{"id":"578a7767-7ca4-4cdd-9874-197dc3bc1d14","arxiv_id":"2607.16562","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":7,"one_line_summary":"AirMoE routes clients via compact prototype statistics and aggregates expert outputs over the air by transmitting them simultaneously, claiming communication savings and better segmentation accuracy.","lead":"A wireless-friendly 'over-the-air' mixture-of-experts system is proposed, where client experts transmit their outputs simultaneously so the radio channel itself computes the weighted sum, and clients report compact statistics instead of raw features for routing. The paper claims this reduces uplink traffic and aggregation latency, with convergence guarantees and gains on semantic segmentation benchmarks.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Experiments omit the wireless channel, so Table II does not validate the over-the-air aggregation claim.","rationale":"The reader's weakest_assumption focuses on perfect CSI and synchronization, which the authors themselves acknowledge as a limitation. My concern is more elementary: the experimental section is compatible with no channel model at all, so the distinctive over-the-air mechanism is empirically unvalidated. This reinforces the REJECT verdict, but through a different route than the stated weakest_assumption. The theory's smooth-surrogate analysis and missing appendices are additional concerns, but the absent channel in the experiments is the most load-bearing gap for the central empirical claim.","tokens_in":27996,"tokens_out":6644,"duration_ms":75675,"concrete_test":"Reproduce the Cityscapes AirMoE configuration from Table II with an explicit channel: i.i.d. Rayleigh fading per client, per-client power P_0, receive SNR 10 dB (so σ_c^2 is set from P_0 and path loss), truncation γ_th = 0.1, perfect CSI, and the Eq. (22)–(23) power control. Report mIoU/mF1 for this noisy over-air run and for the same AirMoE with noiseless aggregation. If the noiseless run reproduces Table II's 44.74/55.99 and the noisy run drops by more than the margin to the best baseline (44.74 vs. 42.08 for SoftMoE), or if the authors cannot provide channel parameters for the original Table II, then the over-the-air empirical claim is not established.","verdict_should_be":"UNCHANGED","load_bearing_attack":"AirMoE's central claim is that Eq. (23) lets the multiple-access channel compute the statistically reweighted MoE fusion in one transmission, and that this system outperforms MoE baselines on four datasets (Table II). The algebraic identity in Eq. (23) is valid under ideal alignment in Eq. (22), but the empirical evidence for the system-level claim does not test the channel. Section VI-A3 lists datasets, metrics, ViT/ASPP architecture, optimizer, hyperparameters, and baselines, but nowhere specifies a channel model: no path-loss/fading distribution, no noise power σ_c^2, no truncation threshold γ_th, no per-client power P_0, no receive scaling, and no CSI error model. Algorithm 1 includes the over-the-air step (lines 10–12), but Table II contains no SNR or noise column. If the experiments used an ideal noiseless channel — or no channel at all — Table II validates only the prototype-statistic routing/FRL and reciprocal-JS weighting; the distinctive over-the-air aggregation is untested. The paper's own Lemma 4 and Fig. 5 state that channel noise creates an irreducible accuracy floor, so nonzero noise must degrade segmentation performance. Without reporting the SNR under which the 44.74 mIoU was obtained, the claimed superiority over SoftMoE/NonLinearMoE cannot be attributed to the wireless aggregation mechanism. The self-acknowledged perfect-CSI idealization is secondary; even under perfect CSI, the channel must be simulated to support the headline claim.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes AirMoE, a cloud-edge wireless Mixture-of-Experts framework. A cloud-hosted frozen large model extracts latent features and broadcasts a compact query to clients; each client maintains a Feature Retrieval Library (FRL) of prototypes and reports a normalized prototype-induced statistic. The cloud selects the TopK clients by Jensen–Shannon divergence and computes reciprocal-distance aggregation weights. Selected clients then transmit their outputs simultaneously over a multiple-access channel with channel-inversion power control, so that the channel superposition approximates the weighted MoE fusion in one analog transmission. The authors provide convergence and ε-stationarity analysis under channel noise, with an explicit noise-induced floor, and report semantic segmentation experiments on Cityscapes, CamVid, Apolloscapes, and CARLA ADV, comparing against MoE and single-model baselines.","tokens_in":28412,"tokens_out":5949,"duration_ms":72910,"significance":"If the empirical claims were fully supported, the paper would make a useful contribution: it identifies a natural mapping from MoE weighted aggregation to over-the-air computation, achieves aggregation latency and bandwidth independent of the number of selected experts, and gives an explicit decomposition of channel-noise effects on optimization (variance inflation and an irreducible stationarity floor). The algebraic construction in Eqs. (22)–(23) is clean, and Lemma 1’s decomposition σ_eff² = σ² + κσ_c²/(2ρ_min) is instructive. These are genuine strengths. However, the central experimental validation is currently missing the wireless channel, and the theoretical claims contain some gaps. The significance is therefore conditional on repair of these load-bearing issues.","major_comments":[{"comment":"The experimental setup never specifies the wireless channel. There is no path-loss/fading model, no noise power σ_c², no receiver noise, no per-client power budget P0, no truncation threshold γ_th, no channel-inversion pre-scaling in the simulation, and no CSI error model. Algorithm 1 includes over-the-air steps (lines 10–12), but Table II reports no SNR, noise level, or channel realization. Since Lemma 4 and Fig. 5 state that channel noise creates an irreducible accuracy floor, the reported 44.74 mIoU on Cityscapes, and the claimed gains over SoftMoE/NonLinearMoE, cannot be attributed to the over-the-air aggregation mechanism. The experiments as presented validate only the FRL-based routing and JS weighting; the distinctive AirComp component is untested.","section":"Section VI-A3 and Table II"},{"comment":"The perfect-CSI and ideal-synchronization assumption is acknowledged as a limitation, but it is load-bearing for the main mechanism. Eq. (22) requires b_j = sqrt(ρ)β_j/γ_j with exact knowledge of γ_j and no phase/alignment error. If CSI is imperfect or synchronization is imperfect, the received signal is no longer the intended weighted sum, Eq. (23) fails, and the entire physical-layer aggregation claim collapses. The paper provides no sensitivity analysis with respect to CSI error or synchronization offset, even in simulation. This is not a minor caveat; it is the central mechanism of the system.","section":"Section VII and Eq. (22)"},{"comment":"Theorem 2 fixes the FRL state M, whereas Algorithm 1 updates M online via Eqs. (11)–(12). Remark 1 states that for η small relative to α_t, M can be treated as quasi-static, but no two-timescale convergence theorem is proved; only boundedness of M is established in Theorem 1. As written, the theory does not cover the algorithm that is implemented. The authors should either provide a genuine two-timescale convergence analysis or explicitly restrict the theorem to the fixed-M setting and soften the claim that it guarantees convergence of AirMoE.","section":"Theorem 2 and Remark 1"},{"comment":"The claim that an ε-stationary point of the noiseless objective is reachable iff ε > ε_floor = 2B_σ² is not justified by the displayed argument. The inequality ‖∇eL‖² ≤ 2‖∇Lbar‖² + 2B_σ² is a one-sided upper bound; it implies a sufficient condition for reaching ε-stationarity of eL and the absence of a guarantee below the floor, but it does not prove that no finite T can reach a point with ‖∇eL‖² ≤ ε for ε ≤ ε_floor. The proof in Appendix G would need to establish a matching lower bound on the gradient bias to support the word \"iff\" and the \"irreducible floor\" interpretation. Otherwise the statement should be weakened to a worst-case/guarantee-based claim.","section":"Lemma 4 and Eq. (37)"}],"minor_comments":[{"comment":"The derivation implicitly assumes the expert outputs y_j are real-valued, or that taking the real part after receiver scaling projects onto the signal space without loss. This should be stated explicitly.","section":"Eq. (23)"},{"comment":"The comparison protocol is reasonable for MoE baselines (same frozen ViT, same expert architecture, same training protocol), but the single-model competitors are trained from scratch while the MoE models use a pretrained frozen ViT. This makes the MoE-vs-single-model comparison uneven; the claim that MoE-based models are inherently better should be tempered.","section":"Section VI-A3"},{"comment":"Fig. 5 is presented as the relationship between iteration complexity and channel parameters, but the parameter values (σ_c², ρ_min) are abstract. Since no simulation settings are given, the figure is purely illustrative. This is acceptable, but the caption should say so explicitly.","section":"Fig. 5 and Section VI-C"},{"comment":"References [45] and [54] point to the same arXiv preprint (arXiv:2405.14131) and should be merged or distinguished. Several references also lack venue details.","section":"References"},{"comment":"Line 12 writes \"Simultaneous transmission ⇒ r^(i)\" but nowhere in the algorithm is the channel realization, noise power, or fading model instantiated. This reflects the same missing wireless simulation as in VI-A3.","section":"Algorithm 1"}],"recommendation":"major_revision","confidential_remarks":"This is a borderline case. The paper has a timely and plausible idea, and the mathematical skeleton is mostly standard. The decisive deficiency is that the experiments never simulate the wireless channel, so the paper's main distinguishing claim—over-the-air aggregation—is not empirically validated. This is fixable with a properly specified channel model, noise level, power budget, truncation threshold, and preferably a sensitivity study around CSI error. The theoretical gaps concerning fixed M and the Lemma 4 \"iff\" should also be addressed. I would be willing to review a revision that repairs these points."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"I read AirMoE with the reader's report in hand, and I side with the reader on the central issue. The paper's genuinely new idea is to map over-the-air computing onto MoE aggregation: the channel physically computes the reweighted sum, and routing uses compact prototype statistics instead of raw features. That combination is not in the cited literature, and decoupling digital routing from analog aggregation is a sensible framing. The convergence analysis is also honest: it is standard nonconvex SGD for a noise-smoothed surrogate, and the bias/floor decomposition in Lemma 4 is a clean way to expose how channel noise creates an irreducible accuracy floor. The authors even flag in the conclusion that they only consider an idealized analog front-end. The soft spot is not the algebra--Equation (23) is correct under ideal alignment--it is that the experiments never specify the wireless channel. No noise power, no fading model, no truncation threshold, no power budget, no SNR. Table II therefore validates the routing/FRL mechanism and the reciprocal-JS weighting under what amounts to an ideal channel, but it does not validate the over-the-air aggregation claim. The paper's own Lemma 4 says nonzero channel noise degrades attainable accuracy, so without reporting the SNR under which the 44.74 mIoU was obtained, the claimed superiority over SoftMoE cannot be attributed to the wireless mechanism. This is a load-bearing gap, not a cosmetic omission. Two smaller issues. The main configuration (10 experts, TopK=5) is worse than the paper's own ablations (TopK=3 gives 45.85; 7 experts gives 45.45), so the reported headline numbers are not the best the method achieves, and this is never explained. And all proofs are in missing appendices; the theory is for a smooth surrogate with fixed FRL state, which is acknowledged but leaves a real gap between the analyzed object and the actual TopK algorithm. Where I would push back on a harsh reading: this is not a fabricated result or a circular argument. The theory has no fitted constants, the FRL stability theorem is plausible, and the authors openly state the perfect-CSI limitation. The problem is that the empirical evidence does not yet test the thing the paper is named after. Who should read this: people working on wireless edge inference or AirComp-based distributed learning will find the architecture worth discussing, and the paper should go to peer review rather than desk reject. It needs a major revision: simulate the channel with explicit SNR and fading, include the proofs, and justify the hyperparameter choice. If that is done, the core idea could become a solid contribution.","headline":"Novel mapping of AirComp onto MoE aggregation, but the experiments never simulate the channel, so the headline accuracy claims are unsupported.","tokens_in":719,"tokens_out":727,"would_cite":false,"duration_ms":30400,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper claims that a Mixture-of-Experts head scattered across wireless clients can be routed by compact statistics and aggregated physically by the channel itself, making MoE fusion latency and bandwidth independent of the number of exp","keywords":["Mixture of Experts","over-the-air computing","edge intelligence","semantic segmentation","feature retrieval library","Jensen-Shannon divergence","wireless aggregation","distributed inference"],"falsifier":"A concrete test: run AirMoE with channel estimation errors (e.g., Gaussian perturbation to gamma_j) or with imperfect timing synchronization, and measure the received sum's deviation from the target MoE weighted sum. If the deviation grows beyond the noise floor and downstream segmentation quality drops sharply, the core over-the-air aggregation claim fails under realistic channel conditions.","tokens_in":27896,"feed_emoji":"📡","tokens_out":1573,"duration_ms":21846,"temperature":0.7,"pith_summary":"AirMoE is a method for running a Mixture-of-Experts model across a cloud server and many wirelessly connected edge devices, where the cloud hosts a frozen large backbone and each device hosts a specialized expert. The paper argues that the two classic MoE operations—routing (which experts to use) and aggregation (how to fuse their outputs)—become prohibitive communication bottlenecks in this setting, and that both can be redesigned. Routing is replaced by each client uploading a tiny prototype-derived statistic instead of raw features, and aggregation is performed in one shot by having selected experts transmit simultaneously, letting the wireless channel superimpose their power-scaled outputs into the desired weighted sum. The paper further proves convergence guarantees for the channel-perturbed training and reports that AirMoE outperforms both MoE baselines and single-model competitors on semantic segmentation across four datasets. The central insight is that the analog multiple-access channel itself can compute the MoE fusion, turning a communication bottleneck into a computational resource.","feed_headline":"Channel computes the MoE fusion in one transmission","feed_subtitle":"Routing via compact statistics and physical-layer aggregation keep cost independent of expert count.","key_machinery":"The key mechanism is the over-the-air aggregation identity: by setting the transmit pre-scaling b_j = sqrt(rho) beta_j / gamma_j, the received signal, after projection and receiver scaling, equals the MoE weighted sum plus zero-mean real-equivalent noise. This turns waveform superposition into the MoE fusion operator, and makes aggregation cost invariant to the number of activated clients. The supporting machinery is the Feature Retrieval Library (FRL)—each client keeps a small set of prototype vectors updated by attention-based read-then-update—and the JS-divergence routing score that selects clients in a shared normalized distribution space, keeping routing digital and low-rate while aggre","core_discovery":"The paper's central claim is that over-the-air computing can realize the statistically reweighted MoE fusion in a single transmission. With channel-inversion power control, each activated client transmits its expert output pre-scaled by its aggregation weight divided by its channel coefficient; the receiver then applies one common scaling, and the superposition of waveforms yields exactly the intended weighted sum plus a zero-mean noise term. The paper combines this physical-layer aggregation with a digitized, statistic-based routing mechanism: each client stores a Feature Retrieval Library of prototypes, retrieves a prototype-induced statistic from a cloud-broadcast compact query, reports i","pith_inferences":["An untested but plausible extension is to apply AirMoE's statistic-based routing and over-the-air aggregation to other tasks such as object detection or multi-label classification, where the same weighted-sum fusion structure appears.","A testable prediction of the paper's theory is that increasing the output dimension of experts raises the stationarity floor quadratically; this implies that reducing output dimensionality (e.g., via low-rank heads or sparsification) should be prioritized over merely reducing noise to improve over-the-air training accuracy.","The paper's reliance on perfect channel state information suggests that an empirical study of AirMoE under realistic channel estimation errors would likely show graceful degradation up to a point, but a hard breakdown once the alignment constraint fails; this is a direction the authors explicitly flag for future work."],"forward_implications":["If AirMoE is correct, a distributed MoE head can be served over bandwidth-limited wireless links with aggregation cost that does not scale with the number of experts, enabling larger expert pools on edge devices.","The convergence analysis shows that channel noise inflates gradient variance linearly but imposes a squared floor on achievable stationarity; this makes the communication-computation tradeoff quantitative and suggests power control and receive-SNR floor as primary design levers.","The statistic-based routing removes the need to upload raw latent features for routing decisions, which could be applied to other distributed inference tasks beyond segmentation where feature transmission is the bottleneck.","The adversarial-condition results (e.g., gains concentrated on low-mIoU datasets) suggest that AirMoE's benefit grows with task difficulty and expert heterogeneity, pointing to its use in diverse, real-world edge deployments."],"fun_headline_variants":["Wireless channel merges MoE experts in one shot","AirMoE: statistics route, channel fuses experts","MoE without uplink flood: stats + waveform sum","One air transmission computes MoE fusion"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The entire aggregation mechanism depends on each client knowing its own uplink channel coefficient exactly, without phase errors or synchronization mismatch, so that its transmitted signal arrives aligned and the received super-position is exactly the intended weighted sum.","fun_headline_variants_meta":{"raw":{"variants":["Wireless channel merges MoE experts in one shot","AirMoE: statistics route, channel fuses experts","MoE without uplink flood: stats + waveform sum","One air transmission computes MoE fusion"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000188,"raw_usage":{"total_tokens":1207,"prompt_tokens":820,"completion_tokens":387,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":564,"completion_tokens_details":{"reasoning_tokens":336}},"tokens_in":564,"tokens_out":387,"duration_ms":5090,"temperature":1.0,"reasoning_tokens":336,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T20:34:21.832414+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A concrete test: run AirMoE with channel estimation errors (e.g., Gaussian perturbation to gamma_j) or with imperfect timing synchronization, and measure the received sum's deviation from the target MoE weighted sum. If the deviation grows beyond the noise floor and downstream segmentation quality drops sharply, the core over-the-air aggregation claim fails under realistic channel conditions.","supporting_citations":[],"review_version":1}