{"id":"49b05c53-edd7-433e-8ee0-3bf836996049","arxiv_id":"2605.19256","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"FSF-DMD replaces the fake-score network in distribution matching distillation with a generator-induced pseudo-velocity surrogate for flow-map generators, showing improved FID on ImageNet-1K 256x256.","lead":"The paper proposes FSF-DMD, a distillation method for flow-map generators that removes the auxiliary fake-score network by substituting a pseudo-velocity signal derived directly from the generator. This targets simpler and potentially more efficient training for few-step image generation models.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"Endpoint pseudo-velocity may not reliably proxy fake-velocity without unstated error bounds or training dynamics analysis","rationale":"The reader's weakest assumption directly matches the load-bearing point. The abstract and summary provide no quantitative validation of proxy fidelity, making the replacement of the auxiliary network the least secure step. Experiments showing FID improvement are consistent with the claim but do not isolate whether gains come from the pseudo-velocity accuracy or from other implementation details; the proposed check would falsify or support the core substitution.","tokens_in":1707,"tokens_out":374,"duration_ms":14502,"concrete_test":"At a mid-training checkpoint (e.g., after 50k iterations on ImageNet-256), independently train a small fake-score network on the current generator samples for 5k steps; compute the average L2 difference between its velocity estimates and the generator's endpoint pseudo-velocity over 1024 held-out latents; if the mean difference exceeds 0.15 (normalized) or if swapping the surrogate for the trained fake-score in the DMD loss changes FID by >3 points, the proxy accuracy claim is unsupported.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim requires that the flow-map generator's endpoint pseudo-velocity supplies a reverse-divergence signal accurate enough to replace an explicit fake-score network. This hinges on an implicit assumption that the pseudo-velocity (derived from the current generator state) remains a sufficiently low-bias estimator of the evolving fake distribution's velocity throughout training. No derivation in the provided sections establishes an error bound between this surrogate and the true fake-velocity; if the generator's flow map deviates from the true transport map (common in early training or under few-step regimes), the resulting objective could optimize against a systematically biased signal, reintroducing the very instability DMD was meant to correct.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The manuscript proposes FSF-DMD, a distribution-matching distillation method for flow-map generators that eliminates the auxiliary fake-score network. It replaces the fake-velocity estimator with a generator-induced endpoint pseudo-velocity surrogate to supply the reverse-divergence signal, derives a practical objective, adds flow-map-consistent backward simulation, and introduces a self-teacher variant for training from scratch. Experiments on ImageNet-1K 256x256 report FID improvements over flow-map baselines and competitive or better results than listed DMD2 comparisons under flow-map initialization, flow-matching initialization, and scratch training.","tokens_in":1859,"tokens_out":522,"duration_ms":40311,"significance":"If the pseudo-velocity proxy is reliable, the work simplifies DMD-style corrections by removing memory and update overhead of a separate network, which is a practical advantage for few-step flow-based generators. The multi-initialization experimental protocol and inclusion of a from-scratch variant provide useful robustness evidence. The approach directly exploits the flow-map structure, which is a clean technical observation.","major_comments":[{"comment":"The central claim that the endpoint pseudo-velocity supplies a sufficiently accurate reverse-divergence signal rests on an unverified assumption that this surrogate tracks the evolving fake distribution without substantial bias. No error bound, bias analysis, or training-dynamics argument is provided to quantify the approximation quality between the pseudo-velocity and the true velocity field of the current generator distribution (see the key observation and objective derivation).","section":"Derivation of the objective"},{"comment":"Table reporting FID scores (ImageNet-1K 256x256): the improvements over DMD2 are stated for the flow-map-initialized setting, but without reported standard deviations across seeds, exact baseline re-implementation details, or an ablation isolating the pseudo-velocity surrogate from the backward-simulation component, it is difficult to attribute gains specifically to the proposed proxy.","section":"Experiments"}],"minor_comments":[{"comment":"Abstract: the phrase 'reaches lower FID than the listed DMD2 comparisons' is imprecise; explicitly name the DMD2 variants and point to the corresponding table/figure for clarity.","section":"Abstract"},{"comment":"Notation: introduce and consistently distinguish 'pseudo-velocity' from true velocity and from the flow-map velocity field at first appearance to prevent reader confusion in the objective and simulation sections.","section":"Notation and preliminaries"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the positive assessment of the work's significance and for the constructive major comments. We address each point below and have revised the manuscript accordingly to strengthen the presentation of the derivation and the experimental evidence.","responses":[{"response":"We acknowledge that the original manuscript presents the pseudo-velocity surrogate as a direct consequence of the flow-map structure without a dedicated error analysis. The derivation relies on the fact that, for a flow-map generator, the endpoint velocity is induced exactly by the generator's own forward mapping, which supplies the reverse-divergence signal by construction. While formal bounds were not derived in the submission, this alignment is consistent with standard Lipschitz assumptions on velocity fields in flow-based models. In the revised manuscript we have expanded the derivation section with a short bias discussion under these assumptions and added empirical training-dynamics plots that track the correlation between the pseudo-velocity and the evolving generator distribution.","revision_made":"yes","referee_comment":"[Derivation of the objective] The central claim that the endpoint pseudo-velocity supplies a sufficiently accurate reverse-divergence signal rests on an unverified assumption that this surrogate tracks the evolving fake distribution without substantial bias. No error bound, bias analysis, or training-dynamics argument is provided to quantify the approximation quality between the pseudo-velocity and the true velocity field of the current generator distribution (see the key observation and objective derivation)."},{"response":"We agree that additional reporting details are necessary to support attribution of the gains. The revised manuscript now includes standard deviations computed over three independent random seeds for all reported FID numbers. We have added an appendix subsection that documents the exact re-implementation of the DMD2 baselines, including optimizer settings, learning-rate schedules, and data-augmentation choices. We have also inserted a new ablation table that isolates the pseudo-velocity surrogate by comparing the full objective against a controlled variant that retains only the flow-map-consistent backward simulation.","revision_made":"yes","referee_comment":"[Experiments] Table reporting FID scores (ImageNet-1K 256x256): the improvements over DMD2 are stated for the flow-map-initialized setting, but without reported standard deviations across seeds, exact baseline re-implementation details, or an ablation isolating the pseudo-velocity surrogate from the backward-simulation component, it is difficult to attribute gains specifically to the proposed proxy."}],"tokens_in":1415,"tokens_out":502,"duration_ms":29779,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The main takeaway is that they replace the auxiliary fake-score network in distribution matching distillation with a generator-induced pseudo-velocity when the model has a flow-map structure. This cuts one extra network and its update cost while keeping the reverse-divergence correction that DMD provides for few-step sampling.","headline":"The paper drops the fake-score network in DMD for flow-map generators by using the model's own endpoint pseudo-velocity as proxy, with decent FID gains on ImageNet but thin support for why the surrogate stays accurate.","tokens_in":2325,"tokens_out":143,"would_cite":false,"duration_ms":19634,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":{"model":"grok-4.3","evidence":[{"relation":"unclear","rs_module":"IndisputableMonolith/Cost/FunctionalEquation.lean","rs_theorem":"washburn_uniqueness_aczel","paper_passage":"the endpoint pseudo-velocity of a flow-map generator provides a tractable proxy for fake-velocity estimation... v_fake(ˆx_t; t) ≈ F_θ(ˆx_t; t, 0) (Eq. 12)"},{"relation":"unclear","rs_module":"IndisputableMonolith/Foundation/AlexanderDuality.lean","rs_theorem":"alexander_duality_circle_linking","paper_passage":"flow-map generator... semigroup property... injectivity and invertibility"}],"headline":"Flow-map surrogate for DMD velocity estimation has no structural overlap with RS cost or forcing machinery","alignment":"orthogonal","rationale":"The paper's core construction replaces an auxiliary fake-score network with the endpoint pseudo-velocity F_θ(ˆx_t; t, 0) derived from the flow-map generator under PF-ODE coupling and injectivity (Eqs. 10-12, Sec. 3.2, A.3). This is a practical ML surrogate for reverse-divergence correction inside a consistency-distillation objective. RS contains no theorems about velocity fields, flow maps, or distribution-matching distillation; its central objects are the reciprocal cost J(x) = ½(x + x⁻¹) − 1 (Cost.FunctionalEquation), the absolute-floor closure (AbsoluteFloorClosure), Alexander-duality forcing of D = 3 (AlexanderDuality), and the 8-tick / φ-ladder derivations of constants. None of these appear in or are paralleled by the paper's machinery.","tokens_in":58391,"confidence":"high","tokens_out":396,"duration_ms":15723,"cache_read_input_tokens":128,"cache_creation_input_tokens":0},"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"Flow-map generators can replace the auxiliary fake-score network in distribution matching distillation by using their own endpoint pseudo-velocity as a reverse-divergence proxy.","keywords":["distribution matching distillation","flow-map generators","pseudo-velocity","fake-score network","few-step generation","reverse-divergence","ImageNet-1K"],"falsifier":"Train matching flow-map generators with the pseudo-velocity objective versus an explicit fake-score network on the same backbone and data; if the final FID scores or distribution match metrics diverge substantially, or if the pseudo-velocity version fails to improve over the plain flow-map baseline, the proxy claim is falsified.","tokens_in":2616,"feed_emoji":"🖼️","tokens_out":703,"duration_ms":62079,"temperature":0.7,"pith_summary":"This paper asks whether the extra fake-score network needed in distribution matching distillation can be removed once the generator itself follows a flow-map structure. It claims that the endpoint pseudo-velocity already produced inside the flow-map generator works as a usable proxy for the fake-velocity signal that supplies the reverse-divergence correction. If the substitution holds, training and memory costs drop because only a single network is needed while the method still performs distribution-level matching rather than pointwise losses. The authors derive a practical objective from this observation, add flow-map-consistent backward simulation, and introduce a self-teacher variant that trains from scratch. Experiments on ImageNet-1K at 256 by 256 resolution show that the resulting FSF-DMD improves flow-map baselines and reaches lower FID than the listed DMD2 comparisons in the flow-map-initialized setting.","feed_headline":"Flow-map endpoint velocity replaces fake-score network","feed_subtitle":"Pseudo-velocity proxy supplies the reverse-divergence signal for distribution matching without extra network overhead.","key_machinery":"Generator-induced pseudo-velocity surrogate, which replaces the auxiliary fake-score estimator by using the flow-map endpoint velocity to deliver the required reverse-divergence correction.","core_discovery":"The endpoint pseudo-velocity of a flow-map generator provides a tractable proxy for fake-velocity estimation, allowing the generator itself to supply the reverse-divergence signal without an explicit auxiliary network.","pith_inferences":["The same endpoint-velocity substitution may simplify other distillation procedures that currently maintain separate score estimators for distribution correction.","If the approximation remains reliable across noise schedules and resolutions, it could reduce the engineering effort required to deploy few-step generators on memory-constrained hardware.","Evaluating the method on conditional or higher-dimensional generation tasks would test whether the flow-map proxy generalizes beyond the static-image setting reported."],"forward_implications":["The derived objective extends DMD-style distribution matching to flow-map generators without the memory and update cost of a second network.","Flow-map-consistent backward simulation can be added to the training loop for greater stability.","A self-teacher variant enables the full method to train from scratch without a separate teacher model.","FSF-DMD reaches lower FID than listed DMD2 comparisons when initialized from flow maps on ImageNet-1K 256x256."],"fun_headline_variants":["Flow-map velocity replaces fake-score network in DMD","Pseudo-velocity enables DMD without fake-score networks","Generator endpoint velocity skips auxiliary fake-score","Flow-map pseudo-velocity supplies reverse-divergence signal","DMD uses generator-induced pseudo-velocity for correction"],"cache_read_input_tokens":64,"weakest_assumption_plain":"The flow-map structure inherently supplies a sufficiently accurate reverse-divergence signal via its endpoint pseudo-velocity without requiring an explicit auxiliary network or additional corrections that would reintroduce similar overhead.","fun_headline_variants_meta":{"raw":{"variants":["Flow-map velocity replaces fake-score network in DMD","Pseudo-velocity enables DMD without fake-score networks","Generator endpoint velocity skips auxiliary fake-score","Flow-map pseudo-velocity supplies reverse-divergence signal","DMD uses generator-induced pseudo-velocity for correction"]},"model":"grok-4.3","cost_usd":0.01301,"raw_usage":{"total_tokens":5625,"prompt_tokens":627,"num_sources_used":0,"completion_tokens":59,"cost_in_usd_ticks":130099500,"prompt_tokens_details":{"text_tokens":627,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":4939,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":627,"tokens_out":59,"duration_ms":54099,"temperature":1.0,"reasoning_tokens":4939,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-05-20T07:13:58.823089+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"Train matching flow-map generators with the pseudo-velocity objective versus an explicit fake-score network on the same backbone and data; if the final FID scores or distribution match metrics diverge substantially, or if the pseudo-velocity version fails to improve over the plain flow-map baseline, the proxy claim is falsified.","supporting_citations":[],"review_version":1}