{"id":"d24cdcd4-f471-4492-b06d-ace49f447a54","arxiv_id":"2607.07925","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"K-Means cluster-weighted training cuts average MSE 34% and lifts R² from 0.54 to 0.80 for a 3D CNN predicting Reynolds stresses on StellarBox quiet-Sun data.","lead":"A CNN trained with K-Means cluster-weighted loss predicts Reynolds stress components from quiet-Sun simulations 34% more accurately than uniform training. The technique could cheapen high-fidelity solar MHD runs by improving subgrid surrogates.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.5","headline":"The 34% MSE / R^{2} gains rest on a single early-error snapshot (T0=10) and a fixed α=3 that is never ablated, so the reported improvement may be an artifact of that particular weighting schedule rather than a robust property of cluster-weighted training.","rationale":"The Reader correctly isolates the early-error weighting step (T0=10, α=3) as the weakest assumption supporting the strongest claim. That assumption is load-bearing: every reported metric improvement flows from the single weight vector produced by that schedule. Because the paper supplies neither an ablation of those two knobs nor a non-clustered reweighting control, the causal attribution to “cluster-weighted training” remains provisional. The concrete grid search above would settle the issue with modest compute and would either confirm robustness or force a more cautious claim. No stronger internal inconsistency is present; the experimental design is otherwise transparent and the physics baselines are properly reported. Hence the Reader’s CONDITIONAL verdict is unchanged, only sharpened by an explicit falsification test.","tokens_in":10729,"tokens_out":619,"duration_ms":7808,"concrete_test":"Re-train the identical CNN architecture under the same data split for a 3\times3 grid of (T0,α) ∈ {5,10,20}\times{1,3,5}, plus one control that assigns the same set of weight values randomly across samples. If the >30% MSE reduction and R^{2}\to0.80 appear only for the published (10,3) pair (or vanish under random reweighting), the headline improvement is schedule-specific and the claim must be narrowed.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim (Table II: average MSE 4.89e18 \to 3.01e18, R^{2} 0.54 \to 0.80) is produced by the exact procedure in Section V-D / Fig. 4: after only T0=10 epochs of ordinary MSE the per-cluster mean squared errors are frozen into permanent sample weights wk=1+α·(MSEk/max MSEj) with the single hand-chosen α=3. No sensitivity study of T0 or α appears, nor is there a control that reweights by a random or uniform-within-cluster schedule of the same magnitude. Consequently it is possible that any sufficiently aggressive reweighting of the high-error tail after a short warm-up would produce a comparable lift, or that a different (T0,α) pair would erase the gain. The claim that “cluster-weighted training” itself is responsible therefore rests on an untested hyper-parameter choice rather than on a demonstrated property of the clustering.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The paper introduces cluster-weighted training for a 3D CNN surrogate that predicts the six Reynolds-stress components τ ij from local macroscopic velocities and density, using high-resolution StellarBox quiet-Sun cubes. After a short uniform-MSE warm-up (T0=10 epochs), K-Means (K=5) partitions of the 82-dimensional input features supply permanent sample weights wk=1+α·(MSEk/max MSEj) with α=3; the CNN is then retrained under the weighted loss. Relative to the identical architecture trained with uniform MSE, average MSE falls from 4.89\times10^18 to 3.01\times10^18 (≈34 %) and average R² rises from 0.54 to 0.80 (Table II), with clearer PDF tails, tighter error histograms and scatter plots (Figs. 5–7) and localized gains in selected clusters and height ranges (Fig. 8). Limited HAC and physics-baseline comparisons are also shown.","tokens_in":11028,"tokens_out":1082,"duration_ms":26015,"significance":"If the reported gains prove robust, the work supplies a practical, architecture-preserving recipe for improving deep surrogates of anisotropic subgrid transport in solar and stellar convection-zone models. The multi-metric evaluation (RMSE per component, PDFs, height-resolved relative improvement) and the explicit algorithmic description (Fig. 4) are strengths that aid reproducibility. The technique is potentially transferable to other regression problems whose error landscape is spatially heterogeneous, which is a useful methodological contribution beyond the immediate solar-physics application.","major_comments":[{"comment":"Section V-D and the algorithm of Fig. 4 freeze the cluster weights after only T0=10 epochs of uniform training and employ a single hand-chosen scaling α=3. No ablation of T0 or α, nor a control that reweights by a random or magnitude-matched schedule of the same strength, is reported. Because Table II attributes the 34 % MSE / R² gains specifically to cluster-weighted training, these untested choices are load-bearing; without them it remains possible that any aggressive reweighting of the high-error tail after a short warm-up would produce comparable lifts.","section":"Section V-D / Fig. 4"},{"comment":"Tables I–II report only point estimates of RMSE, average MSE and R². Despite the authors having already generated 20 K-Means seeds and five representative clusterings, no standard deviations, bootstrap intervals or significance tests accompany the headline numbers. The quantitative claim of a 34 % reduction and an R² jump from 0.54 to 0.80 therefore lacks a statistical foundation.","section":"Tables I–II"},{"comment":"Section VI-A and Table I compare HAC-based weighting only on a 2.5 % random subset of the training data (memory-limited) and with a different power-law scheme. The resulting under-performance relative to full-data K-Means cannot be interpreted as a fair head-to-head of clustering methods, yet the text presents it as supporting evidence for the superiority of K-Means.","section":"Section VI-A / Table I"}],"minor_comments":[{"comment":"Abstract and conclusion state “34 % lower MSE” without citing the precise average-MSE values of Table II; a parenthetical reference would improve traceability.","section":"Abstract"},{"comment":"Figure 2a reports cluster-wise MSE in cm⁴ s⁻⁴ while the later tables use cm² s⁻²; units should be harmonized or explicitly converted.","section":"Fig. 2"},{"comment":"The signed-log transform applied to off-diagonal targets is mentioned only briefly; a short formula or reference would help readers reproduce the preprocessing pipeline.","section":"Section III"},{"comment":"Several figure panels (especially Fig. 7 scatter plots) lack axis labels or color-bar units; adding them would improve readability.","section":"Figs. 5–8"}],"recommendation":"major_revision","confidential_remarks":"The core architecture and data pipeline are taken from the authors’ own SSRN preprint [3]; the novelty is essentially the weighting layer. That is acceptable, but the journal should confirm that the prior work is properly cited and that the present manuscript stands alone. Scope is a reasonable fit for astro-ph.SR given the solar-physics application, though the methodological contribution is more general ML."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The one thing worth knowing is that the authors take their own prior 3-D CNN for StellarBox Reynolds stresses and show a clean, consistent improvement by freezing per-cluster errors after a short uniform warm-up and reweighting the loss. Table II and the supporting PDFs, histograms, and scatter plots (Figs. 5–8) document the same story for every component: average MSE drops from 4.89e18 to 3.01e18 and R^{2} rises from 0.54 to 0.80. That is a real, usable gain for anyone already running these surrogates.\n\nWhat is new is not the idea of hard-example reweighting—that is standard—but the concrete demonstration that K-Means on the 82-dimensional velocity-density neighborhood, followed by the simple wk = 1 + α·(MSEk / max) schedule with α = 3, works on this particular solar data set. They also show that the clusters are not just height bins, that K-Means++ is stable across seeds, and that HAC on a 2.5 % subset fails, which is honest. The architecture, preprocessing, and physics baselines are carried over cleanly from their earlier work, so the comparison is apples-to-apples.\n\nThe soft spots are real but limited. The stress-test note is right that T0 = 10 and α = 3 are never ablated, and there is no random-reweighting control of the same magnitude; so we cannot yet say the lift is uniquely due to “cluster-weighted training” rather than any aggressive focus on the high-error tail. No uncertainty bars or significance tests appear on the metrics, code and data are not released, and the promised insertion into StellarBox (the only real physical validation) is still future work. The closing claim about “diverse environments” is a mild over-reach. None of these sink the central empirical result for the reported setting.\n\nThis paper is for people who already care about data-driven SGS models in solar or stellar convection. It is incremental, transparent, and worth a serious referee’s time. I would send it out; the gaps are fixable and the numbers are worth checking.","headline":"Solid empirical lift from K-Means error reweighting on a solar Reynolds-stress CNN; the 34% MSE / R^{2} gains are real for this setup, but rest on an unablated (T0, α) schedule and still lack closed-loop physics checks.","tokens_in":11646,"tokens_out":584,"would_cite":true,"duration_ms":55866,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"Cluster-weighted training cuts CNN error on solar Reynolds stresses by 34 percent and raises R^{2} from 0.54 to 0.80.","keywords":["surrogate modeling","subgrid turbulence","Reynolds stress tensor","cluster-weighted training","K-Means","3-D CNN","solar convection","quiet Sun"],"falsifier":"Retrain the identical CNN architecture on the same solar cubes using a non-cluster importance scheme (for example, uncertainty-based or residual-based sample weights) and check whether the 34 % MSE reduction and the R^{2} jump to 0.80 still appear; if they do not, the clustering step is essential.","tokens_in":11630,"feed_emoji":"☀️","tokens_out":880,"duration_ms":9971,"temperature":0.7,"pith_summary":"Solar-atmosphere simulations cannot resolve every turbulent eddy, so they need a fast surrogate that estimates the missing Reynolds-stress tensor from large-scale density and velocity. Earlier work already showed that a 3-D convolutional network beats classical gradient and Smagorinsky closures, yet the network still errs systematically in certain flow regimes. This paper partitions the input space with K-Means, measures the residual error of each cluster after a short uniform warm-up, and reweights the loss so that high-error clusters receive more gradient attention. The resulting model recovers both the peaks and the heavy tails of the stress distributions more faithfully, lowers average mean-squared error by more than a third, and lifts the coefficient of determination from 0.54 to 0.80. The same idea is offered as a general remedy for regression problems whose error surface is spatially heterogeneous.","feed_headline":"Cluster weights cut solar-turbulence CNN error by 34%","feed_subtitle":"K-Means reweighting lifts R^{2} from 0.54 to 0.80 on Reynolds stresses without changing the network.","key_machinery":"Cluster-weighted loss: after ten epochs of uniform training, samples belonging to each K-Means cluster are multiplied by a scalar proportional to that cluster’s residual MSE; the reweighted network is then fine-tuned to completion.","core_discovery":"When a 3-D CNN that predicts the six Reynolds-stress components is retrained with a loss reweighted by the mean squared error of five K-Means clusters (weights wk = 1 + 3 · MSEk / max MSEj), average MSE falls from 4.89 × 10^18 to 3.01 × 10^18 and average R^{2} rises from 0.54 to 0.80 relative to the identical architecture trained with ordinary MSE.","pith_inferences":["Because the largest gains concentrate between –1 Mm and the photosphere, the method may be especially useful for near-surface magnetoconvection where gradients are sharpest.","The modest 15–20 % training overhead suggests the technique could be inserted into other large-eddy-simulation pipelines without redesigning the network.","If magnetic-field channels were added to the input features, the same cluster-weighting loop might further reduce errors in magnetized turbulence."],"forward_implications":["Lower-resolution StellarBox runs can replace classical subgrid closures with the cluster-weighted CNN and still recover high-resolution Reynolds-stress statistics.","Off-diagonal stress components, previously only modestly improved by deep networks, now show large error reductions once high-error clusters are emphasized.","The same reweighting recipe can be applied to any regression task whose input space contains regions of systematically higher residual.","Inference cost remains identical to the baseline CNN because only the training loss, not the architecture, is changed."],"fun_headline_variants":["K-Means weights cut solar-turbulence CNN MSE by 34%","Cluster-reweighted loss lifts CNN R² from 0.54 to 0.80","Error-cluster weights drop Reynolds-stress MSE 34% for CNN","K-Means loss reweighting improves quiet-Sun subgrid CNN","Cluster-weighted training raises solar CNN R² to 0.80"],"cache_read_input_tokens":4096,"weakest_assumption_plain":"That the cluster-wise errors measured after only ten epochs of ordinary training give a stable, non-overfitting signal for how much each sample should be weighted thereafter.","fun_headline_variants_meta":{"raw":{"variants":["K-Means weights cut solar-turbulence CNN MSE by 34%","Cluster-reweighted loss lifts CNN R² from 0.54 to 0.80","Error-cluster weights drop Reynolds-stress MSE 34% for CNN","K-Means loss reweighting improves quiet-Sun subgrid CNN","Cluster-weighted training raises solar CNN R² to 0.80"]},"model":"grok-4.5","effort":"low","cost_usd":0.005882,"raw_usage":{"total_tokens":1497,"prompt_tokens":775,"num_sources_used":0,"completion_tokens":103,"cost_in_usd_ticks":58820000,"prompt_tokens_details":{"text_tokens":775,"audio_tokens":0,"image_tokens":0,"cached_tokens":0},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":619,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":775,"tokens_out":103,"duration_ms":54182,"temperature":1.0,"reasoning_tokens":619,"cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-10T15:17:10.835662+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Retrain the identical CNN architecture on the same solar cubes using a non-cluster importance scheme (for example, uncertainty-based or residual-based sample weights) and check whether the 34 % MSE reduction and the R^{2} jump to 0.80 still appear; if they do not, the clustering step is essential.","supporting_citations":[],"review_version":1}