Pith. sign in

REVIEW 3 major objections 6 minor 90 references

Signed Rectified Flow turns a signed measure (positive minus negative distributions) into a valid sampling process that provably avoids the negative regions.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 15:09 UTC pith:MHDOIZK4

load-bearing objection Signed RF is a genuinely new way to put negative distributions into flow guidance, with a clean theory for the idealized dynamics; the practical method is plausible but the exact exclusion guarantee does not transfer to the implemented sampler. the 3 major comments →

arxiv 2607.18516 v1 pith:MHDOIZK4 submitted 2026-07-20 cs.LG cs.CV

Signed Rectified Flow: Negativity-Controlled Generation

Runlong Liao , Baiyu Su , Lizhang Chen , Qiang Liu This is my paper
classification cs.LG cs.CV
keywords signed measuresrectified flowgenerative modelingguidancedensity ratio estimationanti-memorizationconcept suppressioncontinuity equation
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Signed Rectified Flow generalizes Rectified Flow to a signed target πsign = (1+α)π+ − απ−, where π+ is the distribution to promote and π− the distribution to suppress. Although a signed measure cannot be sampled directly, the paper shows that simulating the flow from the usual noise source yields a valid probability law that matches πsign on a 'reachable' subset of its positive region and never enters the negative region. The zero set of the signed marginal acts as a repulsive barrier, and the positive mass that is not reached exactly balances the excluded negative mass. This yields a state-dependent guidance rule whose local strength is set by the density ratio π−/π+, rather than a fixed schedule. The practical payoff is better fidelity–diversity trade-offs in class-conditional generation, reduced training-data replication, and suppression of unsafe concepts with little loss in image quality.

Core claim

The paper's central claim is Theorem B.4: for the exact Signed RF dynamics (under local C1 regularity and nondegenerate zero set), the law of the source-initialized ODE satisfies πsignRF_t(x)=πsign_t(x) 1{x∈Ωr_t}. The flow rectifies a signed density into a valid probability: on the reachable region Ωr_t it preserves the signed density exactly; outside it (ghost and negative regions) it places zero mass. Proposition B.3 shows trajectories never cross the zero set. The same construction gives guidance-form velocity v+_t + λα_t(v+_t−v−_t) with λα_t = α rt/((1+α)−α rt), so the effective guidance scale is a local function of the density ratio rt=π−_t/π+_t. The paper provides two estimators for rt

What carries the argument

The central object is the signed marginal πsign_t = (1+α)π+_t − απ−_t, obtained by writing the usual rectified-flow mixture with a negative coefficient, together with the Signed RF velocity field formed by dividing the signed flux by πsign_t. Because both branches share a Gaussian source, the Tweedie identity shows the signed flux at the zero set points toward the positive side, so the zero set acts as a repulsive barrier. The paper interprets the dynamics as charged particles: positive and negative particles annihilate at the moving zero set, while backward from t=1 only particles from the reachable region connect to the source at t=0. A Liouville/continuity-equation argument along the flow

Load-bearing premise

The load-bearing premise is that the signed density's zero set is always a smooth boundary (nonzero gradient wherever it vanishes), so the signed flux points into the positive region and trajectories can never cross it.

What would settle it

Construct a Gaussian-mixture target where the positive and negative components meet tangentially so that πsign_t(x)=0 and ∇πsign_t(x)=0 at some point, simulate Signed RF with the closed-form ratio, and check whether any trajectory reaches a negative region; any crossing would falsify nonpenetration under degenerate zero sets. In practical settings, measuring signed-density violations along sampled trajectories with finite Euler steps and a learned classifier would test the robustness of the exact guarantee.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If the central theorem holds, negative information enters generation exactly: the induced sampling law is a total-variation-optimal nonnegative approximation of the signed target, equal to the signed density on the reachable region.
  • The guidance scale is locally determined by the density ratio π−_t/π+_t, so repulsion is automatically weak where the state looks positive and strong where it looks negative, replacing fixed guidance schedules with a state-dependent rule.
  • Taking the training set as the negative distribution yields a training-free analytic negative flow; repelling from it shifts nearest-neighbor similarity upward while keeping FID and inception score close to baseline.
  • In safety and concept-suppression settings, subtracting a model of unsafe outputs from a model of safe outputs lowers attack success rates while preserving CLIP and aesthetic scores.
  • Explicitly modeling invalid regions as π− resolves the missing-negative-data problem for planning, removing wall crossings while retaining broad path diversity.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the nondegeneracy assumption (nonzero gradient wherever πsign_t vanishes) fails in high-dimensional learned models, sharp exclusion boundaries may become degenerate, and the exact nonpenetration guarantee weakens to approximate exclusion; this is worth testing in learned latent spaces.
  • Because the guidance scale is a density-ratio function, the same construction could be applied to preference data by treating dispreferred examples as π−, connecting sampling-time suppression to preference-based training objectives.
  • The ghost region means Signed RF silently discards some positive mass; a testable extension is to characterize when the reachable region covers the modes of π+, since diversity depends on this dynamically determined subset.
  • A direct falsification experiment would use an exactly computable ratio in a 2D Gaussian mixture whose zero set has a degenerate tangency and check whether any trajectory crosses into a negative region.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper introduces Signed Rectified Flow (Signed RF), a generalization of Rectified Flow that replaces the target distribution by the signed measure π^sign = (1+α)π^+ − απ^−, where π^+ is promoted and π^− is suppressed. By linearity of the RF marginal and flux, the authors derive a signed velocity field that is well defined away from the zero set of π^sign. They prove that, under Assumption B.2 (C1 regularity and nondegenerate zero set), source-initialized trajectories remain in the positive region and the sampling law coincides with π^sign on a dynamically reachable subset Ω^r_t, while negative and ghost regions are never visited. The paper also derives a guidance-form implementation whose local strength is the density ratio π^−_t/π^+_t, estimated either by a classifier or by online ODE tracking. Experiments cover 2D toys, PointMaze, ImageNet-256 class-conditional generation, anti-memorization via an analytic negative flow, and concept/nudity suppression in Stable Diffusion 3.5, reporting improvements over constant-guidance baselines.

Significance. If the result holds, Signed RF provides a principled way to inject negative information into flow-based generative models, with a clear target-level construction and a nontrivial theoretical characterization of the induced sampling law. The proofs in Appendix B are coherent and the TV-optimality interpretation (Proposition B.1) is a nice contribution. The paper is also strong on the empirical side: the ImageNet, anti-memorization, and safety experiments are extensive and include useful ablations (classifier checkpoint stability, guidance-cap sensitivity, runtime measurements). The main caveat is that the 'provable exclusion' guarantee is established only for the exact Signed RF ODE under Assumption B.2, whereas the implemented pipeline uses estimated ratios, clipping, and caps; the manuscript should make this distinction much more prominent.

major comments (3)
  1. [Assumption B.2 / Proposition B.3 / Remark B.5] The nonpenetration guarantee relies critically on the nondegeneracy condition ∇π^sign_t(x) ≠ 0 whenever π^sign_t(x) = 0. For π^sign_t = (1+α)π^+_t − απ^−_t where π^±_t are learned densities or empirical Gaussian mixtures, this condition is not enforced and can fail at degenerate points; Remark B.5 explicitly concedes such cases are not covered. The abstract and Sec. 2.2 state 'provably excluding' without this qualification. Please add the regularity condition to the main-text claims, or provide a generic-perturbation argument showing degenerate zero sets are negligible, or empirically test nondegeneracy in the evaluated settings.
  2. [Sec. 2.3 / Eq. (5) and stabilization] The implemented sampler replaces the exact velocity with a classifier- or online-estimated density ratio, then applies the denominator lower bound max((1+α)−α r_t, ε) and the guidance cap λ_max. These modifications mean the simulated trajectories are not governed by the signed flux, so the density identity and nonpenetration results (Prop. B.3, Thm. B.4) do not apply to the evaluated algorithm. The main text should state this explicitly in Sec. 2.3 and soften the abstract/conclusion accordingly. It would strengthen the paper to quantify the effect of clipping (e.g., bound on the distance from the zero set, or a post-hoc check that no samples land in Ω^−_1).
  3. [Sec. 3.2 / Fig. 6 / Table 1] The ImageNet FID numbers are reported as best-FID after sweeping α (and ω for CFG) on the same 50K evaluation set. Selecting the best hyperparameter on the test set can overestimate improvements because FID has sampling variance. The stability tables (Tables 2 and 3) are helpful, but they do not address the selection-bias issue. Please report a held-out hyperparameter selection procedure, or give mean±std over multiple seeds for a fixed α chosen on a validation split, and include the full sweep curves so readers can assess the operating points.
minor comments (6)
  1. [Sec. 2.2 / Figure 2] The text says 'the black curve shows the signed marginal', but the figure is in color; please adjust the color reference or label the curves directly.
  2. [Sec. 2.3 / Eq. (6)] In the online ratio-tracking ODE, λ^α_t(z_t) is used before its definition at the tracked ratio; clarify that λ^α_t is evaluated at r_t(z_t)=exp(u_t).
  3. [Sec. 3.2] The negative distribution for the ImageNet classifier is a mixture of low-quality and wrong-class samples, not a fixed π^− as in the theory. The density-ratio interpretation is therefore approximate; please state this explicitly in the main text.
  4. [Tables 2 and 3] The header 'Step α' is ambiguous; rename it to 'α' and clarify that FID is computed on the same 50K sample set.
  5. [Figure 8] The legend 'RF Ours' is unclear; it should be split into distinct labels for the base model and the proposed method.
  6. [Appendix A] The sentence 'under the stated assumptions, the strict nonpenetration result applies to the exact Signed RF dynamics' is an important caveat that should appear in Sec. 2.3 or 2.2, not only in the related-work appendix.

Circularity Check

0 steps flagged

No circular derivation: Signed RF's sampling-law theorem follows from the continuity equation and flow-map mass conservation; the only self-citations are standard background, and the flagged caveats are assumptions, not input–output identification.

full rationale

Walked the paper's derivation chain. The signed marginal πsign_t and signed flux are defined by linear combination of the two branch RF marginals/fluxes (Eqs. 8–9), and the Signed RF velocity is defined as flux divided by signed density (Eq. 10). The central identity πsignRF_t(x) = πsign_t(x) 1{x∈Ωr_t} (Eq. 4 / Theorem B.4) is not definitional: Ωr_t is defined as the image of the source-initialized flow map, and the equality of densities is proved via the continuity equation plus Liouville's formula, using πsign_0 = π0. This is a genuine transport calculation, not a restatement of the input. The nonpenetration result (Proposition B.3) is a separate flux-sign argument using the Gaussian-source identity (∇πsign_t)ᵀ jsign_t = ((1−t)/t)‖∇πsign_t‖² > 0 on the zero set; it does not assume the conclusion. Assumption B.2's nondegenerate zero-set condition is a real premise, and Remark B.5 explicitly concedes that degenerate zero sets are 'not covered by the present result,' so the limitation is disclosed rather than hidden. The empirical machinery (classifier-based ratio estimation, online ratio tracking, clipping/capping, α sweeps) is standard estimation and hyperparameter tuning; no fitted parameter is renamed as a prediction. In anti-memorization, the nearest-neighbor SSCD metric is aligned with the method's explicit objective of repelling from the training set, so improvements on that metric are unsurprising, but the paper also reports an independent latent-L2 metric and does not frame the experiment as a first-principles prediction. Self-citations (Rectified Flow [20,21], lecture notes [24], Momentum Guidance [8]) supply standard background or baselines and are not invoked to rule out alternatives. No circular step meeting the quoted-reduction standard was found.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 3 invented entities

The method introduces a signed target and derives a velocity field from it; the main free parameters are the suppression strength α (swept per task), guidance caps, and manual negative-prompt choices. The theory rests on standard RF marginal preservation plus an explicit regularity/nondegeneracy assumption (B.2). The ghost region and negative mass are internal mathematical constructs, not independently verified entities.

free parameters (4)
  • α (suppression strength) = swept per task; e.g., ImageNet α≈0.07–0.10, anti-memorization α=0.1–7.5
    Controls the balance between positive and negative targets; selected by validation/test FID or ASR in each experiment, not predicted by the theory.
  • λ_max (guidance cap) = e.g., 10, 5, 3 in ImageNet
    Numerical stabilization; chosen by hand to avoid excessive guidance when the denominator (1+α)−αr_t becomes small.
  • ε (denominator lower bound) = 1e-3
    Clipping constant in the stabilization procedure; ad hoc numerical safeguard.
  • negative prompt / negative distribution choice = manually selected per concept (e.g., 'Mickey Mouse, Disney character')
    The definition of π− is a modeling choice in concept suppression and safety experiments; it determines what gets suppressed but is not derived.
axioms (5)
  • standard math Rectified Flow marginal preservation and linearity of marginals/flux for convex mixtures
    Borrowed from RF literature [20,21]; used to define the signed marginal πsign_t and the velocity field in Eq. 3.
  • domain assumption Common Gaussian source π0=N(0,I) and Tweedie identity s±_t=(t v±_t − x)/(1−t)
    Used in the flux barrier proof (Prop B.3) and in the online density-ratio tracking update (Eq. 6, Eq. 14).
  • ad hoc to paper Assumption B.2: local C1 regularity of πsign_t and signed flux, and nondegeneracy ∇πsign_t≠0 on the zero set
    Required for the nonpenetration theorem and the sampling-law theorem; not guaranteed by RF theory and can fail for empirical/singular targets at t=1.
  • standard math ODE well-posedness on compact intervals [0,T]⊂[0,1)
    Needed to define the flow map Φ_t and justify the change-of-variables argument in Theorem B.4.
  • domain assumption Both RF branches share the same source distribution
    Ensures πsign_0=π0>0 and initializes the log-ratio u0=0; standard in RF setups where π+ and π− are transported from the same noise source.
invented entities (3)
  • Ghost region Ωg_t no independent evidence
    purpose: Explains the unsampled positive mass that balances the excluded negative mass; partitions the positive region into reachable and unreachable parts.
    Defined as Ω+_t \ Ωr_t; its existence follows from the mass balance, but its location is determined implicitly by the dynamics and it has no independent falsifiable handle.
  • Signed measure / negative mass no independent evidence
    purpose: Intermediate mathematical object encoding suppression by subtracting απ− from (1+α)π+.
    πsign is not a probability measure; used as a construction device. Its 'negative mass' has no empirical handle outside the flow construction.
  • Exclusion barrier (zero set Ω0_t) no independent evidence
    purpose: Singular boundary that prevents source-initialized trajectories from entering negative regions.
    A dynamical consequence of the flux identity and nondegeneracy assumption; not an independently measurable physical object.

pith-pipeline@v1.3.0-alltime-deepseek · 30239 in / 11750 out tokens · 131236 ms · 2026-08-01T15:09:15.631215+00:00 · methodology

0 comments
read the original abstract

We introduce Signed Rectified Flow (Signed RF), a generalization of Rectified Flow that targets the signed measure $\pi^{sign} = (1+\alpha)\pi^+ - \alpha\pi^-$, where $\alpha>0$, $\pi^+$ is the distribution to promote, and $\pi^-$ is the distribution to suppress. Although direct sampling from a signed measure is not well-defined, Signed RF induces a valid generative process that concentrates probability in regions where the signed measure is positive while provably excluding regions dominated by its negative component. It therefore provides a principled framework for incorporating negative information and exclusion constraints into generative modeling. We analyze the signed continuity equation underlying Signed RF and use a charged-particle interpretation to explain how negative mass forms exclusion barriers. This theory further motivates practical adaptive guidance algorithms. Across several applications, Signed RF improves the fidelity-diversity trade-off on ImageNet, reduces nearest-neighbor similarity in anti-memorization experiments, and reduces nudity induced by adversarial prompts in Stable Diffusion 3.5 while preserving CLIP and aesthetic scores.

Figures

Figures reproduced from arXiv: 2607.18516 by Baiyu Su, Lizhang Chen, Qiang Liu, Runlong Liao.

Figure 1
Figure 1. Figure 1: Signed Rectified Flow steers samples toward the [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: 1D evolution of the signed density π sign t and the induced sampling law π signRF t . Signed RF rejects negative regions and matches the signed density on a reachable subset of the positive region. • Trajectories remain in the positive region. The source-initialized trajectories remain in the posi￾tive region Ω + t :={x : π sign t (x) > 0} and never enter the negative region Ω − t :={x : π sign t (x) < 0}.… view at source ↗
Figure 3
Figure 3. Figure 3: Signed RF sample-region decomposition. Left: Signed RF dynamics partition the space into reachable, negative (rejection), and ghost regions. Right: the terminal signed density π sign 1 . The induced sampling law is supported on the reachable region, where it coincides with the signed density, while the negative and ghost regions remain unobserved by the source-initialized sampler. • The ghost region. The r… view at source ↗
Figure 4
Figure 4. Figure 4: 2D Gaussian-mixture comparison. Top: constant guidance with varying scales. Bottom: Signed RF with varying α. The background visualizes the signed target π sign = (1 + α)π + − απ−, with blue and pink indicating positive and negative regions, respectively [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: The missing negative data. A positive-only planner produces wall-crossing trajectories (a). Constant guidance trades safety for path diversity: weak guidance leaves violations (b), while strong guidance collapses the trajectories (c). Signed RF avoids obstacles while preserving diverse feasible paths (d). The Pareto plot compares the number of failures and path diversity (e). 3.2 Signed RF as State-Aware C… view at source ↗
Figure 6
Figure 6. Figure 6: ImageNet-256 comparison. Left: best-FID operating points from the parameter sweeps. Right: Pareto frontiers; markers denote the corresponding best-FID points. a better fidelity–diversity trade-off: while both methods improve precision at the expense of recall, Signed RF maintains higher recall at comparable precision. We further verify that the density-ratio-based guidance is stable in practice. Performanc… view at source ↗
Figure 7
Figure 7. Figure 7: Memorization examples on ImageNet. Each column uses the same initial seed, with the second and fourth rows showing the nearest training images. The base model closely reproduces training examples, including dataset-specific artifacts such as borders, watermarks, and backgrounds. In contrast, Data Repulsive Flow suppresses these artifacts while preserving class semantics, produc￾ing samples that are visibly… view at source ↗
Figure 8
Figure 8. Figure 8: Quantitative comparison for anti-memorization on ImageNet. Left: SSCD L2 nearest￾neighbor histograms for four classes on the same high-risk seeds. Upper right: generation quality under the standard 50K-sample evaluation. Lower right: memorization statistics for class 248, summarized by the 5%, 10%, and 25% quantiles and mean SSCD L2 distance (higher is better). where wi(x, t) are the normalized Gaussian ke… view at source ↗
Figure 9
Figure 9. Figure 9: Quantitative comparison for nudity prevention on SD 3.5-Medium. Left: quantitative results on Ring-A-Bell and COCO-30K. Signed RF reduces unsafe generation while maintaining comparable image quality. Right: qualitative comparisons. positive branch v + that models safe outputs and a negative branch v − that models the corresponding unsafe outputs. The training data are constructed from paired prompts in ViS… view at source ↗
Figure 10
Figure 10. Figure 10: Qualitative comparison of concept suppression on Z-Image. [PITH_FULL_IMAGE:figures/full_fig_p009_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Density-ratio estimation ablation. Columns compare constant guidance, the analytic ratio, the classifier-based estimate, and the online-tracked estimate, using the same initial Gaussian noise and velocity fields v + t and v − t . Panels (e)–(g) show the mean guidance scale λ α t , the mean log-density ratio log rt, and the sample-wise guidance scales along individual trajectories. Despite small estimation… view at source ↗
Figure 12
Figure 12. Figure 12: Additional density-ratio estimation ablation in a different setup. [PITH_FULL_IMAGE:figures/full_fig_p023_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Additional density-ratio estimation ablation in a different setup. [PITH_FULL_IMAGE:figures/full_fig_p024_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Additional density-ratio estimation ablation in a different setup. [PITH_FULL_IMAGE:figures/full_fig_p024_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Additional 2D examples across multiple setups. [PITH_FULL_IMAGE:figures/full_fig_p025_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Effective guidance scale λ α t along trajectories for four ImageNet classes. Each color corresponds to the same initial seed shared across panels. Despite identical seeds, the guidance profiles differ substantially across classes, illustrating the nontrivial and class-dependent behavior of the guidance [PITH_FULL_IMAGE:figures/full_fig_p028_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Runtime and peak CUDA memory of the vectorized Gaussian analytic flow on an [PITH_FULL_IMAGE:figures/full_fig_p032_17.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

90 extracted references · 23 linked inside Pith

  1. [1]

    Finetuned language models are zero-shot learners.arXiv preprint arXiv:2109.01652, 2021

    Jason Wei, Maarten Bosma, Vincent Y Zhao, Kelvin Guu, Adams Wei Yu, Brian Lester, Nan Du, Andrew M Dai, and Quoc V Le. Finetuned language models are zero-shot learners.arXiv preprint arXiv:2109.01652, 2021

    Pith/arXiv arXiv 2021

  2. [2]

    Multitask prompted training enables zero-shot task generalization.arXiv preprint arXiv:2110.08207, 2021

    Victor Sanh, Albert Webson, Colin Raffel, Stephen H Bach, Lintang Sutawika, Zaid Alyafeai, Antoine Chaffin, Arnaud Stiegler, Teven Le Scao, Arun Raja, et al. Multitask prompted training enables zero-shot task generalization.arXiv preprint arXiv:2110.08207, 2021

    Pith/arXiv arXiv 2021

  3. [3]

    Classifier-free diffusion guidance.arXiv preprint arXiv:2207.12598, 2022

    Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance.arXiv preprint arXiv:2207.12598, 2022

    Pith/arXiv arXiv 2022

  4. [4]

    Diffusion models beat gans on image synthesis

    Prafulla Dhariwal and Alexander Nichol. Diffusion models beat gans on image synthesis. Advances in neural information processing systems, 34:8780–8794, 2021

    2021

  5. [5]

    High- resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High- resolution image synthesis with latent diffusion models. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10684–10695, 2022

    2022

  6. [6]

    Glide: Towards photorealistic image generation and editing with text-guided diffusion models.arXiv preprint arXiv:2112.10741, 2021

    Alex Nichol, Prafulla Dhariwal, Aditya Ramesh, Pranav Shyam, Pamela Mishkin, Bob McGrew, Ilya Sutskever, and Mark Chen. Glide: Towards photorealistic image generation and editing with text-guided diffusion models.arXiv preprint arXiv:2112.10741, 2021

    Pith/arXiv arXiv 2021

  7. [7]

    Photorealistic text-to-image diffusion models with deep language understanding.Advances in neural information processing systems, 35:36479–36494, 2022

    Chitwan Saharia, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily L Denton, Kamyar Ghasemipour, Raphael Gontijo Lopes, Burcu Karagol Ayan, Tim Salimans, et al. Photorealistic text-to-image diffusion models with deep language understanding.Advances in neural information processing systems, 35:36479–36494, 2022

    2022

  8. [8]

    Momentum guidance: Plug-and-play guidance for flow models.arXiv preprint arXiv:2602.20360, 2026

    Runlong Liao, Jian Yu, Baiyu Su, Chi Zhang, Lizhang Chen, and Qiang Liu. Momentum guidance: Plug-and-play guidance for flow models.arXiv preprint arXiv:2602.20360, 2026

    Pith/arXiv arXiv 2026

  9. [9]

    Training a helpful and harmless assistant with reinforcement learning from human feedback.arXiv preprint arXiv:2204.05862, 2022

    Yuntao Bai, Andy Jones, Kamal Ndousse, Amanda Askell, Anna Chen, Nova DasSarma, Dawn Drain, Stanislav Fort, Deep Ganguli, Tom Henighan, et al. Training a helpful and harmless assistant with reinforcement learning from human feedback.arXiv preprint arXiv:2204.05862, 2022

    Pith/arXiv arXiv 2022

  10. [10]

    Constitutional ai: Harmlessness from ai feedback.arXiv preprint arXiv:2212.08073, 2022

    Yuntao Bai, Saurav Kadavath, Sandipan Kundu, Amanda Askell, Jackson Kernion, Andy Jones, Anna Chen, Anna Goldie, Azalia Mirhoseini, Cameron McKinnon, et al. Constitutional ai: Harmlessness from ai feedback.arXiv preprint arXiv:2212.08073, 2022

    Pith/arXiv arXiv 2022

  11. [11]

    Safe latent diffusion: Mitigating inappropriate degeneration in diffusion models

    Patrick Schramowski, Manuel Brack, Björn Deiseroth, and Kristian Kersting. Safe latent diffusion: Mitigating inappropriate degeneration in diffusion models. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 22522–22531, 2023

    2023

  12. [12]

    Training language models to follow instructions with human feedback.Advances in neural information processing systems, 35:27730–27744, 2022

    Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, et al. Training language models to follow instructions with human feedback.Advances in neural information processing systems, 35:27730–27744, 2022

    2022

  13. [13]

    Direct preference optimization: Your language model is secretly a reward model

    Rafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D Manning, Stefano Ermon, and Chelsea Finn. Direct preference optimization: Your language model is secretly a reward model. Advances in neural information processing systems, 36:53728–53741, 2023

    2023

  14. [14]

    Training diffusion models with reinforcement learning

    Kevin Black, Michael Janner, Yilun Du, Ilya Kostrikov, and Sergey Levine. Training diffusion models with reinforcement learning. InThe Twelfth International Conference on Learning Representations, 2024. 10

    2024

  15. [15]

    Diffusion model alignment using direct preference optimization

    Bram Wallace, Meihua Dang, Rafael Rafailov, Linqi Zhou, Aaron Lou, Senthil Purushwalkam, Stefano Ermon, Caiming Xiong, Shafiq Joty, and Nikhil Naik. Diffusion model alignment using direct preference optimization. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8228–8238, 2024

    2024

  16. [16]

    Aligning text-to-image models using human feedback.arXiv preprint arXiv:2302.12192, 2023

    Kimin Lee, Hao Liu, Moonkyung Ryu, Olivia Watkins, Yuqing Du, Craig Boutilier, Pieter Abbeel, Mohammad Ghavamzadeh, and Shixiang Shane Gu. Aligning text-to-image models using human feedback.arXiv preprint arXiv:2302.12192, 2023

    Pith/arXiv arXiv 2023

  17. [17]

    Denoising diffusion probabilistic models.Advances in neural information processing systems, 33:6840–6851, 2020

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models.Advances in neural information processing systems, 33:6840–6851, 2020

    2020

  18. [18]

    Score-based generative modeling through stochastic differential equations.arXiv preprint arXiv:2011.13456, 2020

    Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equations.arXiv preprint arXiv:2011.13456, 2020

    Pith/arXiv arXiv 2011

  19. [19]

    Denoising diffusion implicit models.arXiv preprint arXiv:2010.02502, 2020

    Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising diffusion implicit models.arXiv preprint arXiv:2010.02502, 2020

    Pith/arXiv arXiv 2010

  20. [20]

    Flow straight and fast: Learning to generate and transfer data with rectified flow.arXiv preprint arXiv:2209.03003, 2022

    Xingchao Liu, Chengyue Gong, and Qiang Liu. Flow straight and fast: Learning to generate and transfer data with rectified flow.arXiv preprint arXiv:2209.03003, 2022

    Pith/arXiv arXiv 2022

  21. [21]

    Rectified flow: A marginal preserving approach to optimal transport.arXiv preprint arXiv:2209.14577, 2022

    Qiang Liu. Rectified flow: A marginal preserving approach to optimal transport.arXiv preprint arXiv:2209.14577, 2022

    Pith/arXiv arXiv 2022

  22. [22]

    Flow matching for generative modeling.arXiv preprint arXiv:2210.02747, 2022

    Yaron Lipman, Ricky TQ Chen, Heli Ben-Hamu, Maximilian Nickel, and Matt Le. Flow matching for generative modeling.arXiv preprint arXiv:2210.02747, 2022

    Pith/arXiv arXiv 2022

  23. [23]

    Building normalizing flows with stochastic interpolants.arXiv preprint arXiv:2209.15571, 2022

    Michael S Albergo and Eric Vanden-Eijnden. Building normalizing flows with stochastic interpolants.arXiv preprint arXiv:2209.15571, 2022

    Pith/arXiv arXiv 2022

  24. [24]

    Let Us Flow Together

    Qiang Liu. Let Us Flow Together. Lecture notes, https://www.cs.utexas.edu/~lqiang/ PDF/flow_book.pdf, 2024. Version dated December 24, 2024

    2024

  25. [25]

    A stochastic estimator of the trace of the influence matrix for laplacian smoothing splines.Communications in Statistics-Simulation and Computation, 18(3):1059– 1076, 1989

    Michael F Hutchinson. A stochastic estimator of the trace of the influence matrix for laplacian smoothing splines.Communications in Statistics-Simulation and Computation, 18(3):1059– 1076, 1989

    1989

  26. [26]

    Younis, Rodrigo Perez-Vicente, John U

    Omar G. Younis, Rodrigo Perez-Vicente, John U. Balis, Will Dudley, Alex Davey, and Jordan K Terry. Minari, September 2024

    2024

  27. [27]

    D4rl: Datasets for deep data-driven reinforcement learning.arXiv preprint arXiv:2004.07219, 2020

    Justin Fu, Aviral Kumar, Ofir Nachum, George Tucker, and Sergey Levine. D4rl: Datasets for deep data-driven reinforcement learning.arXiv preprint arXiv:2004.07219, 2020

    Pith/arXiv arXiv 2004

  28. [28]

    Imagenet: A large- scale hierarchical image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large- scale hierarchical image database. In2009 IEEE conference on computer vision and pattern recognition, pages 248–255. Ieee, 2009

    2009

  29. [29]

    Pytorch rectifiedflow, 2024

    Qiang Liu, Runlong Liao, Bo Liu, and Xixi Hu. Pytorch rectifiedflow, 2024

    2024

  30. [30]

    Gans trained by a two time-scale update rule converge to a local nash equilibrium.Advances in neural information processing systems, 30, 2017

    Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and Sepp Hochreiter. Gans trained by a two time-scale update rule converge to a local nash equilibrium.Advances in neural information processing systems, 30, 2017

    2017

  31. [31]

    Improved techniques for training gans.Advances in neural information processing systems, 29, 2016

    Tim Salimans, Ian Goodfellow, Wojciech Zaremba, Vicki Cheung, Alec Radford, and Xi Chen. Improved techniques for training gans.Advances in neural information processing systems, 29, 2016

    2016

  32. [32]

    As- sessing generative models via precision and recall.Advances in neural information processing systems, 31, 2018

    Mehdi SM Sajjadi, Olivier Bachem, Mario Lucic, Olivier Bousquet, and Sylvain Gelly. As- sessing generative models via precision and recall.Advances in neural information processing systems, 31, 2018

    2018

  33. [33]

    Improving rectified flow with boundary conditions

    Xixi Hu, Runlong Liao, Keyang Xu, Bo Liu, Yeqing Li, Eugene Ie, Hongliang Fei, and Qiang Liu. Improving rectified flow with boundary conditions. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 18177–18186, 2025. 11

    2025

  34. [34]

    A self- supervised descriptor for image copy detection

    Ed Pizzi, Sreya Dutta Roy, Sugosh Nagavara Ravindra, Priya Goyal, and Matthijs Douze. A self- supervised descriptor for image copy detection. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 14532–14542, 2022

    2022

  35. [35]

    Shielded diffusion: Generating novel and diverse images using sparse repellency

    Michael Kirchhof, James Thornton, Louis Béthune, Pierre Ablin, Eugene Ndiaye, and Marco Cuturi. Shielded diffusion: Generating novel and diverse images using sparse repellency. InProceedings of the 42nd International Conference on Machine Learning, volume 267 of Proceedings of Machine Learning Research, pages 30911–30942, 2025

    2025

  36. [36]

    Scaling rectified flow trans- formers for high-resolution image synthesis

    Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas Müller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, et al. Scaling rectified flow trans- formers for high-resolution image synthesis. InForty-first international conference on machine learning, 2024

    2024

  37. [37]

    Safe-CLIP: Removing NSFW Concepts from Vision-and-Language Models

    Samuele Poppi, Tobia Poppi, Federico Cocchi, Marcella Cornia, Lorenzo Baraldi, and Rita Cucchiara. Safe-CLIP: Removing NSFW Concepts from Vision-and-Language Models. In Proceedings of the European Conference on Computer Vision, 2024

    2024

  38. [38]

    Ring-a-bell! how reliable are concept removal methods for diffusion models?arXiv preprint arXiv:2310.10012, 2023

    Yu-Lin Tsai, Chia-Yi Hsu, Chulin Xie, Chih-Hsun Lin, Jia-You Chen, Bo Li, Pin-Yu Chen, Chia-Mu Yu, and Chun-Ying Huang. Ring-a-bell! how reliable are concept removal methods for diffusion models?arXiv preprint arXiv:2310.10012, 2023

    Pith/arXiv arXiv 2023

  39. [39]

    SAFREE: Training- free and adaptive guard for safe text-to-image and video generation

    Jaehong Yoon, Shoubin Yu, Vaidehi Patil, Huaxiu Yao, and Mohit Bansal. SAFREE: Training- free and adaptive guard for safe text-to-image and video generation. InThe Thirteenth Interna- tional Conference on Learning Representations, 2025

    2025

  40. [40]

    Training-free safe denoisers for safe use of diffusion models

    Mingyu Kim, Dongjun Kim, Amman Yusuf, Stefano Ermon, and Mijung Park. Training-free safe denoisers for safe use of diffusion models. InAdvances in Neural Information Processing Systems, volume 38, 2025

    2025

  41. [41]

    Z-Image Team, Huanqia Cai, Sihan Cao, Ruoyi Du, Peng Gao, Aiming Hao, Steven C. H. Hoi, Zhaohui Hou, Shijie Huang, Dengyang Jiang, Yuming Jiang, Xin Jin, Liangchen Li, et al. Z-Image: An efficient image generation foundation model with single-stream diffusion transformer.arXiv preprint arXiv:2511.22699, 2025

    Pith/arXiv arXiv 2025

  42. [42]

    Classifier-free guidance is a predictor-corrector.Trans- actions on Machine Learning Research, 2025

    Arwen Bradley and Preetum Nakkiran. Classifier-free guidance is a predictor-corrector.Trans- actions on Machine Learning Research, 2025

    2025

  43. [43]

    Conditional diffusion models with classifier-free gibbs-like guidance.arXiv preprint arXiv:2505.21101, 2025

    Badr Moufad, Yazid Janati, Alain Durmus, Ahmed Ghorbel, Eric Moulines, and Jimmy Ols- son. Conditional diffusion models with classifier-free gibbs-like guidance.arXiv preprint arXiv:2505.21101, 2025

    Pith/arXiv arXiv 2025

  44. [44]

    Guiding a diffusion model with a bad version of itself.Advances in Neural Information Processing Systems, 37:52996–53021, 2024

    Tero Karras, Miika Aittala, Tuomas Kynkäänniemi, Jaakko Lehtinen, Timo Aila, and Samuli Laine. Guiding a diffusion model with a bad version of itself.Advances in Neural Information Processing Systems, 37:52996–53021, 2024

    2024

  45. [45]

    Applying guidance in a limited interval improves sample and distribution quality in diffusion models

    Tuomas Kynkäänniemi, Miika Aittala, Tero Karras, Samuli Laine, Timo Aila, and Jaakko Lehtinen. Applying guidance in a limited interval improves sample and distribution quality in diffusion models. InAdvances in Neural Information Processing Systems, 2024

    2024

  46. [46]

    Cfg++: Manifold-constrained classifier free guidance for diffusion models

    Hyungjin Chung, Jeongsol Kim, Geon Yeong Park, Hyelin Nam, and Jong Chul Ye. Cfg++: Manifold-constrained classifier free guidance for diffusion models. InThe Thirteenth Interna- tional Conference on Learning Representations, 2025

    2025

  47. [47]

    Seyedmorteza Sadat, Otmar Hilliges, and Romann M. Weber. Eliminating oversaturation and artifacts of high guidance scales in diffusion models. InThe Thirteenth International Conference on Learning Representations, 2025

    2025

  48. [48]

    Yeh, and Ziwei Liu

    Weichen Fan, Amber Yijia Zheng, Raymond A. Yeh, and Ziwei Liu. Cfg-zero*: Improved classifier-free guidance for flow matching models.arXiv preprint arXiv:2503.18886, 2025

    Pith/arXiv arXiv 2025

  49. [49]

    Angle domain guidance: Latent diffusion requires rotation rather than extrapolation

    Cheng Jin, Zhenyu Xiao, Chutao Liu, and Yuantao Gu. Angle domain guidance: Latent diffusion requires rotation rather than extrapolation. InInternational Conference on Machine Learning, pages 28187–28212. PMLR, 2025. 12

    2025

  50. [50]

    Dynamic negative guidance of diffusion models

    Felix Koulischer, Johannes Deleu, Gabriel Raya, Thomas Demeester, and Luca Ambrogioni. Dynamic negative guidance of diffusion models. InThe Thirteenth International Conference on Learning Representations, 2025

    2025

  51. [51]

    Compositional visual generation with energy-based models

    Yilun Du, Shuang Li, and Igor Mordatch. Compositional visual generation with energy-based models. InAdvances in Neural Information Processing Systems, 2020

    2020

  52. [52]

    Tenenbaum

    Nan Liu, Shuang Li, Yilun Du, Antonio Torralba, and Joshua B. Tenenbaum. Compositional visual generation with composable diffusion models. InEuropean Conference on Computer Vision, 2022

    2022

  53. [53]

    Sega: Instructing text-to-image models using semantic guidance

    Manuel Brack, Felix Friedrich, Dominik Hintersdorf, Lukas Struppek, Patrick Schramowski, and Kristian Kersting. Sega: Instructing text-to-image models using semantic guidance. In Advances in Neural Information Processing Systems, 2023

    2023

  54. [54]

    Erasing concepts from diffusion models

    Rohit Gandikota, Joanna Materzynska, Jaden Fiotto-Kaufman, and David Bau. Erasing concepts from diffusion models. InProceedings of the IEEE/CVF International Conference on Computer Vision, 2023

    2023

  55. [55]

    Unified concept editing in diffusion models

    Rohit Gandikota, Hadas Orgad, Yonatan Belinkov, Joanna Materzynska, and David Bau. Unified concept editing in diffusion models. InWinter Conference on Applications of Computer Vision (WACV), 2024

    2024

  56. [56]

    Mace: Mass concept erasure in diffusion models

    Shilin Lu, Zilan Wang, Leyang Li, Yanzhu Liu, and Adams Wai-Kin Kong. Mace: Mass concept erasure in diffusion models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024

    2024

  57. [57]

    Defensive unlearning with adversarial training for robust concept erasure in diffusion models

    Yimeng Zhang, Xin Chen, Jinghan Jia, Yihua Zhang, Chongyu Fan, Jiancheng Liu, Mingyi Hong, Ke Ding, and Sijia Liu. Defensive unlearning with adversarial training for robust concept erasure in diffusion models. InAdvances in Neural Information Processing Systems, 2024

    2024

  58. [58]

    SAeuron: Interpretable concept unlearning in diffusion models with sparse autoencoders

    Bartosz Cywi´nski and Kamil Deja. SAeuron: Interpretable concept unlearning in diffusion models with sparse autoencoders. InForty-Second International Conference on Machine Learning, 2025

    2025

  59. [59]

    Extracting training data from diffusion models

    Nicholas Carlini, Jamie Hayes, Milad Nasr, Matthew Jagielski, Vikash Sehwag, Florian Tramèr, Borja Balle, Daphne Ippolito, and Eric Wallace. Extracting training data from diffusion models. In32nd USENIX Security Symposium (USENIX Security 23), pages 5253–5270, 2023

    2023

  60. [60]

    Diffu- sion art or digital forgery? investigating data replication in diffusion models

    Gowthami Somepalli, Vasu Singla, Micah Goldblum, Jonas Geiping, and Tom Goldstein. Diffu- sion art or digital forgery? investigating data replication in diffusion models. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023

    2023

  61. [61]

    Understanding and mitigating copying in diffusion models

    Gowthami Somepalli, Vasu Singla, Micah Goldblum, Jonas Geiping, and Tom Goldstein. Understanding and mitigating copying in diffusion models. InAdvances in Neural Information Processing Systems, 2023

    2023

  62. [62]

    Detecting, explaining, and mitigating memorization in diffusion models

    Yuxin Wen, Yuchen Liu, Chen Chen, and Lingjuan Lyu. Detecting, explaining, and mitigating memorization in diffusion models. InThe Twelfth International Conference on Learning Representations, 2024

    2024

  63. [63]

    Differentially private diffusion models.Transactions on Machine Learning Research, 2023

    Tim Dockhorn, Tianshi Cao, Arash Vahdat, and Karsten Kreis. Differentially private diffusion models.Transactions on Machine Learning Research, 2023

    2023

  64. [64]

    Dimakis, and Adam Klivans

    Giannis Daras, Kulin Shah, Yuval Dagan, Aravind Gollakota, Alexandros G. Dimakis, and Adam Klivans. Ambient diffusion: Learning clean distributions from corrupted data. In Advances in Neural Information Processing Systems, 2023

    2023

  65. [65]

    Classifier-free guidance inside the attraction basin may cause memorization

    Anubhav Jain, Yuya Kobayashi, Takashi Shibuya, Yuhta Takida, Nasir Memon, Julian Togelius, and Yuki Mitsufuji. Classifier-free guidance inside the attraction basin may cause memorization. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2025. 13

    2025

  66. [66]

    Seyedmorteza Sadat, Jakob Buhmann, Derek Bradley, Otmar Hilliges, and Romann M. Weber. Cads: Unleashing the diversity of diffusion models through condition-annealed sampling. In The Twelfth International Conference on Learning Representations, 2024

    2024

  67. [67]

    Cambridge University Press, 2012

    Masashi Sugiyama, Taiji Suzuki, and Takafumi Kanamori.Density Ratio Estimation in Machine Learning. Cambridge University Press, 2012

    2012

  68. [68]

    Gutmann and Aapo Hyvärinen

    Michael U. Gutmann and Aapo Hyvärinen. Noise-contrastive estimation of unnormalized statistical models, with applications to natural image statistics.Journal of Machine Learning Research, 13:307–361, 2012

    2012

  69. [69]

    Density ratio estimation via infinitesimal classification

    Kristy Choi, Chenlin Meng, Yang Song, and Stefano Ermon. Density ratio estimation via infinitesimal classification. InInternational Conference on Artificial Intelligence and Statistics, 2022

    2022

  70. [70]

    Refining generative process with discriminator guidance in score-based diffusion models

    Dongjun Kim, Yeongmin Kim, Se Jung Kwon, Wanmo Kang, and Il-Chul Moon. Refining generative process with discriminator guidance in score-based diffusion models. InInternational Conference on Machine Learning, 2023

    2023

  71. [71]

    Ricky T. Q. Chen, Yulia Rubanova, Jesse Bettencourt, and David Duvenaud. Neural ordinary differential equations. InAdvances in Neural Information Processing Systems, 2018

    2018

  72. [72]

    Will Grathwohl, Ricky T. Q. Chen, Jesse Bettencourt, Ilya Sutskever, and David Duvenaud. Ffjord: Free-form continuous dynamics for scalable reversible generative models. InInterna- tional Conference on Learning Representations, 2019

    2019

  73. [73]

    Diffusion-npo: Negative preference optimization for better preference-aligned generation of diffusion models

    Fu-Yun Wang, Yunhao Shui, Jingtan Piao, Keqiang Sun, and Hongsheng Li. Diffusion-npo: Negative preference optimization for better preference-aligned generation of diffusion models. InThe Thirteenth International Conference on Learning Representations, 2025

    2025

  74. [74]

    Stochastic interpolants: A unifying framework for flows and diffusions.arXiv preprint arXiv:2303.08797, 2023

    Michael S Albergo, Nicholas M Boffi, and Eric Vanden-Eijnden. Stochastic interpolants: A unifying framework for flows and diffusions.arXiv preprint arXiv:2303.08797, 2023

    Pith/arXiv arXiv 2023

  75. [75]

    Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980, 2014

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980, 2014

    Pith/arXiv arXiv 2014

  76. [76]

    Representation alignment for generation: Training diffusion transformers is easier than you think

    Sihyun Yu, Sangkyung Kwak, Huiwon Jang, Jongheon Jeong, Jonathan Huang, Jinwoo Shin, and Saining Xie. Representation alignment for generation: Training diffusion transformers is easier than you think. InThe Thirteenth International Conference on Learning Representations, 2025

    2025

  77. [77]

    An image is worth 16x16 words: Transformers for image recognition at scale.arXiv preprint arXiv:2010.11929, 2020

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. An image is worth 16x16 words: Transformers for image recognition at scale.arXiv preprint arXiv:2010.11929, 2020

    Pith/arXiv arXiv 2010

  78. [78]

    Scalable diffusion models with transformers

    William Peebles and Saining Xie. Scalable diffusion models with transformers. InProceedings of the IEEE/CVF international conference on computer vision, pages 4195–4205, 2023

    2023

  79. [79]

    Decoupled weight decay regularization.arXiv preprint arXiv:1711.05101, 2017

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization.arXiv preprint arXiv:1711.05101, 2017. A Related Work Inference-time guidance and its target distribution.Classifier guidance augments the diffusion score with the gradient of an external classifier, whereas classifier-free guidance (CFG) forms an affine combination of conditional and...

    Pith/arXiv arXiv 2017

  80. [80]

    3.3, we construct a separate analytic negative branch for each ImageNet class

    / one_minus_t[:, None] 53velocity = velocity_flat.reshape(x_shape) 54 55if not return_logp: 56return velocity, None 57 58log_mixture = ( 59torch.logsumexp(log_weights.float(), dim=1) 60- math.log(float(num_data)) 61) 62log_normalizer = ( 63-0.5 * latent_dim * math.log(2.0 * math.pi) 64- latent_dim * torch.log(one_minus_t.float()) 65) 66logp = log_mixture ...

    2048

Showing first 80 references.