication system for visual trans- mission,

Y . Zhao, Y . Yue, S. Hou, B. Cheng, and Y . Huang, “LaMoSC: Large language model-driven semantic communication system for visual trans- mission,”IEEE Trans. Cogn. Commun. Netw., vol. 10, no. 6, pp. 2005– 2018, Dec. 2024

2005
[8]

Leveraging MoE-based large language model for zero-shot multi-task semantic communication,

S.-Y . Huang, R. Liao, and V . W. S. Wong, “Leveraging MoE-based large language model for zero-shot multi-task semantic communication,” in Proc. of IEEE Int. Conf. Commun. (ICC), Montreal, Canada, Jun. 2025

2025

Show all 15 references

[9]

Learning transferable visual models from natural language supervi- sion,

A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark, G. Krueger, and I. Sutskever, “Learning transferable visual models from natural language supervi- sion,” inProc. of Int’l Conf. on Machine Learning (ICML), Jul. 2021

2021
[10]

Generative adversarial nets,

I. J. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y . Bengio, “Generative adversarial nets,” in Proc. of Adv. Neural Inf. Process. Syst., Montreal, Canada, Dec. 2014

2014
[11]

VQA: Visual Question Answering,

S. Antol, A. Agrawal, J. Lu, M. Mitchell, D. Batra, C. L. Zitnick, and D. Parikh, “VQA: Visual Question Answering,” inProf. of Int’l Conf. Comput. Vis. (ICCV), Santiago, Chile, Dec. 2015

2015
[12]

U-Net: Convolutional net- works for biomedical image segmentation,

O. Ronneberger, P. Fischer, and T. Brox, “U-Net: Convolutional net- works for biomedical image segmentation,” inProc. of Int’l Conf. Med. Ima. Comput. Comput.-Assisted Intervention (MICCAI), Munich, Germany, Oct. 2015

2015
[13]

Feature-wise transformations,

V . Dumoulin, E. Perez, N. Schucher, F. Strub, H. de Vries, A. Courville, and Y . Bengio, “Feature-wise transformations,”Distill, Jul. 2018

2018
[14]

Visual instruction tuning,

H. Liu, C. Li, Q. Wu, and Y . J. Lee, “Visual instruction tuning,” inProc. of Advances in Neural Inf. Process. Syst. (NeurIPS), New Orleans, LA, Dec. 2023

2023
[15]

Information Technology - JPEG 2000 Image Coding System Part 1: Core Coding System,

ISO/IEC 15444-1, “Information Technology - JPEG 2000 Image Coding System Part 1: Core Coding System,” 5th edition, Nov. 2024

2000

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.

[1] [7]

Semantic communication based on large language model for underwa- ter image transmission,

W. Chen, W. Xu, H. Chen, X. Zhang, Z. Qin, Y . Zhang, and Z. Han, “Semantic communication based on large language model for underwa- ter image transmission,”arXiv preprint arXiv:2408.12616, 2024

arXiv 2024

[2] [1]

Semantic communication: A survey on research landscape, challenges, and future directions,

T. M. Getu, G. Kaddoum, and M. Bennis, “Semantic communication: A survey on research landscape, challenges, and future directions,”Proc. of the IEEE, vol. 112, no. 11, pp. 1649–1685, Nov. 2024

2024

[3] [2]

Beyond transmitting bits: Context, semantics, and task-oriented communications,

D. G ¨und¨uz, Z. Qin, I. E. Aguerri, H. S. Dhillon, Z. Yang, A. Yener, K. K. Wong, and C.-B. Chae, “Beyond transmitting bits: Context, semantics, and task-oriented communications,”IEEE J. Sel. Areas Commun., vol. 41, no. 1, pp. 5–41, Jan. 2023

2023

[4] [3]

SCAN: Semantic communication with adaptive channel feedback,

G. Zhang, Q. Hu, Y . Cai, and G. Yu, “SCAN: Semantic communication with adaptive channel feedback,”IEEE Trans. Cogn. Commun. Netw., vol. 10, no. 5, pp. 1759–1773, Oct. 2024

2024

[5] [4]

Toward semantic communications: Deep learning-based image semantic coding,

D. Huang, F. Gao, X. Tao, Q. Du, and J. Lu, “Toward semantic communications: Deep learning-based image semantic coding,”IEEE J. Sel. Areas Commun., vol. 41, no. 1, pp. 55–71, Jan. 2023

2023

[6] [5]

Visual language model based cross-modal semantic communication systems,

F. Jiang, C. Tang, L. Dong, K. Wang, K. Yang, and C. Pan, “Visual language model based cross-modal semantic communication systems,” IEEE Trans. Wireless Commun., vol. 24, no. 5, pp. 3937–3948, May 2025

2025

[7] [6]

LaMoSC: Large language model-driven semantic communication system for visual trans- mission,

Y . Zhao, Y . Yue, S. Hou, B. Cheng, and Y . Huang, “LaMoSC: Large language model-driven semantic communication system for visual trans- mission,”IEEE Trans. Cogn. Commun. Netw., vol. 10, no. 6, pp. 2005– 2018, Dec. 2024

2005

[8] [8]

Leveraging MoE-based large language model for zero-shot multi-task semantic communication,

S.-Y . Huang, R. Liao, and V . W. S. Wong, “Leveraging MoE-based large language model for zero-shot multi-task semantic communication,” in Proc. of IEEE Int. Conf. Commun. (ICC), Montreal, Canada, Jun. 2025

2025