Pith. sign in

REVIEW 4 major objections 5 minor 34 references

DISCD: Distributed Lossy Semantic Communication for Logical Deduction of Hypothesis

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

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

Pith's one-line read Selecting messages by semantic informativeness is claimed to concentrate the posterior on the true world state and improve hypothesis deduction under communication constraints.

desk verdict Distributed extension of the single-node semantic communication idea, with a checkable empirical win over random selection, but the main theorem is unproven as stated and the proof contains a normalization error; worth reviewing, not accepting as is. read the letter →

arxiv 2502.05744 v1 pith:TT2GZUVJ submitted 2025-02-09 cs.IT math.IT

Ahmet Faruk Saz , Siheng Xiong , Faramarz Fekri This is my paper · ORCID
classification cs.ITmath.IT
keywords semanticcommunicationdistributedinferencehypothesisdeductionfirst-orderlogicinductivecont-informationlossyPACbound
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 distributed communication protocol in which nodes that each see only part of a logically described world exchange the most semantically informative sentences through a server, rather than random ones. It claims that this selection rule, based on cont-information (the remaining uncertainty about a state after evidence), concentrates the posterior distribution on the true constituent of the world faster, and therefore improves each node's ability to deduce which hypothesis the world supports. The authors prove a comparison theorem (Theorem 4) stating that cont-selected evidence gives the true minimal constituent at least as high posterior probability as random evidence and a tighter PAC bound, and they demonstrate on a custom logical-deduction dataset that the protocol reaches target accuracy in fewer rounds and fewer transmitted bits. If correct, the result gives a principled way to decide what to transmit under a strict communication budget in semantic communication systems.

What carries the argument

The load-bearing device is cont-information, $cont(S_i; e) = 1 - c(S_i, e)$, where $c(S_i, e) = p(S_i \mid e)$ is the inductive logical probability (degree of confirmation) that evidence $e$ supports state $S_i$. The protocol selects, at each round, the sentence $m$ that minimizes cont given the evidence accumulated so far (node-side equation 12; server-side equation 13), and then updates the posterior over constituents via Bayesian updating (equations 15-18). The priors are assigned using a gamma-function form over constituent widths, and the convergence and PAC analysis rests on the minimal constituent receiving probability approaching 1 as evidence accumulates (Theorem 3). The optimization is made computationally tractable in experiments by counting the number of states each FOL sentence satisfies with an exact Boolean model counter.

What would settle it

Enumerate all $B$-sentence evidence sets in a small finite world (e.g., two predicates and two entities, giving 16 state descriptions), compute the posterior of the true minimal constituent under the cont-selection rule and under random selection, and check Theorem 4(a) and the PAC-bound sum in equation (22); any instance where random evidence gives a strictly higher posterior, or a smaller $\epsilon'$, would falsify the claimed advantage.

Watch

Extended reading notes

Core claim

The central discovery is that when each communication round is used to transmit the FOL sentence that minimizes cont-information relative to the observer's accumulated evidence, the inductive-logical posterior over constituents becomes more concentrated on the true minimal constituent than under uniform random transmission. The authors formalize this as Theorem 4: for a fixed budget of $B$ sentences, the posterior probability of the minimal constituent under cont-selected evidence satisfies $c(C^c, e_{j,cont}) \ge c(C^c, e_{j,rand})$, and the PAC-bound error $\epsilon'$ is no larger under cont-selected evidence. The reason offered is that cont-selected evidence yields more discriminative likelihood ratios: the likelihood of the true constituent rises while those of incompatible constituents fall or become zero. This concentration of the posterior lowers the Bayes risk of each node's hypothesis deduction task.

Load-bearing premise

The claimed advantage rests on the assumption that the cont-selection rules in equations (12) and (13) produce evidence whose likelihood ratios are more discriminative for the true minimal constituent; the proof of Theorem 4 asserts this rather than deriving it.

Editorial extensions

If this is right

  • Under the same communication budget, a node using DISCD reaches a target hypothesis-deduction accuracy in fewer rounds than random transmission, so the same accuracy costs fewer bits (the paper reports average savings of roughly 270 bits per node for one-sentence messages and 316 bits for two-sentence messages).
  • As the number of communication rounds grows, each node's posterior over constituents converges toward the true distribution, so the Bayes risk of the deduced hypothesis decreases; this holds for both finite worlds (state descriptions) and infinite worlds (constituents).
  • The PAC bound in Theorem 3 guarantees that the minimal constituent is approximately correct with high probability once enough evidence is accumulated, and Theorem 4 tightens that bound when evidence is cont-selected rather than random.
  • Semantic selection can render some constituents impossible (zero likelihood), which makes the posterior sharply peaked and further improves deduction in later rounds.

Reading between the lines

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

  • If the advantage really comes from sharper likelihood ratios, the same selection principle could be ported to any probabilistic state-space model, not just first-order-logic constituents, as long as each message's effect on the posterior can be scored.
  • The paper compares against random transmission only; a natural next test is to compare cont-selection against other principled heuristics (e.g., selecting the message with maximum mutual information with the state space) to see whether the gain is specific to the cont measure or generic to informative selection.
  • The theory is stated from a single node's perspective; a multi-node extension would need to show that the server's aggregation step preserves the likelihood-ratio ordering when nodes hold overlapping evidence.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes DISCD, a distributed semantic communication framework for hypothesis deduction. Nodes with partial first-order-logic observations exchange messages selected to maximize Carnapian cont-information; a central server aggregates the nodes' updates and broadcasts the most informative FOL sentence. The paper's central theoretical claim, Theorem 4, states that cont-selected evidence yields a posterior probability for the true minimal constituent at least as high as random evidence and a tighter PAC bound. The authors report experiments on a custom logical-deduction dataset showing that DISCD reaches target deduction accuracies in fewer communication rounds than random selection.

Significance. The paper connects Carnap-Hintikka inductive logic to a concrete communication protocol, which is a useful and relatively uncommon model-theoretic alternative to neural semantic communication. The formulation of the selection problem with a cont-information objective and the use of #SAT tools for state counting are concrete engineering contributions. The experiments on a custom dataset, with code released on GitHub, demonstrate an advantage over random message selection in the reported setting. However, the central theoretical guarantee is currently unsupported: the proof of Theorem 4 is a sketch that assumes the very likelihood-ratio dominance it needs to establish, and the experimental evidence lacks error bars and a nontrivial baseline. If the theorem can be properly proved or replaced with a precise convergence statement under explicit conditions, the framework would be a significant contribution to task-oriented semantic communication.

major comments (4)
  1. [III-E, proof of Theorem 4 (Eqs. (20)-(22))] The proof of the central claim is a bare assertion. It states that cont-selected evidence e_{j,cont} 'affects the likelihoods such that P(e_{j,cont}|C^c) is higher' and that incorrect constituents get lower likelihood, but no property of the selection rule (12)-(13), of the inductive logical probability (10), or of the prior/likelihood (16)-(17) is used to derive this monotone likelihood-ratio dominance. The rule chooses messages maximizing the degree of confirmation under the current posterior; from a sufficiently skewed prior, it can select evidence that reinforces a wrong constituent. The theorem requires a proof that cont-selection concentrates posterior mass on the true minimal constituent, not merely an assertion that it does.
  2. [III-E, Eq. (21)] The proof claims P(C^w | e_{j,cont}) >= P(C^w | e_{j,rand}) for every constituent w. Summing both sides over all constituents gives 1 >= 1, so no strict gain for C^c is possible if the inequality is applied to all w. The theorem's statement (a) only needs the inequality at C^c, but the proof's global inequality is inconsistent with the claimed concentration improvement. This normalization issue is load-bearing and must be repaired.
  3. [III-D, Theorem 3 and Eq. (22)] The PAC bound in Theorem 3 is quoted from [31] with no derivation, and the paper explicitly leaves the asymptotic convergence proof to [31]. Part (b) of Theorem 4 depends on this result, but Eq. (22) is a new expression that does not follow from Eq. (19); no algebraic derivation connects the selection mechanism to the claimed smaller ratio P(e | C^{c+i})/P(e | C^c). This is a second unproved step in the paper's main theoretical claim.
  4. [IV, Experiment Results] The experimental comparison uses only random message selection as a baseline and reports single success-rate numbers with no error bars, confidence intervals, or number of repeated runs. The random baseline is not described in enough detail (e.g., whether results are averaged over random seeds, and how many). The dataset is custom and its scale, difficulty, and relation to the protocol's assumptions are not specified. Consequently, the empirical support for the claimed advantage over random messaging is not statistically substantiated.
minor comments (5)
  1. [II-C, Definition 6 and Eqs. (12)-(13)] The argument order of cont is inconsistent: Definition 6 defines cont(S_i; e) = 1 - c(S_i, e), but Eqs. (12) and (13) write cont(e_j ∧ ..., m), where the state/evidence roles are unclear. Please define cont on arbitrary sentences and evidence consistently.
  2. [II-B, Definition 5 and Eq. (10)] The notation \(\lambda(w_{S_i})\) and \(w_{S_i}\) is not defined explicitly; in particular, \(w_{S_i}\) is introduced as the 'weight assigned to state' but its relationship to the prior-induced width parameter in Eq. (17) and to \(\alpha\) in Eq. (16) is never stated.
  3. [III-B, System Model] The phrase 'it updates it's perceived SotW' contains a typo ('it's' for 'its'); there are several other grammatical slips throughout the paper that should be corrected in a revised version.
  4. [IV, Fig. 2 and Table I] Figure 2 would benefit from error bars or shaded confidence regions, and Table I should state explicitly what 'N/A' means (e.g., the target success rate was never reached within the tested rounds) and over how many runs the costs are averaged.
  5. [III-D, Section text] The sentence 'Despite omitted from this manuscript due to space concerns, it had been proven...' is an explicit acknowledgement that a key proof is missing; if space is the constraint, the full derivation should be placed in an appendix or supplementary document rather than cited away.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: Theorem 4 is unsupported, but the claimed advantage is not identical to its input by construction.

full rationale

The derivation chain is not circular. The cont-selection rule (12)-(13) chooses messages by minimizing cont(e_j, m), which is a function of the inductive confirmation of messages under the current evidence; Theorem 4 concludes an inequality about the posterior c(C^c, e) of the minimal constituent. These are distinct quantities, linked only by the Carnap-Hintikka model (16)-(18), so the theorem does not follow from the selection rule by definition. The proof of Theorem 4 asserts, without derivation from (12)-(13) or (16)-(17), that cont-selected evidence makes P(e_cont | C^c) higher and P(e_cont | C^w) lower; this unsupported monotone-likelihood-ratio premise is a correctness gap (and Eq. (21) as stated for every w is inconsistent with normalization), but it is not a case where the output equals the input by construction. The only self-citations ([33], [34]) describe implementation details of the optimization algorithm and are not load-bearing for the main theorem; the PAC bound in Theorem 3 is cited to external work [31]. No fitted parameter is renamed as a prediction. Therefore no definitional, fitting, or self-citation circularity is present.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The central claim rests on the Carnap-Hintikka inductive logic model with free parameters alpha and lambda(w), on the domain assumption that the true world is the minimal constituent, and on the unproven premise that cont-informative messages improve posterior concentration. No new entities are postulated.

free parameters (2)
  • lambda(w)
    Prior coefficient in the inductive logical probability (eq. 10) and likelihood (eq. 17). Its value is never specified or estimated; it changes all posterior and selection computations, so the reported gains depend on it.
  • alpha
    Prior belief parameter in the constituent prior (eq. 16) and the PAC bound (eq. 19). No value or fitting procedure is provided.
assumptions (5)
  • domain assumption The true state of the world is the minimal constituent C^c (Section III-E, 'Assume the minimal constituent C^c corresponds to the true state of the world.').
    This underpins the convergence analysis and Theorem 4; if the world is not the minimal constituent, the claimed domination may fail.
  • domain assumption The inductive logical probability model with prior (16) and likelihood (17) is a correct generative model for evidence.
    All posterior updates and message selections are computed from this model, but the paper provides no evidence that this model matches real data.
  • domain assumption Observations from different nodes are independent given the state of the world, so combined Bayesian updates are valid.
    The protocol assumes messages from different nodes can be aggregated via (18), which requires conditional independence.
  • standard math The PAC bound and asymptotic convergence of the posterior to the minimal constituent hold as stated in [31].
    The paper explicitly relies on [31] for convergence, saying the proof is omitted due to space.
  • ad hoc to paper Selecting messages that maximize cont-information reduces the observer's uncertainty about the true constituent.
    This is the premise of the protocol and the conclusion of Theorem 4, effectively assumed rather than proven.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DISCD: Distributed Lossy Semantic Communication for Logical Deduction of Hypothesis." pith.science (2026). https://pith.science/paper/TT2GZUVJ

@misc{pith2026250205744,
  author       = {Pith},
  title        = {Pith review of: DISCD: Distributed Lossy Semantic Communication for Logical Deduction of Hypothesis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TT2GZUVJ}},
  note         = {Machine review of arXiv:2502.05744}
}
read the original abstract

In this paper, we address hypothesis testing in a distributed network of nodes, where each node has only partial information about the State of the World (SotW) and is tasked with determining which hypothesis, among a given set, is most supported by the data available within the node. However, due to each node's limited perspective of the SotW, individual nodes cannot reliably determine the most supported hypothesis independently. To overcome this limitation, nodes must exchange information via an intermediate server. Our objective is to introduce a novel distributed lossy semantic communication framework designed to minimize each node's uncertainty about the SotW while operating under limited communication budget. In each communication round, nodes determine the most content-informative message to send to the server. The server aggregates incoming messages from all nodes, updates its view of the SotW, and transmits back the most semantically informative message. We demonstrate that transmitting semantically most informative messages enables convergence toward the true distribution over the state space, improving deductive reasoning performance under communication constraints. For experimental evaluation, we construct a dataset designed for logical deduction of hypotheses and compare our approach against random message selection. Results validate the effectiveness of our semantic communication framework, showing significant improvements in nodes' understanding of the SotW for hypothesis testing, with reduced communication overhead.

Figures

Figures reproduced from arXiv: 2502.05744 by the authors.

Figure 1
Figure 1. Distributed semantic communication framework. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Success rate in hypothesis deduction task vs. commu [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 31 canonical work pages

  1. [31]

    Knowledge, acceptance, and induc- tive logic

    Jaakko Hintikka and Risto Hilpinen. Knowledge, acceptance, and induc- tive logic. In Jaakko Hintikka and Patrick Suppes, editors, Aspects of Inductive Logic , volume 43 of Studies in Logic and the Foundations of Mathematics, pages 1–20. Elsevier, 1966

    work page 1966

  2. [1]

    Dhillon, Zhaohui Yang, Aylin Yener, Kai Kit Wong, and Chan-Byoung Chae

    Deniz G ¨und¨uz, Zhijin Qin, Inaki Estella Aguerri, Harpreet S. Dhillon, Zhaohui Yang, Aylin Yener, Kai Kit Wong, and Chan-Byoung Chae. Beyond transmitting bits: Context, semantics, and task-oriented communi- cations. IEEE Journal on Selected Areas in Communications, 41(1):5–41, 2023

    work page 2023

  3. [2]

    Choi, Seong-Lyun Kim, and Mehdi Bennis

    Jihong Park, Jinho D. Choi, Seong-Lyun Kim, and Mehdi Bennis. Enabling the wireless metaverse via semantic multiverse communication. 2023 20th Annual IEEE International Conference on Sensing, Communi- cation, and Networking (SECON) , pages 85–90, 2022

    work page 2023

  4. [3]

    Towards a decentralized metaverse: Synchronized orchestration of digital twins and sub-metaverses

    Omar Hashash, Christina Chaccour, Walid Saad, Kei Sakaguchi, and Tao Yu. Towards a decentralized metaverse: Synchronized orchestration of digital twins and sub-metaverses. ICC 2023 - IEEE International Conference on Communications , pages 1905–1910, 2022

    work page 2023

  5. [4]

    A two-dimensional continuum of inductive methods*

    Jaakko Hintikka. A two-dimensional continuum of inductive methods*. Studies in logic and the foundations of mathematics , 43:113–132, 1966

    work page 1966

  6. [5]

    Joint source- channel coding over additive noise analog channels using mixture of variational autoencoders

    Yashas Malur Saidutta, Afshin Abdi, and Faramarz Fekri. Joint source- channel coding over additive noise analog channels using mixture of variational autoencoders. IEEE Journal on Selected Areas in Commu- nications, 39(7):2000–2013, 2021

    work page 2000

  7. [6]

    Semantic communications for speech recognition

    Zhenzi Weng, Zhijin Qin, and Geoffrey Ye Li. Semantic communications for speech recognition. In 2021 IEEE Global Communications Conference (GLOBECOM), pages 1–6, 2021

    work page 2021

  8. [7]

    Goal-oriented communications for the iot and application to data compression

    Chao Zhang, Hang Zou, Samson Lasaulce, Walid Saad, Marios Koun- touris, and Mehdi Bennis. Goal-oriented communications for the iot and application to data compression. IEEE Internet of Things Magazine , 5(4):58–63, 2022

    work page 2022

Show all 34 references
  1. [8]

    The semantic com- munication game

    Bas ¸ak G¨uler, Aylin Yener, and Ananthram Swami. The semantic com- munication game. IEEE Transactions on Cognitive Communications and Networking, 4(4):787–802, 2018

    2018

  2. [9]

    Zhijin Qin, Xiaoming Tao, Jianhua Lu, and Geoffrey Y . Li. Semantic communications: Principles and challenges. ArXiv, abs/2201.01389, 2021

    2021 arXiv

  3. [10]

    Distributed variational repre- sentation learning

    Inaki Estella Aguerri and Abdellatif Zaidi. Distributed variational repre- sentation learning. IEEE transactions on pattern analysis and machine intelligence, 43(1):120–138, 2019

    2019

  4. [11]

    Distributed deep variational information bottleneck

    Abdellatif Zaidi and Inaki Estella Aguerri. Distributed deep variational information bottleneck. In 2020 IEEE 21st International Workshop on Signal Processing Advances in Wireless Communications (SPAWC), pages 1–5. IEEE, 2020

    2020

  5. [12]

    On the acceleration of deep learning model parallelism with staleness

    An Xu, Zhouyuan Huo, and Heng Huang. On the acceleration of deep learning model parallelism with staleness. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 2088–2097, 2020

    2020

  6. [13]

    Training recommender systems at scale: Communication-efficient model and data parallelism

    Vipul Gupta, Dhruv Choudhary, Ping Tak Peter Tang, Xiaohan Wei, Xing Wang, Yuzhen Huang, Arun Kejariwal, Kannan Ramchandran, and Michael W Mahoney. Training recommender systems at scale: Communication-efficient model and data parallelism. arXiv preprint arXiv:2010.08899, 2020

    2010 arXiv

  7. [14]

    Communication-efficient split learning based on analog communication and over the air aggregation

    Mounssif Krouka, Anis Elgabli, Chaouki ben Issaid, and Mehdi Bennis. Communication-efficient split learning based on analog communication and over the air aggregation. In 2021 IEEE Global Communications Conference (GLOBECOM), pages 1–6. IEEE, 2021

    2021

  8. [15]

    Li, and Biing-Hwang Juang

    Huiqiang Xie, Zhijin Qin, Geoffrey Y . Li, and Biing-Hwang Juang. Deep learning enabled semantic communication systems. IEEE Transactions on Signal Processing , 69:2663–2675, 2020

    2020

  9. [16]

    A theory of semantic communication

    Yulin Shao, Qingqing Cao, and Deniz G ¨und¨uz. A theory of semantic communication. ArXiv, abs/2212.01485, 2022

    2022 arXiv

  10. [17]

    Gangtao Xin, Pingyi Fan, and Khaled B. Letaief. Semantic communica- tion: A survey of its theoretical development. Entropy, 26(2), 2024

    2024

  11. [18]

    Reliable semantic communication system enabled by knowledge graph

    Shengteng Jiang, Yueling Liu, Yichi Zhang, Peng Luo, Kuo Cao, Jun Xiong, Haitao Zhao, and Jibo Wei. Reliable semantic communication system enabled by knowledge graph. Entropy, 24(6), 2022

    2022

  12. [19]

    Outline of a theory of strongly semantic information

    Luciano Floridi. Outline of a theory of strongly semantic information. SSRN Electronic Journal , 01 2004

    2004

  13. [20]

    Exk-sc: A semantic communication model based on information framework expansion and knowledge collision

    Gangtao Xin and Pingyi Fan. Exk-sc: A semantic communication model based on information framework expansion and knowledge collision. Entropy, 24(12), 2022

    2022

  14. [21]

    Vincent Poor

    Zhouxiang Zhao, Zhaohui Yang, Mingzhe Chen, Zhaoyang Zhang, and H. Vincent Poor. A joint communication and computation design for probabilistic semantic communications. Entropy, 26(5), 2024

    2024

  15. [22]

    Huiqiang Xie, Zhijin Qin, and Geoffrey Y . Li. Semantic communication with memory. IEEE Journal on Selected Areas in Communications , 41:2658–2669, 2023

    2023

  16. [23]

    Semantic-native communication: A simplicial complex perspec- tive

    Qiyang Zhao, Mehdi Bennis, M ´erouane Debbah, and Daniella Harth da Costa. Semantic-native communication: A simplicial complex perspec- tive. 2022 IEEE Globecom Workshops (GC Wkshps) , pages 1513–1518, 2022

    2022

  17. [24]

    Tilp: Differentiable learning of temporal logical rules on knowledge graphs

    Siheng Xiong, Yuan Yang, Faramarz Fekri, and James Clayton Kerce. Tilp: Differentiable learning of temporal logical rules on knowledge graphs. In The Eleventh International Conference on Learning Repre- sentations

  18. [25]

    Teilp: Time prediction over knowledge graphs via logical reason- ing

    Siheng Xiong, Yuan Yang, Ali Payani, James C Kerce, and Faramarz Fekri. Teilp: Time prediction over knowledge graphs via logical reason- ing. In Proceedings of the AAAI Conference on Artificial Intelligence , volume 38, pages 16112–16119, 2024

    2024

  19. [26]

    Temporal inductive logic reasoning over hypergraphs

    Yuan Yang, Siheng Xiong, Ali Payani, James C Kerce, and Faramarz Fekri. Temporal inductive logic reasoning over hypergraphs. In Pro- ceedings of the Thirty-Third International Joint Conference on Artificial Intelligence, pages 3613–3621, 2024

    2024

  20. [27]

    Harnessing the power of large language models for natural language to first-order logic translation

    Yuan Yang, Siheng Xiong, Ali Payani, Ehsan Shareghi, and Faramarz Fekri. Harnessing the power of large language models for natural language to first-order logic translation. In Lun-Wei Ku, Andre Martins, and Vivek Srikumar, editors, Proceedings of the 62nd Annual Meeting of th...

    2024

  21. [28]

    Large language models can learn temporal reasoning

    Siheng Xiong, Ali Payani, Ramana Kompella, and Faramarz Fekri. Large language models can learn temporal reasoning. In Lun-Wei Ku, Andre Martins, and Vivek Srikumar, editors, Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long...

    2024

  22. [29]

    An outline of a theory of semantic information

    Rudolf Carnap and Yehoshua Bar-Hillel. An outline of a theory of semantic information. 1952

    1952

  23. [30]

    Distributive normal forms in first-order logic

    Jaakko Hintikka. Distributive normal forms in first-order logic. In J.N. Crossley and M.A.E. Dummett, editors, Formal Systems and Recursive Functions, volume 40 of Studies in Logic and the Foundations of Mathematics, pages 48–91. Elsevier, 1965

    1965

  24. [32]

    Accessed: 2024-11-28

    https://github.com/ahmetfsaz/DISCD. Accessed: 2024-11-28

    2024

  25. [33]

    Lossy semantic communication for the logical deduction of the state of the world

    Ahmet Faruk Saz, Siheng Xiong, and Faramarz Fekri. Lossy semantic communication for the logical deduction of the state of the world. arXiv preprint arXiv:2410.01676, 2024

    2024 arXiv

  26. [34]

    Model-theoretic logic for mathematical theory of semantic infor- mation and communication

    Ahmet Faruk Saz, Siheng Xiong, Yashas Malur Saidutta, and Faramarz Fekri. Model-theoretic logic for mathematical theory of semantic infor- mation and communication. arXiv preprint arXiv:2401.17556 , 2024

    2024

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.