Collaborative Knowledge Sharing-empowered Effective Semantic Rate Maximization for Two-tier Semantic-Bit Communication Networks

Fang Fang; Hong Chen; Xianbin Wang

REVIEW 3 major objections 5 minor 16 references

Collaborative Knowledge Sharing-empowered Effective Semantic Rate Maximization for Two-tier Semantic-Bit Communication Networks

T0 review · 3 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 The paper establishes that a two-tier edge network with collaborative knowledge sharing—uploading or downloading mismatched knowledge and using bit fallback—achieves the exact maximum of a generalized effective semantic transmission rate…

desk verdict The collaborative knowledge-sharing framework is a legitimate extension, but the simulations rest on a self-contradictory accuracy model, so the performance claims don't hold as written. read the letter →

arxiv 2501.01637 v1 pith:AII2UKEU submitted 2025-01-03 eess.SP

Hong Chen , Fang Fang , Xianbin Wang This is my paper · ORCID

classification eess.SP

keywords semanticcommunicationsknowledgebasemismatchsharingeffectivetransmissionrateresourceallocationbranchandboundKuhn-Munkresalgorithmfractionalprogramming

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 addresses knowledge base mismatch in task-oriented semantic communications, where transmitters and receivers must share the same knowledge to exchange semantics and misalignment degrades performance. It proposes collaborative sharing in a two-tier edge network: mismatched knowledge is uploaded from mobile devices or downloaded from the macro base station, with unshared knowledge transmitted as ordinary bits. To measure performance, it defines the generalized effective semantic transmission rate (GESTR), which combines semantic accuracy with the overhead of knowledge sharing. The core claim is that the resulting mixed-integer nonlinear program can be solved to global optimality by splitting it into per-association subproblems solved by fractional-programming branch-and-bound and an assignment subproblem solved by a modified Kuhn-Munkres algorithm. Simulations show this design outperforms baselines without collaboration or knowledge sharing, with the largest gains when signal-to-noise ratios are low.

What carries the argument

The driving object is the generalized effective semantic transmission rate (GESTR), a ratio objective that counts successfully delivered semantic information in suts per second, with knowledge-sharing time treated as overhead in the denominator. The solution mechanism has three parts: a fractional-programming transform (the parametric subtractive form of the ratio, updated iteratively) that turns each relaxed branch-and-bound subproblem into a linear program; the linearization of the product of binary transmission-mode and sharing-manner variables via an auxiliary variable $\tilde{b} = a b$; and branch-and-bound over the relaxed binaries whose bounds prune the search. Once these joint subproblems are solved, the remaining assignment of devices to subchannels is a bipartite matching on the computed GESTR weights, solved by a modified Kuhn-Munkres algorithm. The key identity that makes the decomposition work is that, for fixed association, the objective and constraints factor per device, so each device's joint subproblem can be solved independently before matching.

What would settle it

A brute-force enumeration of all feasible associations, transmission modes, sharing manners, and extraction ratios on a small network instance would settle the optimality claim: if any enumerated solution has total GESTR higher than the FP-BnB plus modified Kuhn-Munkres output, the claimed global optimum is false.

Watch

Extended reading notes

Core claim

The central claim is that a two-tier semantic-bit network's total generalized effective semantic transmission rate is maximized by jointly choosing, for every device, whether each mismatched knowledge class is shared via an MD-to-BS upload or an MBS-to-SBS download, how much semantic data is extracted (the extraction ratio), and which base station and subchannel serve the device. The paper shows that given an association, the remaining subproblem separates across devices and can be solved optimally by relaxing the binary transmission-mode and sharing-manner variables, applying fractional programming to the ratio objective, and branching over fractional solutions with branch-and-bound. The optimal subproblem values are then fed to a bipartite matching solved by a modified Kuhn-Munkres algorithm to assign base stations and subchannels. The result is a claimed global optimum for the original MINLP, and the paper demonstrates that the collaborative knowledge-sharing mechanism yields higher GESTR than no-collaboration and no-knowledge-sharing baselines, particularly in low-SNR conditions.

Load-bearing premise

The entire optimization and all numerical results rest on the closed-form semantic-accuracy surrogate $\epsilon'(\xi)$ in equation (15), whose four parameters are fitted by nonlinear least squares to data from reference [9] rather than derived from the actual neural semantic codecs used for the tasks; if that curve does not match the real accuracy-versus-extraction-ratio relation, the optimal decisions computed from it may not be optimal in practice.

Editorial extensions

If this is right

The joint decomposition yields the exact optimum, so it supplies a performance upper bound against which heuristic or learning-based resource allocators in semantic networks can be benchmarked.
The collaborative scheme shows that when device-to-SBS channels are weak, downloading mismatched knowledge from the MBS to the SBS is the better sharing manner, which transfers the burden from power-limited devices to the wired backhaul.
The GESTR metric, including knowledge-sharing overhead and semantic accuracy, offers a single objective that trades off semantic fidelity against the cost of alignment; this can serve as a design target for future hybrid semantic-bit systems.
Since bit communications act as a fallback for unshared knowledge, the formulation naturally covers mixed semantic and bit transmission without forcing every knowledge class to be shared, which matters for delay-constrained tasks.

Reading between the lines

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

Editorial inference: the FP-BnB-plus-matching pipeline is not specific to GESTR; any ratio objective that separates given an assignment, with binary variables entering linearly after a product linearization, can use the same decomposition, so the method likely transfers to other joint mode-selection and resource-allocation problems.
Editorial inference: the paper's optimality is relative to the fitted semantic-accuracy curve; a natural extension is to replace the closed-form surrogate with a lookup table or a differentiable neural accuracy model and keep the same optimization machinery, or to close the loop by fine-tuning knowledge-sharing decisions from measured accuracy feedback.
Editorial inference: the constraint that one device gets at most one subchannel and one subchannel is used by at most one device makes the allocation a simple matching; in multi-carrier or OFDMA settings with frequency reuse, the allocation becomes a harder generalized assignment, and the exactness of the Kuhn-Munkres step would be lost, though the FP-BnB subproblems would remain valid.
Editorial inference: the low-SNR advantage suggests collaborative sharing acts as a form of diversity; an experimental study varying the MBS-to-SBS backhaul capacity or the ratio of stored KB classes could quantify when downloading from the MBS ceases to help.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Colleague,

The interesting part of this paper is the collaborative knowledge-sharing mechanism: when an MD associates with an SBS, mismatched KB classes can be uploaded from the MD or downloaded from the MBS via backhaul, and unshared classes fall back to bit communications. That is a sensible extension of the KB-mismatch literature, and the GESTR metric that includes sharing overhead is a reasonable way to evaluate it. The problem formulation is detailed, and the decomposition into per-link joint subproblems plus a weighted matching allocation is structurally sound.

But there is a load-bearing flaw in the numerical section. Equation (15) defines semantic accuracy as ε = −θ1 e^{θ2(1−ξ)} + θ3 e^{−θ4(1−ξ)} with θ1, θ2, θ3, θ4 ≥ 0 and states ε is monotonically increasing in ξ. Yet Section IV adopts (θ1, θ2, θ3, θ4) = (−6.205e−8, 16.45, 0.9228, −0.06917), which violates θ1 ≥ 0 and θ4 ≥ 0. With those values ε(0) ≈ 1.85 and ε(1) ≈ 0.92, so accuracy exceeds 1 and decreases with ξ, the opposite of the stated premise. The threshold ξ^th derived from monotonicity is therefore meaningless, and the GESTR curves in Figures 3–4 are computed from an invalid model. This is not a subtle issue; it is an internal contradiction in the paper's own equations.

The rest of the paper has soft spots too. The transformation from (20) to (21) using the auxiliary variable ~b is plausible but the recovery of binary b* from ~b* is not clearly specified. The FP-BnB algorithm is claimed to find the global optimum, but no convergence or complexity analysis specific to this non-convex problem is provided. Those are secondary; the surrogate invalidates the main performance claims.

Who is this for? Researchers working on resource allocation in semantic communications. The collaborative sharing idea deserves attention, and the framework could be useful after the surrogate is fixed and simulations re-run with a valid accuracy model (or with real semantic codecs). I would send this to a serious referee, but only with the expectation of major revision. As it stands, the paper is not publishable without redoing the numerical results.

Recommendation: engage with the work, but insist on replacing or re-validating the semantic accuracy model and re-running the experiments.

Referee Report

3 major / 5 minor

Summary. The paper studies a two-tier semantic-bit communication network in which mobile devices (MDs) with mismatched knowledge bases (KBs) can share missing knowledge either by uploading it to the serving base station or, for small-cell associations, downloading it from the macro base station, with remaining data carried by bit communication. The authors define a generalized effective semantic transmission rate (GESTR) that includes semantic accuracy and knowledge-sharing overhead, formulate a mixed-integer nonlinear program (MINLP) to maximize total GESTR over knowledge-sharing decisions, extraction ratios, and BS/subchannel allocations, and propose a decomposition: per-association subproblems solved by a fractional-programming branch-and-bound (FP-BnB) method, followed by a modified Kuhn-Munkres (K-M) matching for allocation. Simulations compare the proposed solution with no-collaboration and no-knowledge-sharing baselines, reporting gains that are larger at low SNR and low bandwidth.

Significance. If the correctness issues were resolved, the paper would make a moderate contribution: the decomposition into independent per-association joint subproblems plus a weighted bipartite matching is natural and mostly well executed, and the explicit treatment of knowledge-sharing overhead in the semantic rate is a useful modeling step. The Dinkelbach-style fractional programming inside a branch-and-bound framework is standard but is applied here to a relevant problem, and the reduction of the final allocation to maximum-weight matching is clean. However, the paper does not provide machine-checked proofs or code, and the numerical validation as written rests on an internally inconsistent semantic-accuracy model, so the empirical claims are not currently established.

major comments (3)

[Section IV, Eq. (15)] The simulation parameters (θ1, θ2, θ3, θ4) = (−6.205e−8, 16.45, 0.9228, −0.06917) violate the non-negativity constraints θ1 ≥ 0 and θ4 ≥ 0 stated in Eq. (15). With these values, ε(0) ≈ 1.85 and ε(1) ≈ 0.92, so the semantic accuracy exceeds 1 and decreases with ξ, directly contradicting the monotonicity premise used in Section III-A to derive the lower bound ξ ≥ ξ^th and to define the search range [ξ^th, 1]. Consequently, constraint (16) is not enforced as written, and all GESTR values reported in Figs. 3 and 4 are computed from an invalid accuracy model. The authors must either supply fitted parameters that satisfy θ1, θ2, θ3, θ4 ≥ 0 and are consistent with a monotone increasing ε(ξ), or correct Eq. (15) to match the sign convention of the source [9], and then re-run all simulations.
[Section III-A and Section IV, Table I] The claimed global optimality of the joint solution is not established for the continuous variable ξ. The manuscript replaces ξ ∈ [ξ^th, 1] by M equal segments and searches only the discrete points ξ^(m), but the objective γ_{i,j,k} depends on ξ nonlinearly through ε(ξ), the semantic transmission time tS ∝ ξ, and the computing time tR ∝ ξ^{−ρ}; no monotonicity or convexity argument is given to guarantee that an optimum occurs at a grid point, and no error bound is provided for finite M. The value of M is also never reported in the simulation setup. The authors should either prove that the discretization is exact, provide a rigorous approximation guarantee, or revise the claim from 'joint optimum' to an approximate solution, and they should report M.
[Section IV, baselines] The two baseline schemes, 'no collaboration' and 'no knowledge sharing', are only identified by citations to [10] and [15]; the optimization problems solved for these baselines are not stated. Since the central performance claim is that the proposed mechanism outperforms these baselines, the comparison is only meaningful if the reader can verify which variables are optimized and which constraints are retained. Please specify the baseline formulations explicitly, or state clearly that they are taken verbatim from the cited papers with their original parameter settings.

minor comments (5)

[Table I] The entry 'f C n [4, 2] G Hz' appears to be a typographical artifact; please clarify the computing speed f_C^j and its units.
[Section II-C, after Eq. (16)] The sentence 'Each MD m has a minimum semantic accuracy requirement' uses the symbol m, but MDs are indexed by i; please correct the notation.
[Section III-A] The phrase 'one-dimension l inear search' contains a missing space and should read 'one-dimensional linear search'.
[Section III-A, Eq. (20)-(21)] The substitution b̃ = ab is acceptable because Y(a,b) and T(a,b) depend on b only through a and ab; it would improve readability to state this explicitly when introducing b̃.
[Problem (18)] The objective allows δi,j,k = 0 for all k for some MDs, so the 'total GESTR of all MDs' may be maximized by not serving some MDs; if the intent is to serve every MD, a minimum-service constraint should be added.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the GESTR derivation and FP-BnB/K-M optimality claims are structurally self-contained; the fitted semantic-accuracy surrogate is an explicit external modeling input, not a back-fitted prediction.

full rationale

The paper's central analytical claims are (i) that problem (18) decomposes, for fixed δ, into independent joint subproblems (19) whose variables (a, b, ξ) can be optimized by FP-BnB, and (ii) that the resulting allocation problem (24) is a bipartite weighted matching solvable by a modified Kuhn-Munkres algorithm. These claims follow from the problem structure (constraints (5), (6), (14), (16) separate over (i, j, k) once δ is fixed) and from standard fractional-programming and matching arguments; they do not import the paper's own conclusions as inputs. The semantic accuracy model (15) is explicitly presented as an approximate surrogate with tuning parameters fitted to external data [9], and the GESTR (17) is defined directly in terms of this surrogate. Using a fitted surrogate inside an optimization objective is a modeling choice, not a circular prediction: the paper never fits (15) to the same quantities it then claims to predict, and the optimality of the resource allocation is conditional on that stated model. The self-citations [13] and [16] are to a standard K-M algorithm and to a conventional channel/parameter setup; neither is used to prove uniqueness or to forbid alternatives, so they are not load-bearing. A serious internal inconsistency exists: Section IV adopts (θ1, θ2, θ3, θ4) = (−6.205e−8, 16.45, 0.9228, −0.06917), which violates the non-negativity constraints stated for Eq. (15) and makes ε non-monotone and greater than 1; however, this is a correctness issue with the fitted surrogate, not a circularity in the derivation chain.

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

The central derivation rests on a fitted semantic accuracy model, a simplified additive delay model, and a matching algorithm whose optimality relies on details not fully proven. The free parameters are few but the most load-bearing one (the ε(ξ) fit) is imported from previous work and not validated here.

free parameters (3)

θ1, θ2, θ3, θ4 = (-6.205e-8, 16.45, 0.9228, -0.06917)
Tuning parameters in the semantic accuracy model (15), fitted via nonlinear least squares to data from [9]. They are used directly in the objective function and constraints, and all numerical results inherit this fitted model.
ρ = 1 (default)
Constant parameter in ω_i,j,k = 1/ξ_i,j,k^ρ, varying with adopted semantic models. It is set to 1 in simulations without tuning justification.
M (segments in one-dimensional search) = not specified
The number of equal segments in the linear search for ξ_i,j,k is not specified, affecting the exact optimality claim. It is a discretization parameter chosen by hand.

assumptions (3)

domain assumption The semantic accuracy ε_i,j,k can be accurately approximated by the exponential model in (15) with parameters fitted to data from [9].
The paper states that closed-form ε(ξ) is intractable and adopts this surrogate. The correctness of all results depends on this approximation being valid for the simulated tasks.
domain assumption Knowledge sharing and semantic transmission can be modeled as additive time components, with no interference or congestion effects beyond the subchannel exclusivity constraints (5)-(6).
The task completion time (13) is a simple sum of transmission, computing, and knowledge sharing times. The model ignores queueing, retransmissions, and other real-world effects.
standard math The fractional programming convergence proof in [14] applies to the specific non-convex relaxed problem (22) with the additional coupled constraints (21c).
The paper cites [14] for convergence but does not verify that the assumptions of that proof (e.g., quasi-concavity or specific structure) hold for the transformed problem here.

invented entities (1)

Generalized effective semantic transmission rate (GESTR), γ_i,j,k
purpose: A new performance metric that combines semantic accuracy, semantic information, knowledge sharing overhead, and bit fallback in one rate.
GESTR is defined in equation (17) but is not independently validated against any physical or semantic metric. It is a rational combination of existing quantities, so its usefulness is a modeling choice.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Collaborative Knowledge Sharing-empowered Effective Semantic Rate Maximization for Two-tier Semantic-Bit Communication Networks." pith.science (2026). https://pith.science/paper/AII2UKEU

@misc{pith2026250101637,
  author       = {Pith},
  title        = {Pith review of: Collaborative Knowledge Sharing-empowered Effective Semantic Rate Maximization for Two-tier Semantic-Bit Communication Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AII2UKEU}},
  note         = {Machine review of arXiv:2501.01637}
}

read the original abstract

Effective task-oriented semantic communications relies on perfect knowledge alignment between transmitters and receivers for accurate recovery of task-related semantic information, which can be susceptible to knowledge misalignment and performance degradation in practice. To tackle this issue, continual knowledge updating and sharing are crucial to adapt to evolving task and user related demands, despite the incurred resource overhead and increased latency. In this paper, we propose a novel collaborative knowledge sharing-empowered semantic transmission mechanism in a two-tier edge network, exploiting edge cooperations and bit communications to address KB mismatch. By deriving a generalized effective semantic transmission rate (GESTR) that considers both semantic accuracy and overhead, we formulate a mixed integer nonlinear programming problem to maximize GESTR of all mobile devices by optimizing knowledge sharing decisions, extraction ratios, and BS/subchannel allocations, subject to task accuracy and delay requirements. The joint optimum solution can be obtained by proposed fractional programming based branch and bound algorithm and modified Kuhn-Munkres algorithm efficiently. Simulation results demonstrate the superior performance of proposed solution, especially in low signal-to-noise conditions.

Figures

Figures reproduced from arXiv: 2501.01637 by the authors.

**Figure 4.** Total GESTR versus minimum semantic accuracy requir [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗

**Figure 3.** Total GESTR versus maximum delay tolerance. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

16 extracted references · 15 canonical work pages

[9]

Adaptable semantic compres- sion and resource allocation for task-oriented communicat ions,

C. Liu, C. Guo, Y . Y ang, and N. Jiang, “Adaptable semantic compres- sion and resource allocation for task-oriented communicat ions,” IEEE Transactions on Cognitive Communications and Networking , vol. 10, no. 3, pp. 769–782, 2024

work page 2024
[10]

Task-oriented and semantic-aware heterogeneous network s for artiﬁcial intelligence of things: Performance analysis and optimiza tion,

X. Xu, B. Xu, S. Han, C. Dong, H. Xiong, R. Meng, and P . Zhan g, “Task-oriented and semantic-aware heterogeneous network s for artiﬁcial intelligence of things: Performance analysis and optimiza tion,” IEEE Internet of Things Journal , vol. 11, no. 1, pp. 228–242, 2024

work page 2024
[15]

Knowledge base aware semantic communication in vehicular networks,

L. Xia, Y . Sun, D. Niyato, K. Ma, J. Kang, and M. A. Imran, “ Knowledge base aware semantic communication in vehicular networks,” in ICC 2023 - IEEE International Conference on Communications , 2023, pp. 3989– 3994

work page 2023
[1]

J. Guo, H. Chen, B. Song, Y . Chi, C. Y uen, F. R. Y u, G. Y . Li, and D. Niyato, “Distributed task-oriented communication n etworks with 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 100 150 200 250 300 350 Fig. 4. Total GESTR versus minimum semantic accuracy requir ements. multimodal semantic relay and edge intelligence,” IEEE Communica- tions Magazine, vol. ...

work page 2024
[2]

Towar d in- telligent resource allocation on task-oriented semantic c ommunication,

H. Zhang, H. Wang, Y . Li, K. Long, and V . C. M. Leung, “Towar d in- telligent resource allocation on task-oriented semantic c ommunication,” IEEE Wireless Communications , vol. 30, no. 3, pp. 70–77, 2023

work page 2023
[3]

Robust semantic communications with masked VQ-V AE enabled codebo ok,

Q. Hu, G. Zhang, Z. Qin, Y . Cai, G. Y u, and G. Y . Li, “Robust semantic communications with masked VQ-V AE enabled codebo ok,” IEEE Transactions on Wireless Communications , vol. 22, no. 12, pp. 8707–8722, 2023

work page 2023
[4]

Semantic know ledge base-enabled zero-shot multi-level feature transmission optimization,

Y . Sun, H. Chen, X. Xu, P . Zhang, and S. Cui, “Semantic know ledge base-enabled zero-shot multi-level feature transmission optimization,” IEEE Transactions on Wireless Communications , vol. 23, no. 5, pp. 4904–4917, 2024

work page 2024
[5]

Wireless Resource Optimization in Hybrid Semantic/Bit Communication Networks

L. Xia, Y . Sun, D. Niyato, L. Zhang, and M. A. Imran, “Wirel ess resource optimization in hybrid semantic/bit communicati on networks,” arXiv preprint arXiv:2404.04162 , 2024

work page Pith review arXiv 2024

Show all 16 references

[6]

Generative AI-driven semantic communication networks: A rchitecture, 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: A rchitecture, technologies and applications,” IEEE Transactions on Cognitive Com- munications and Networking , pp. 1–1, 2024

2024
[7]

Computing ofﬂo ading and semantic compression for intelligent computing tasks in ME C systems,

Y . Zheng, T. Zhang, R. Huang, and Y . Wang, “Computing ofﬂo ading and semantic compression for intelligent computing tasks in ME C systems,” in 2023 IEEE Wireless Communications and Networking Conferen ce (WCNC), 2023, pp. 1–6

2023
[8]

Online resource allocation for semantic-aware edge compu ting sys- tems,

Y . Cang, M. Chen, Z. Y ang, Y . Hu, Y . Wang, C. Huang, and Z. Zh ang, “Online resource allocation for semantic-aware edge compu ting sys- tems,” IEEE Internet of Things Journal , pp. 1–1, 2023

2023
[11]

The levenberg-marquardt algorithm for no nlinear least squares curve-ﬁtting problems,

H. P . Gavin, “The levenberg-marquardt algorithm for no nlinear least squares curve-ﬁtting problems,” Department of Civil and Environmental Engineering Duke University August , vol. 3, 2019

2019
[12]

Resource all ocation for text semantic communications,

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

2022
[13]

Joint computatio n ofﬂoading and radio resource allocations in wireless cellular networ ks,

H. Chen, D. Zhao, Q. Chen, and R. Chai, “Joint computatio n ofﬂoading and radio resource allocations in wireless cellular networ ks,” in 2018 10th International Conference on Wireless Communications and Signal Processing (WCSP), 2018, pp. 1–6

2018
[14]

Fractional programming for communic ation systems—part I: Power control and beamforming,

K. Shen and W. Y u, “Fractional programming for communic ation systems—part I: Power control and beamforming,” IEEE Transactions on Signal Processing , vol. 66, no. 10, pp. 2616–2630, 2018

2018
[16]

Semantic extraction mode l selection for IoT devices in edge-assisted semantic communications,

H. Chen, F. Fang, and X. Wang, “Semantic extraction mode l selection for IoT devices in edge-assisted semantic communications, ” IEEE Com- munications Letters , vol. 28, no. 7, pp. 1733–1737, 2024

2024

Pith tools

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

[1] [9]

Adaptable semantic compres- sion and resource allocation for task-oriented communicat ions,

C. Liu, C. Guo, Y . Y ang, and N. Jiang, “Adaptable semantic compres- sion and resource allocation for task-oriented communicat ions,” IEEE Transactions on Cognitive Communications and Networking , vol. 10, no. 3, pp. 769–782, 2024

work page 2024

[2] [10]

Task-oriented and semantic-aware heterogeneous network s for artiﬁcial intelligence of things: Performance analysis and optimiza tion,

X. Xu, B. Xu, S. Han, C. Dong, H. Xiong, R. Meng, and P . Zhan g, “Task-oriented and semantic-aware heterogeneous network s for artiﬁcial intelligence of things: Performance analysis and optimiza tion,” IEEE Internet of Things Journal , vol. 11, no. 1, pp. 228–242, 2024

work page 2024

[3] [15]

Knowledge base aware semantic communication in vehicular networks,

L. Xia, Y . Sun, D. Niyato, K. Ma, J. Kang, and M. A. Imran, “ Knowledge base aware semantic communication in vehicular networks,” in ICC 2023 - IEEE International Conference on Communications , 2023, pp. 3989– 3994

work page 2023

[4] [1]

J. Guo, H. Chen, B. Song, Y . Chi, C. Y uen, F. R. Y u, G. Y . Li, and D. Niyato, “Distributed task-oriented communication n etworks with 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 100 150 200 250 300 350 Fig. 4. Total GESTR versus minimum semantic accuracy requir ements. multimodal semantic relay and edge intelligence,” IEEE Communica- tions Magazine, vol. ...

work page 2024

[5] [2]

Towar d in- telligent resource allocation on task-oriented semantic c ommunication,

H. Zhang, H. Wang, Y . Li, K. Long, and V . C. M. Leung, “Towar d in- telligent resource allocation on task-oriented semantic c ommunication,” IEEE Wireless Communications , vol. 30, no. 3, pp. 70–77, 2023

work page 2023

[6] [3]

Robust semantic communications with masked VQ-V AE enabled codebo ok,

Q. Hu, G. Zhang, Z. Qin, Y . Cai, G. Y u, and G. Y . Li, “Robust semantic communications with masked VQ-V AE enabled codebo ok,” IEEE Transactions on Wireless Communications , vol. 22, no. 12, pp. 8707–8722, 2023

work page 2023

[7] [4]

Semantic know ledge base-enabled zero-shot multi-level feature transmission optimization,

Y . Sun, H. Chen, X. Xu, P . Zhang, and S. Cui, “Semantic know ledge base-enabled zero-shot multi-level feature transmission optimization,” IEEE Transactions on Wireless Communications , vol. 23, no. 5, pp. 4904–4917, 2024

work page 2024

[8] [5]

Wireless Resource Optimization in Hybrid Semantic/Bit Communication Networks

L. Xia, Y . Sun, D. Niyato, L. Zhang, and M. A. Imran, “Wirel ess resource optimization in hybrid semantic/bit communicati on networks,” arXiv preprint arXiv:2404.04162 , 2024

work page Pith review arXiv 2024