{"id":"ad0d9576-0ec4-4cd6-bdeb-4538c66d575e","arxiv_id":"2501.01635","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"high","formal_verification":"none","parameter_count":2,"one_line_summary":"A hybrid semantic/bit transmission scheme with proactive knowledge sharing is optimized to maximize the effective semantic rate in multi-cell networks under delay and accuracy constraints.","lead":"Small cell networks can fail at semantic communication when the phone and the base station do not share the same background knowledge. This paper proposes letting the phone upload the missing knowledge pieces, and then optimizing how much data is sent as meaning versus as ordinary bits to maximize the network's semantic rate.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The monotone-transform claim in §III-B2 is contradicted by the paper's own Table I semantic-accuracy parameters: with θ1=-6.205e-8 and θ4=-0.06917, Eq. (12) is strictly decreasing in ξ, so the Polyblock optimality and the ξ≥ξth interval collapse.","rationale":"The reader's weakest assumption is the monotonicity and parameter constraints of the semantic accuracy model. I independently computed the derivative of Eq. (12) and confirmed that the Table I parameters make ε strictly decreasing in ξ. This is load-bearing because the entire monotonic optimization transformation, the feasible interval ξth ≤ ξ ≤ 1, and the claimed optimality of the Polyblock algorithm depend on ε being increasing. The paper itself states the parameter constraint θi ≥ 0 and omits the derivative proof, so the inconsistency is internal, not a matter of disagreeing with an external consensus. The central algorithmic claim and the simulation results that rely on it are therefore unsupported as written. Since the reader already recommended REJECT with moderate confidence and my analysis identifies the same concern, no change to the verdict is needed.","tokens_in":21375,"tokens_out":3606,"duration_ms":35040,"concrete_test":"Evaluate Eq. (12) and its derivative at ξ=0 and ξ=1 using the Table I parameters (θ1=−6.205e−8, θ2=16.45, θ3=0.9228, θ4=−0.06917). If ε(0) > ε(1) and ε′(ξ) < 0, then the monotone-increasing assumption in §III-B2 is false; also check whether ξ ≥ ξth is feasible for the simulated εth values. This numerical check settles whether the Polyblock reduction applies to the model actually used in the paper.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section III-B2 reduces subproblem (21) to the monotonic problem (22) by asserting that εm,n(ξ) from Eq. (12) is monotonically increasing in ξ, with the derivative proof omitted. Eq. (12) gives ε′(ξ) = θ1θ2 e^{θ2(1−ξ)} + θ3θ4 e^{−θ4(1−ξ)}. The paper states θ1,θ2,θ3,θ4 ≥ 0, which would make this derivative nonnegative. However, Table I sets (θ1,θ2,θ3,θ4) = (−6.205e−8, 16.45, 0.9228, −0.06917), violating that constraint. With these values both derivative terms are negative, so ε is strictly decreasing on [0,1]. Consequently h(ξ) = R(Σ Im,k ε(ξ) + const) is decreasing, not increasing; constraint (14) yields an upper bound ξ ≤ ξth, not the lower bound ξ ≥ ξth used in (22d); and the Polyblock transformation, whose convergence and optimality require a monotone objective over the stated box, is unsupported. The claim that Algorithm 2 returns the optimum, and the simulation comparisons built on it, therefore do not follow from the model as written.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript studies a task-oriented semantic communication network in which mobile devices may share mismatched knowledge with a selected small-cell base station, and any unshared knowledge is served by conventional bit communication. Its system model leads to a mixed-integer nonlinear program (problem (16)) maximizing a generalized effective semantic transmission rate over the knowledge-sharing set KCu_{m,n}, the semantic extraction ratio ξ_{m,n}, and the SBS association x_{m,n}, subject to delay, accuracy, and cloudlet-capacity constraints. The authors decompose the problem into per-device/per-SBS KUER subproblems and a separate association problem. For each fixed knowledge-sharing set, the reduced one-variable subproblem (21) is transformed into a supposedly monotonic optimization problem (22)-(23) and solved by a Polyblock outer approximation (Algorithm 2); the outer enumeration over KCu is proposed as the optimum solver, and the association problem is solved by Kuhn-Munkres matching. The efficient algorithm replaces the outer enumeration by a two-tier class partitioning (Algorithm 1). Numerical experiments compare the efficient algorithm to the exhaustive optimum and to a no-knowledge-sharing baseline.","tokens_in":21686,"tokens_out":13563,"duration_ms":146090,"significance":"Should the derivations hold, the paper would offer a useful and well-structured treatment of a realistic KB-mismatch scenario, and the decomposition into per-pair optimization plus matching is a sensible architecture. The paper is also transparent in giving complete pseudo-code and a default-parameter table. However, the central optimality claim depends on the monotonicity of the semantic accuracy model and on the monotone structure of the feasible set; both are problematic. The accuracy model parameters used in the simulations contradict the model's own nonnegativity assumption, and the delay constraint is not monotone in ξ. As a result, the claimed optimality of Algorithm 2 and the simulation comparisons are not supported by the present manuscript.","major_comments":[{"comment":"The reduction to a monotonic optimization problem in Section III-B2 rests on the assertion that ε_{m,n}(ξ) in Eq. (12) is monotonically increasing in ξ, 'which can be proved by the first-order derivative of the function,' but the derivative is not shown. For Eq. (12) one obtains dε/dξ = θ1θ2 e^{θ2(1−ξ)} + θ3θ4 e^{−θ4(1−ξ)}. With the stated condition θ1,θ2,θ3,θ4 ≥ 0 this derivative is nonnegative, which is presumably the intended argument. However, Table I and the text of Section IV set (θ1,θ2,θ3,θ4) = (−6.205e−8, 16.45, 0.9228, −0.06917), which violates the stated nonnegativity constraint. With these values both terms of the derivative are negative, so ε is strictly decreasing on [0,1]; moreover ε(0) ≈ 1.85, violating the model property 0 ≤ ε ≤ 1 stated in Section II-D. Consequently, h(ξ) in problem (21) is decreasing in ξ, constraint (14) gives an upper bound ξ ≤ ξth rather than the lower bound ξ ≥ ξth used in (22d), and the Polyblock transformation and optimality claim in Algorithm 2 do not follow. Since Figures 5-10 are generated with this same parameter set, the simulation results do not support the paper's claims as written.","section":"III-B2, Eq. (12), Table I"},{"comment":"Even if ε were increasing, the claim that problem (21) can be transformed into a monotonic optimization problem is not justified, because the delay constraint (22c) is not monotone in ξ. With KCu_{m,n} fixed, the left-hand side of that constraint has the form α ξ + β ξ^{−ρ} with α,β > 0, where α comes from tS_{m,n} in Eq. (3) and β from tR_{m,n} in Eqs. (5)-(6). This function is convex and minimized at an interior point, so its feasible set is generally an interval [r1,r2] rather than a normal (downward-closed) set. After intersecting with ξ ≥ ξth, the feasible set need not be normal, and the projection operation used in Algorithm 3, whose binary search assumes that feasibility along the ray from vmin is prefix-monotone, can fail when ξth < r1. The paper does not establish the normality required for Polyblock outer approximation, so the convergence to a global optimum of problem (21) and hence the optimality of the exhaustive enumeration plus Polyblock procedure is unsupported even under the intended monotone ε model.","section":"III-B2, Eqs. (3), (5), (6), (21)-(23), Algorithm 3"}],"minor_comments":[{"comment":"The fitted model is written as ε′_{m,n}(ξ|θ1,θ2,θ3,θ4), but the prime is also used for derivatives elsewhere; please use a distinct symbol, for example ε̂ or ε̃, to avoid confusion.","section":"Eq. (12)"},{"comment":"The text states that each MD associates with one and only one SBS, but constraint (10) is ∑_n x_{m,n} ≤ 1, which allows an MD to remain unassociated. The matching formulation in Section III-C should clarify whether unassociated MDs are permitted and how the bipartite graph handles the capacities Smax_n.","section":"II-C"},{"comment":"The parameters (θ1,θ2,θ3,θ4) are given only in the text rather than in Table I; placing them in the table would make the inconsistency with the stated nonnegativity constraint in Eq. (12) immediately visible.","section":"IV"},{"comment":"There are several typographical errors, including 'semenatic' in the Fig. 1 caption, 'intergrading' in the introduction, and 'complimental' in Section II-B; these should be corrected.","section":"General"},{"comment":"The simulation 'validity' claim rests on comparing the proposed algorithms with each other and with a no-knowledge-sharing scheme under the same approximate ε model; an external benchmark or real model-evaluation data would be needed to substantiate the physical accuracy of the semantic accuracy model.","section":"IV"}],"recommendation":"reject","confidential_remarks":"The stress-test concern is valid and, in my reading, under-sells the problem: the monotone reduction in Section III-B2 also fails because the delay constraint is not monotone in ξ, independent of the signs of the θ parameters. The two issues together mean that both the optimality proof and the simulation evidence are unsupported as written. I would be open to a substantially revised manuscript that corrects the semantic accuracy model, re-derives the monotone transformation or solves the one-dimensional problem directly, and reruns all simulations with valid parameters."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: the joint formulation is new and the decomposition is mostly sensible, but the paper has a load-bearing contradiction in its semantic accuracy model. The monotonicity claim in §III-B2 is asserted without proof, and the authors' own Table I parameters falsify it.\n\nThey define ε(ξ) = -θ1 exp(θ2(1−ξ)) + θ3 exp(−θ4(1−ξ)) and state θ1,θ2,θ3,θ4 ≥ 0. With that, ε′(ξ) = θ1θ2 exp(θ2(1−ξ)) + θ3θ4 exp(−θ4(1−ξ)) ≥ 0, so ε is increasing. Table I sets (θ1,θ2,θ3,θ4) = (−6.205e−8, 16.45, 0.9228, −0.06917). Two of those are negative. Plugging in, both terms of ε′ are negative, so ε is strictly decreasing on [0,1]. That inverts constraint (14), turning ξ ≥ ξth into ξ ≤ ξth in (22d), and makes h(ξ) decreasing. The Polyblock method in Algorithm 2 relies on a monotone objective over the box; without it, the claimed optimality and the \"close-to-optimum\" simulation comparisons are unsupported.\n\nCredit where it's due. The system model is carefully built, and the decomposition into per-pair KUER subproblems plus an SBS-association matching is a reasonable way to attack the MINLP. The generalized rate metric and the joint optimization of knowledge sharing, extraction ratio, and association do appear to be new relative to the cited literature; the two-tier class partitioning heuristic is a plausible low-complexity design. The paper is not sloppy overall — the error is localized to the parameter assumptions and the omitted derivative check.\n\nSecondary soft spots: the simulations validate against the same fitted ε model used in the derivation, with no independent accuracy benchmark, so \"validity\" is really self-consistency. And several parameter values are imported from [26] without checking whether they respect the model's own θ ≥ 0 constraint. Both are fixable.\n\nWho this is for: researchers working on task-oriented semantic communication resource allocation. The formulation is worth having in the literature once the accuracy model is corrected and the simulations rerun. As submitted, I would not cite it.\n\nRecommendation: send it to peer review — the novelty and scope justify referee time and the core flaw is identifiable and repairable — but the authors should be asked to fix the monotonicity argument, respect their parameter constraints, and rerun the evaluation. This is a major-revision situation, not a desk reject.","headline":"A solid new formulation undercut by a load-bearing sign error: the paper's own simulation parameters make its semantic accuracy model decreasing in ξ, so the monotonic-optimization and optimality claims in §III-B2 do not hold as written.","tokens_in":22140,"tokens_out":4419,"would_cite":false,"duration_ms":42377,"reading_group":"maybe","serious_thinker":"no","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Jointly choosing which knowledge to share, how much to compress, and which base station to serve each device maximizes the generalized effective semantic transmission rate in multi-cell hybrid semantic/bit networks.","keywords":["semantic communications","knowledge sharing","hybrid semantic-bit networks","task-oriented communications","semantic extraction ratio","multi-cell networks","monotonic optimization","SBS association"],"falsifier":"Compute Eq. (12) with the Table I parameters $\\theta_1=-6.205\\times 10^{-8}$, $\\theta_2=16.45$, $\\theta_3=0.9228$, $\\theta_4=-0.06917$ at $\\xi=0.5$, $0.75$, and $1$; if $\\varepsilon(0.5)>\\varepsilon(1)$, then the accuracy curve decreases over the range the optimization uses, and the monotonic transformation plus the $\\xi\\ge\\xi^{\\mathrm{th}}$ reduction would need a different justification.","tokens_in":21186,"feed_emoji":"📡","tokens_out":12953,"duration_ms":112496,"temperature":0.7,"pith_summary":"This paper proposes a transmission mechanism in which a mobile device first uploads the task-related knowledge its chosen base station is missing, and then sends data that relies on that shared knowledge through semantic communications while sending the rest as ordinary bits. The authors claim that jointly optimizing three decisions — which mismatched knowledge classes to share, how aggressively to extract semantic information (the semantic extraction ratio), and which small base station to associate with — maximizes the total effective semantic transmission rate of all devices, while respecting each task's semantic accuracy and delay requirements. The resulting mixed-integer nonlinear program is decomposed into per-pair knowledge-and-extraction subproblems and an association subproblem, with the per-pair part solved by enumeration plus monotonic optimization and the association part solved exactly as weighted bipartite matching. A cheaper two-tier partitioning heuristic is shown in simulation to nearly match the optimum and to outperform a baseline that shares no knowledge. If correct, the paper gives a concrete resource-allocation rule for task-oriented semantic networks operating with mismatched knowledge bases.","feed_headline":"Joint choice of sharing, compression, and cell maximizes semantic rate","feed_subtitle":"Multi-cell hybrid scheme with proactive knowledge sharing beats no-sharing baselines in simulation.","key_machinery":"The load-bearing object is the generalized effective semantic transmission rate $\\gamma_{m,n}$ defined in (15), a ratio of useful semantic information (semantically recovered data weighted by accuracy $\\varepsilon_{m,n}(\\xi_{m,n})$ plus bit-transmitted data for unshared knowledge classes) to total transmission time. The argument runs on the monotonicity of the semantic accuracy function $\\varepsilon_{m,n}(\\xi_{m,n})$ of Eq. (12): with a monotone curve, the reduced subproblem in $\\xi_{m,n}$ becomes a monotonic optimization problem (22)–(23), solvable globally by Polyblock outer approximation, and the accuracy constraint (14) collapses to the lower bound $\\xi_{m,n} \\ge \\xi^{\\mathrm{th}}_{m,n}$. The second piece is the two-tier class partitioning heuristic (Algorithm 1), which sorts mismatched knowledge classes by the data-benefit ratio $\\Phi^B_{m,k} = d^T_{m,k}/d^K_{m,k}$ and then by the time-ratio $\\Phi^D_{m,n,k}$, turning exponential enumeration over sharing subsets into a linear search over prefix sets. The third piece maps the SBS association subproblem to weighted bipartite matching, solved optimally by the modified Kuhn-Munkres algorithm.","core_discovery":"On its own terms, the paper establishes a rate-maximization formulation and a solution method. The generalized effective semantic transmission rate $\\gamma_{m,n}$ counts semantically transmitted information weighted by accuracy plus bit-transmitted information, normalized by total transmission time, so the hybrid mechanism is evaluated as one integrated metric. The paper claims that the optimal operating point is found by jointly setting the knowledge sharing set $K^{\\mathrm{Cu}}_{m,n}$, the semantic extraction ratio $\\xi_{m,n}$, and the binary association $x_{m,n}$; no single one of these suffices. For a fixed sharing set, the reduced problem becomes monotonic in $\\xi_{m,n}$ and is solved globally by Polyblock outer approximation; for fixed per-pair solutions, the association problem becomes a maximum-weight matching and is solved optimally by the modified Kuhn-Munkres algorithm. Simulations then support the efficiency claim for the two-tier heuristic and the benefit of knowledge sharing over the no-sharing baseline.","pith_inferences":["The same per-pair-plus-matching decomposition would likely extend to other task-oriented objectives, such as energy consumption or task success probability, because the rate objective and the SBS-association layer separate cleanly.","The prefix structure of the two-tier partitioning suggests an online policy: as channel rates vary, an MD could decide which knowledge classes to upload in order of data-benefit ratio, re-optimizing the cutoff each coherence interval.","Because the objective counts bit-transmitted data equally with semantically transmitted data, re-weighting the two terms by task importance or energy cost would shift the optimal sharing set and stress-test how robust the policy is.","Because the Polyblock optimality proof is built on monotonicity of the accuracy curve, treating the fitted curve as non-monotone would be a direct stress test of the global-optimality claim."],"forward_implications":["The association layer can be solved optimally by weighted bipartite matching once each MD–SBS pair's best sharing set and extraction ratio are known, so the hard combinatorial part is confined to the per-pair subproblems.","The proposed two-tier heuristic matches the optimum closely in small networks, and the gap shrinks when cloudlet computation capacity or delay tolerance grows.","Knowledge sharing pays off most when knowledge data are small relative to the raw data they unlock, and the benefit disappears once knowledge sizes grow large, giving a practical threshold for switching to pure bit transmission.","The total semantic rate saturates as delay tolerance, bandwidth, or cloudlet count grows, so provisioning beyond those saturation points yields little additional benefit."],"supporting_citations":[{"why":"Defines the no-knowledge-sharing baseline and the KB-mismatch setting that the proposed mechanism is compared against.","marker":"[16]"},{"why":"Supplies the semantic accuracy model, the nonlinear least-squares fitting method, and the default model parameters used in the simulations.","marker":"[26]"},{"why":"Provides the monotonic optimization and Polyblock outer approximation framework that solves the reduced subproblem optimally.","marker":"[37]"},{"why":"Demonstrates the application of the Polyblock algorithm to communication resource allocation, cited alongside [37].","marker":"[38]"},{"why":"Gives the modified Kuhn-Munkres algorithm that solves the SBS association subproblem as weighted bipartite matching.","marker":"[35]"},{"why":"Provides the prior hybrid semantic/bit network model with mode selection that this work extends to multi-cell scenarios with knowledge sharing.","marker":"[14]"},{"why":"Introduces the semantic unit (sut) used to define the generalized effective semantic transmission rate.","marker":"[33]"},{"why":"Supplies the linear-search task-class partitioning idea underlying the two-tier heuristic.","marker":"[36]"}],"fun_headline_variants":["Knowledge sharing unlocks higher semantic rate in multi-cell hybrid nets","Joint design of sharing, extraction, and cell association maximizes semantic rate","Proactive knowledge sharing beats no-sharing in semantic-bit hybrid networks","Optimal sharing, compression, and association boost semantic rate in multi-cell","Multi-cell semantic rate maximized via knowledge sharing and hybrid bits"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"Everything hinges on the assumption that semantic accuracy rises with the extraction ratio $\\xi$, a property the paper states for its fitted curve but that its own Table I parameters contradict because two of the fitted coefficients are negative, which makes accuracy fall as extraction rises.","fun_headline_variants_meta":{"raw":{"variants":["Knowledge sharing unlocks higher semantic rate in multi-cell hybrid nets","Joint design of sharing, extraction, and cell association maximizes semantic rate","Proactive knowledge sharing beats no-sharing in semantic-bit hybrid networks","Optimal sharing, compression, and association boost semantic rate in multi-cell","Multi-cell semantic rate maximized via knowledge sharing and hybrid bits"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000859,"raw_usage":{"total_tokens":3740,"prompt_tokens":968,"completion_tokens":2772,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":584,"completion_tokens_details":{"reasoning_tokens":2694}},"tokens_in":584,"tokens_out":2772,"duration_ms":18567,"temperature":1.0,"reasoning_tokens":2694,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T22:23:48.034172+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute Eq. (12) with the Table I parameters $\\theta_1=-6.205\\times 10^{-8}$, $\\theta_2=16.45$, $\\theta_3=0.9228$, $\\theta_4=-0.06917$ at $\\xi=0.5$, $0.75$, and $1$; if $\\varepsilon(0.5)>\\varepsilon(1)$, then the accuracy curve decreases over the range the optimization uses, and the monotonic transformation plus the $\\xi\\ge\\xi^{\\mathrm{th}}$ reduction would need a different justification.","supporting_citations":[{"cited_title":"Adaptable semanti c compres- sion and resource allocation for task-oriented communicat ions,","cited_arxiv_id":null,"evidence_quote":"Supplies the semantic accuracy model, the nonlinear least-squares fitting method, and the default model parameters used in the simulations."},{"cited_title":"Monotonic optimization in communication and networking systems,","cited_arxiv_id":null,"evidence_quote":"Provides the monotonic optimization and Polyblock outer approximation framework that solves the reduced subproblem optimally."},{"cited_title":"Resourc e alloca- tion for multi-user downlink MISO OFDMA-URLLC systems,","cited_arxiv_id":null,"evidence_quote":"Demonstrates the application of the Polyblock algorithm to communication resource allocation, cited alongside [37]."},{"cited_title":"Bipartite matching based user grouping for grouped ofdm-idma,","cited_arxiv_id":null,"evidence_quote":"Gives the modified Kuhn-Munkres algorithm that solves the SBS association subproblem as weighted bipartite matching."},{"cited_title":"Task cla ss par- titioning for mobile computation offloading,","cited_arxiv_id":null,"evidence_quote":"Supplies the linear-search task-class partitioning idea underlying the two-tier heuristic."}],"review_version":1}