Zero-Forget Preservation of Semantic Communication Alignment in Distributed AI Networks

Geoffrey Ye Li; Jingzhi Hu

REVIEW 3 major objections 5 minor 18 references

Zero-Forget Preservation of Semantic Communication Alignment in Distributed AI Networks

T0 review · 3 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Distributed AIs can adapt to local domains while preserving perfect semantic-communication alignment by making sparse, switchable parameter patches rather than dense updates.

desk verdict A genuine and clean idea—carrying domain adaptation as switchable sparse deltas—but the empirical claim of matching full DA at 1% sparsity is confounded by unequal training budgets and missing error bars. read the letter →

arxiv 2411.19385 v1 pith:YYQ7KAKG submitted 2024-11-28 cs.LG cs.AIeess.SP

Jingzhi Hu , Geoffrey Ye Li This is my paper · ORCID

classification cs.LGcs.AIeess.SP

keywords semanticcommunicationsdomainadaptationsparseadditivemodificationsneuralmodelalignmentstraight-throughestimationimagetransmissiondistributedAI

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper proposes a way for distributed AI systems to keep their semantic communication link intact even after each AI adapts to a different local domain. The method, zero-forget domain adaptation, performs adaptation through sparse additive modifications to the neural network parameters instead of changing the parameters themselves. Because the patch touches less than one percent of the parameters, it can be stored cheaply and switched off at any time, which returns both transmitter and receiver to the original pre-trained encoder–decoder pair that is perfectly aligned. The authors show experimentally on image transmission that this sparse adaptation matches or slightly beats full domain adaptation while eliminating the misalignment loss entirely.

What carries the argument

The central object is the sparse additive modification (SAM): a set of per-parameter changes written as a binary mask multiplied by a continuous modification vector, with the mask selecting the top-gamma fraction of parameters by an importance score. The SAM is optimized by straight-through gradient estimation, treating the mask as identity for backpropagation so that scores and modifications can be updated by gradient descent, and by a layer-wise linear sparsity allocation that distributes the parameter budget proportionally to input-plus-output dimensions rather than parameter counts. Its load-bearing property is switch-off: because the adapted weights are pre-trained weights plus the SAM, subtracting the SAM gives back the aligned pre-trained weights exactly, making the misalignment loss zero without any joint retraining or equalization.

What would settle it

Take the same autoencoder and compare a 1%-sparse SAM against full-parameter domain adaptation on a shift that changes global image statistics, such as Gaussian noise, blur, or weather effects applied to the same image set; if full DA beats the SAM by more than a small margin on domain PSNR, the claim that sparse patches can carry the adaptation value would be false on that shift.

Watch

Extended reading notes

Core claim

The central claim is that the optimal alignment process for a pair of adapted AIs is simply to subtract the sparse additive modifications: $A^*(\hat{\theta}_T, \hat{\phi}_R) = (\hat{\theta}_T - \Delta\hat{\theta}_T, \hat{\phi}_R - \Delta\hat{\phi}_R) = (\theta^*, \phi^*)$. In other words, if each AI achieves domain adaptation by adding a sparse patch to the shared pre-trained weights, switching the patch off restores the original aligned model with zero forgetting. The empirical counterpart is that with at most 1% of parameters modified, the optimized patch attains domain-reconstruction PSNR comparable to or better than full baseline domain adaptation on four synthetic image shifts, and the saved patch costs less than 1% of the model size in additional memory.

Load-bearing premise

The framework assumes that modifying at most 1% of a shared model's parameters can carry essentially all of the domain-adaptation value; this is tested on one autoencoder and four synthetic image shifts, with no error bars and no comparison to other parameter-efficient fine-tuning approaches.

Editorial extensions

If this is right

Tx and Rx AIs can toggle between a perfectly aligned pre-trained state and a locally adapted state at will, with no alignment traffic.
SC alignment overhead is reduced from joint training or latent-space equalization to storing a patch smaller than 1% of the model.
Domain adaptation quality does not suffer: at sparsity ratios up to 1%, SAM-based adaptation matches or exceeds full DA on all four tested shifts.
The optimized mask and linear sparsity distribution are worth measurable gains (0.73 dB and 1.13 dB respectively on the tested domains).
Since the SAM is additive and separable, one model could host multiple domain patches, each independently storable and removable.

Reading between the lines

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

Editorial: the same switch-off idea could be applied to parameter-efficient fine-tuning methods such as adapters or low-rank updates, provided the residual is sparse enough to store; the paper does not compare against such methods.
Editorial: for real-world shifts that change low-level statistics broadly, such as sensor noise or weather corruption, a 1% parameter budget may be too tight; stress-testing on naturally corrupted datasets would map where the sparsity assumption breaks.
Editorial: if sparse additive patches generalize, a single shared pre-trained backbone could serve many users in a semantic network, with each user's domain represented by a small patch and alignment guaranteed by simply not loading patches at the other end.
Editorial: the linear sparsity allocation rule comes from lottery-ticket literature; a direct test would vary architecture depth and width to see whether the rule remains optimal.

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,

Read the ZFDA paper. The core trick is genuinely nice: instead of trying to realign two adapted models after the fact, you carry each model's domain adaptation as a sparse additive delta on the shared pretrained weights, and you switch the delta off to restore the original semantic communication alignment. Equation (4) is true by construction, so the authors are right that the real claim to scrutinize is whether a ≤1%-sparse delta can match full-model domain adaptation.

What's new: in the semantic communication literature, alignment is usually done via joint tuning or latent-space equalizers. Carrying DA as switchable supermask-style deltas is a new application, and the formulation (scores, mask, modifications) is clean. The paper is honest about what it imports—the mechanism comes from Ramanujan et al. and Bengio et al., and the layer-wise budget from Evci et al. Proposition 1 is a plausible extension of the known result, though the proof is a sketch.

The soft spot is exactly the one the stress-test flags. ZFDA gets 30 epochs on the domain; the baseline gets 10. That is three times the gradient steps. So Fig. 4's "comparable or better than baseline DA" could just be longer training, not a benefit of sparsity. The paper attributes the gain to reduced overfitting, but that attribution is not identifiable from this setup. Missing error bars or multiple seeds make it worse: the displayed gaps could be run-to-run variance. That's fixable, but it is a load-bearing flaw in the experimental support.

Two smaller issues. The "less than 1% memory" claim needs accounting: storing the binary mask plus the non-zero values can double the overhead for extreme sparsity unless you compress indices, which is not discussed. And the baselines are weak: no comparison to LoRA/adapter-style parameter-efficient fine-tuning, which is the natural alternative. One autoencoder and four synthetic shifts is a thin platform for a generality claim about realistic domains.

None of this kills the idea. The framework is sound, the writing is clear, and the zero-forget property is a legitimate selling point. But the paper's stated empirical guarantee—matching full DA at 1% sparsity—is not yet established. It needs a fair comparison (matched epochs, seeds), memory accounting, and at least one stronger baseline. I'd send it to review: the idea deserves referee time, and the experiment is fixable. A careful reviewer should ask for exactly the matched-budget comparison that's missing.

Recommendation: peer review, with major revision expected.

Referee Report

3 major / 5 minor

Summary. The paper proposes a zero-forget domain adaptation (ZFDA) framework for semantic communication systems in which transmitter and receiver AIs inherit the same pretrained encoder/decoder. Instead of fine-tuning the full model, each AI learns a sparse additive modification (SAM) to its parameters. Because the SAM can be switched off, the original pretrained parameters are restored exactly, so the semantic alignment loss defined in Eq. (2) is reduced to zero by Eq. (4). The optimization problem (P1) is tackled by decomposing each SAM into a binary mask and a continuous vector, using a score-based straight-through estimator for the mask, and allocating layerwise sparsity budgets via Eq. (11). Experiments on an autoencoder-based image transmission system with four synthetic domain shifts report that sparsity ratios below 1% achieve domain-adaptation PSNR comparable to or better than full baseline DA, while preserving alignment for free.

Significance. If the empirical claims hold, the paper offers a clean conceptual insight: alignment preservation can be made definitional by storing domain adaptations as removable sparse patches rather than as permanent parameter changes. The formulation in (P1) and the decomposition in Eq. (7) are clear, and the zero-forget statement in Eq. (4) is a direct, machine-checkable consequence of the construction. The paper also includes a proof attempt for Proposition 1 and a useful ablation of mask optimization and sparsity distribution. However, the non-tautological part of the central claim is that a <=1%-sparse SAM can match full fine-tuning DA quality; that claim is currently supported by a single autoencoder, four synthetic shifts, unequal training budgets, and no error bars or multiple seeds. The advertised storage cost of 'less than 1% additional memory' also appears to omit the cost of the mask or indices. These issues are fixable, but they are load-bearing for the paper's main conclusions.

major comments (3)

[Section V, Fig. 4] The central DA-performance claim is confounded by unequal training budgets. The text states that baseline DA trains the full pretrained model for 10 epochs, while the ZFDA SAM is trained for 30 epochs with alpha_s=1 and alpha_v=1e-4. ZFDA therefore receives three times as many gradient steps on the domain, so the observation that ZFDA sometimes exceeds baseline DA cannot be attributed to sparsity or to the regularizing effect of limiting parameter changes. Since no seeds or error bars are reported, the differences in Fig. 4 may also be run-to-run variance. Please rerun the comparison with equal training budgets (e.g., the same number of epochs or the same number of gradient evaluations) and report mean and standard deviation over at least five seeds.
[Abstract, Section V, Section VI] The claim of 'less than 1% additional memory' equates the sparsity ratio with storage overhead, but a SAM must store both the nonzero modification values and the mask or indices that identify them. A binary mask alone costs N bits versus 32N bits for 32-bit parameters, i.e., 3.125% overhead at gamma=1%, and explicit index storage typically adds another gamma*N index entries. Thus the actual memory overhead at a 1% sparsity ratio is above 1% unless a compressed or implicit representation is specified. Please specify the storage format used in the experiments and recompute the memory ratio, or revise the abstract and conclusion accordingly.
[Section IV-B, Appendix A] Proposition 1 is not established for arbitrary neural architectures as stated. The proof in Appendix A uses a first-order Taylor expansion of the loss with respect to mask entries and ignores higher-order terms, and it does not account for the magnitudes of the modification values v_i, even though a mask switch from 0 to 1 changes the network by m_i * v_i, which can be large. For a nonlinear network, a finite score update does not guarantee a monotone loss decrease without additional assumptions such as sufficiently small step sizes or sufficiently small v. Please restate Proposition 1 as a heuristic motivation for the score update, or provide a proof under explicit assumptions that match the experiments.

minor comments (5)

[Section II] There are two typos: 'identify function' should be 'identity function', and 'cellar' should be 'cellular'.
[Section IV-B] Equation (9) approximates the gradient through the binary mask using the straight-through estimator; please state explicitly that m_i is treated as the identity for gradient flow, since this is a nonstandard but well-known approximation.
[Section V, Fig. 4] The x-axis labels omit the percent sign on the last tick, and the legend entries for the four cases (linear/uniform sparsity, optimized/fixed mask) are not visually tied to the curves; please clarify the figure or caption.
[Section V] The description of the local domain dataset is ambiguous: CIFAR-100 has 100 classes, the pretraining uses 80 classes, and the domain uses 10 classes 'unused in the pre-training'; please report the exact class split and the number of samples per domain.
[Section V] The paper does not compare ZFDA with other parameter-efficient domain adaptation methods such as adapters or low-rank adaptations; adding such a comparison would help contextualize the DA-quality claim and the memory-efficiency claim.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: zero-forget is a definitional design property, DA performance is measured on held-out data, and the sole self-citation is non-load-bearing background.

full rationale

Walking the claimed derivation chain, there is no step where a prediction reduces to an input by construction. The zero-forget property in Eq. (4) is an architectural identity: the SAM is defined as an additive modification, and A* is defined as subtracting it, so A*(theta*_T, phi*_R)=(theta*, phi*) follows from the framework's definitions; the paper does not treat this identity as a fitted or statistically forced result, and the empirical content of the paper is the Fig. 4 claim that a <=1%-sparse SAM reaches baseline DA quality, which is evaluated on held-out CIFAR-100 test images after training on domain data. The mask-optimization scheme is imported from external references [15], [16], [17] rather than from the authors' own prior work, and Proposition 1 is proved in the appendix. The only self-citation, Ref. [3] (co-author G.Y. Li), supports general background on the SC paradigm and is not load-bearing for the ZFDA derivation. The unequal training budgets of baseline DA (10 epochs) and ZFDA (30 epochs) in Fig. 4 are a fairness/correctness concern about whether the DA-comparability claim is established, but they are not circularity. Overall the derivation is self-contained with respect to its empirical claims.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central claim rests on the empirical hypothesis that sparse additive modifications at or below 1% can carry domain adaptation, plus standard shared-pretraining and reliable-channel assumptions. No new physical entities are introduced; the SAM is a sparse delta vector, and the importance score is a known auxiliary variable from [15].

free parameters (4)

Sparsity ratio gamma = 0.03% to 1% (swept)
Controls the trade-off between DA performance and memory; chosen by hand across values, not predicted.
Score learning rate alpha_s = 1
Set manually in Sec. V; affects the mask selection dynamics.
Modification learning rate alpha_v = 0.0001
Set manually in Sec. V; controls the size of additive modifications.
SAM training epochs = 30 (vs 10 for baseline DA)
ZFDA is trained longer than the baseline, confounding the comparison of sparsity effects with training budget.

assumptions (5)

standard math First-order Taylor expansion is a valid approximation of the loss change from mask flips (Appendix A).
The proof of Proposition 1 uses a first-order expansion; monotonic decrease is only approximate for finite score steps.
domain assumption Straight-through estimation provides usable gradients through the binary mask (Sec. IV-B).
The gradient of the indicator function is zero almost everywhere; the paper adopts the approximation from [15].
domain assumption The communication channel is an identity function, i.e., semantics arrive as reliable bits (Sec. II).
The paper assumes above 99.9% reliability and ignores channel distortion, so alignment loss is measured without noise.
domain assumption AIs inherit the same pretrained model from a shared knowledge base (Sec. I).
Initial alignment and the ability to switch off SAM both require the base weights to be available and identical.
domain assumption Linear layer-wise sparsity allocation improves sparse training (Eq. 11, from [17]).
The allocation rule is imported from Rigging the Lottery; the paper does not re-derive or test it independently.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Zero-Forget Preservation of Semantic Communication Alignment in Distributed AI Networks." pith.science (2026). https://pith.science/paper/YYQ7KAKG

@misc{pith2026241119385,
  author       = {Pith},
  title        = {Pith review of: Zero-Forget Preservation of Semantic Communication Alignment in Distributed AI Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YYQ7KAKG}},
  note         = {Machine review of arXiv:2411.19385}
}

read the original abstract

Future communication networks are expected to connect massive distributed artificial intelligence (AI). Exploiting aligned priori knowledge of AI pairs, it is promising to convert high-dimensional data transmission into highly-compressed semantic communications (SC). However, to accommodate the local data distribution and user preferences, AIs generally adapt to different domains, which fundamentally distorts the SC alignment. In this paper, we propose a zero-forget domain adaptation (ZFDA) framework to preserve SC alignment. To prevent the DA from changing substantial neural parameters of AI, we design sparse additive modifications (SAM) to the parameters, which can be efficiently stored and switched-off to restore the SC alignment. To optimize the SAM, we decouple it into tractable continuous variables and a binary mask, and then handle the binary mask by a score-based optimization. Experimental evaluations on a SC system for image transmissions validate that the proposed framework perfectly preserves the SC alignment with almost no loss of DA performance, even improved in some cases, at a cost of less than 1% of additional memory.

Figures

Figures reproduced from arXiv: 2411.19385 by the authors.

**Figure 2.** Samples from the original domain and the other varied local domains. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** , although lots of resources are devoted to SC alignment, the achievable performance is still unsatisfactory: 2.41 dB and 0.80dB average increase in PSNR contributed by tuning-based and equalizer-based methods, respectively. In stark contrast, our ZFDA framework suffers from zero loss with no overheads for gaining the SC alignment as SAMs can be switched-off. A natural question to follow is how ZFDA performs in adap… view at source ↗

Figures from the paper (1 more)

**Figure 4.** Figure 4: PSNR for the four local domains versus the sparsity rate of the SAM, [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

18 extracted references · 12 canonical work pages

[1]

From connected people, connected things, to connected intelligence,

Y . Chen, P. Zhu, G. He, X. Yan, H. Baligh, and J. Wu, “From connected people, connected things, to connected intelligence,” in Proc. 6G SUMMIT, Levi, Finland, Mar. 2020

work page 2020
[2]

Tong and P

W. Tong and P. Zhu, 6G: The Next Horizon . Cambridge: Cambridge University Press, 2021

work page 2021
[3]

Deep learning enabled semantic communication systems,

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

2021
[4]

Semantic communications: Overview, open issues, and future research directions,

X. Luo, H.-H. Chen, and Q. Guo, “Semantic communications: Overview, open issues, and future research directions,” IEEE Wireless Commun. , vol. 29, no. 1, pp. 210–219, Feb. 2022

2022
[5]

Deep joint source- channel coding for wireless image transmission,

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

2019
[6]

Deep learning- enabled semantic communication systems with task-unaware transmitter and dynamic data,

H. Zhang, S. Shao, M. Tao, X. Bi, and K. B. Letaief, “Deep learning- enabled semantic communication systems with task-unaware transmitter and dynamic data,” IEEE J. Sel. Areas Commun. , vol. 41, no. 1, pp. 170–185, Jan. 2023

2023
[7]

Post-deployment fine- tunable semantic communication,

P. Si, R. Liu, L. Qian, J. Zhao, and K.-Y . Lam, “Post-deployment fine- tunable semantic communication,” IEEE Trans. Wireless Commun., Aug. 2024, early access

work page 2024
[8]

Synchronization mechanism: Preliminary attempt to enable wireless semantic communication with heterogeneous knowledge bases,

X. Xu, Z. Bian, B. Wang, Y . Chen, S. Han, C. Sun, W. Zhang, and P. Zhang, “Synchronization mechanism: Preliminary attempt to enable wireless semantic communication with heterogeneous knowledge bases,” IEEE Commun. Lett. , vol. 28, no. 8, pp. 1815–1819, Aug. 2024

work page 2024

Show all 18 references

[9]

Semantics alignment via split learning for resilient multi-user semantic communication,

J. Choi, J. Park, S.-W. Ko, J. Choi, M. Bennis, and S.-L. Kim, “Semantics alignment via split learning for resilient multi-user semantic communication,” IEEE Trans. Veh. Technol., vol. 73, no. 10, pp. 15 815– 15 819, Oct. 2024

2024
[10]

Semantic channel equalizer: Modelling language mismatch in multi-user semantic communications,

M. Sana and E. C. Strinati, “Semantic channel equalizer: Modelling language mismatch in multi-user semantic communications,” in Proc. IEEE GlobeCom, Kuala Lumpur, Malaysia, Dec. 2023

2023
[11]

Dynamic relative representations for goal-oriented semantic communications,

S. Fiorellino, C. Battiloro, E. C. Strinati, and P. Di Lorenzo, “Dynamic relative representations for goal-oriented semantic communications,” arXiv:2403.16986, Mar. 2024

2024 arXiv
[12]

Soft partitioning of latent space for semantic channel equalization,

T. H ¨uttebr¨aucker, M. Sana, and E. C. Strinati, “Soft partitioning of latent space for semantic channel equalization,” in Proc. IEEE ISWCS, Rio de Janeiro, Brazil, Jul. 2024

2024
[13]

Supermasks in superposition,

M. Wortsman, V . Ramanujan, R. Liu, A. Kembhavi, M. Rastegari, J. Yosinski, and A. Farhadi, “Supermasks in superposition,” in Proc. NeurIPS, Online, Dec. 2020

2020
[14]

5G; Service requirements for next generation new services and markets (3GPP TS 22.261 version 15.7.0 Release 15),

ETSI, “5G; Service requirements for next generation new services and markets (3GPP TS 22.261 version 15.7.0 Release 15),” European Telecommunications Standards Institute (ETSI), Sophia Antipolis, France, Tech. Rep. TS 122 261 V15.7.0, March 2019. [Online]. Available: http://ww...

2019
[15]

What’s hidden in a randomly weighted neural network?

V . Ramanujan, M. Wortsman, A. Kembhavi, A. Farhadi, and M. Raste- gari, “What’s hidden in a randomly weighted neural network?” in Proc. IEEE/CVF CVPR, Online, Jun. 2020

2020
[16]

Estimating or propagat- ing gradients through stochastic neurons for conditional computation,

Y . Bengio, N. L ´eonard, and A. Courville, “Estimating or propagat- ing gradients through stochastic neurons for conditional computation,” arXiv:1308.3432, Aug. 2013

2013 arXiv
[17]

Rigging the lottery: Making all tickets winners,

U. Evci, T. Gale, J. Menick, P. S. Castro, and E. Elsen, “Rigging the lottery: Making all tickets winners,” in Proc. ICML, Online, Jun. 2020

2020
[18]

Learning multiple layers of features from tiny images,

A. Krizhevsky, G. Hinton et al. , “Learning multiple layers of features from tiny images,” 2009, Technical Report TR-2009

2009

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.

[1] [1]

From connected people, connected things, to connected intelligence,

Y . Chen, P. Zhu, G. He, X. Yan, H. Baligh, and J. Wu, “From connected people, connected things, to connected intelligence,” in Proc. 6G SUMMIT, Levi, Finland, Mar. 2020

work page 2020

[2] [2]

Tong and P

W. Tong and P. Zhu, 6G: The Next Horizon . Cambridge: Cambridge University Press, 2021

work page 2021

[3] [3]

Deep learning enabled semantic communication systems,

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

2021

[4] [4]

Semantic communications: Overview, open issues, and future research directions,

X. Luo, H.-H. Chen, and Q. Guo, “Semantic communications: Overview, open issues, and future research directions,” IEEE Wireless Commun. , vol. 29, no. 1, pp. 210–219, Feb. 2022

2022

[5] [5]

Deep joint source- channel coding for wireless image transmission,

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

2019

[6] [6]

Deep learning- enabled semantic communication systems with task-unaware transmitter and dynamic data,

H. Zhang, S. Shao, M. Tao, X. Bi, and K. B. Letaief, “Deep learning- enabled semantic communication systems with task-unaware transmitter and dynamic data,” IEEE J. Sel. Areas Commun. , vol. 41, no. 1, pp. 170–185, Jan. 2023

2023

[7] [7]

Post-deployment fine- tunable semantic communication,

P. Si, R. Liu, L. Qian, J. Zhao, and K.-Y . Lam, “Post-deployment fine- tunable semantic communication,” IEEE Trans. Wireless Commun., Aug. 2024, early access

work page 2024

[8] [8]

Synchronization mechanism: Preliminary attempt to enable wireless semantic communication with heterogeneous knowledge bases,

X. Xu, Z. Bian, B. Wang, Y . Chen, S. Han, C. Sun, W. Zhang, and P. Zhang, “Synchronization mechanism: Preliminary attempt to enable wireless semantic communication with heterogeneous knowledge bases,” IEEE Commun. Lett. , vol. 28, no. 8, pp. 1815–1819, Aug. 2024

work page 2024