TFG-Flow: Training-free Guidance in Multimodal Generative Flow

Haotian Ye; Haowei Lin; Jianzhu Ma; Shanda Li; Stefano Ermon; Yiming Yang; Yitao Liang

REVIEW 4 major objections 3 minor 22 references

TFG-Flow: Training-free Guidance in Multimodal Generative Flow

T0 review · 4 major / 3 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read TFG-Flow claims to construct a guided flow for multimodal generation that preserves the unguided flow's marginals, guides discrete atom types with a logarithmic Monte-Carlo estimator, and steers continuous coordinates equivariantly…

desk verdict Genuinely novel discrete guidance for multimodal flow with a real sample-complexity bound, but the continuous-guidance algorithm is a heuristic that Theorem 3.2 does not actually cover, and the main-table numbers disagree with the appendix. read the letter →

arxiv 2501.14216 v3 pith:BJ43L4CX submitted 2025-01-24 cs.LG cs.AIcs.CE

Haowei Lin , Shanda Li , Haotian Ye , Yiming Yang , Stefano Ermon , Yitao Liang , Jianzhu Ma This is my paper · ORCID

classification cs.LGcs.AIcs.CE

keywords training-freeguidanceflowmatchingmultimodalmoleculardesigndiscreteequivariantgraphneuralnetworkimportancesamplingSO(3)-invariance

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper claims that guidance of a pretrained flow model can be extended from purely continuous data to multimodal data—coordinates plus discrete atom types—without any additional training. It proves that a guided flow exists that keeps the original flow's marginals while aligning the final sample with a target property, and derives the corresponding guided velocity and rate matrix. The key practical claim is that discrete guidance can be done with a self-normalised Monte-Carlo estimator using only a logarithmic number of samples, avoiding the exponential cost of exact expectations over atom-type assignments. Continuous guidance is performed by projected gradient ascent and provably preserves rotation invariance. If these claims hold, training-free guidance becomes applicable to molecular design and other scientific settings where discrete and continuous variables coexist, and the tests on quantum properties, structural similarity, and pocket-based drug design support that picture.

What carries the argument

The central object is the guided multimodal flow p_t(G_t|c): a joint distribution over the flow trajectory {G_t}_{t in [0,1]} and the target condition c, built so that p(c|G_1)=f_c(G_1), the trajectory marginals equal the unguided flow's for t<1, and G_t and c are independent given G_1. The guided velocity v_t(x_t|c)=E_{p_{1|t}(x_1|x_t,c)}[v_{t|1}(x_t|x_1)] and the guided rate matrix R_t(a_t,b|c)=E_{p_{1|t}(a_1|a_t,c)}[R_{t|1}(a_t,b|a_1)] then generate the guided flow through the Fokker-Planck and Kolmogorov equations. The discrete part becomes tractable because the guided rate matrix equals a ratio of unguided posterior expectations, E[f_c(G_1)R_{t|1}]/E[f_c(G_1)], which the algorithm estimates with K i.i.d. samples from the flow model's posterior; Theorem 3.4 bounds the uniform error by O(sqrt((1/K) log(n|A|/delta))). The continuous part is steered by repeating X_t <- Project_Gamma(X_t + rho_t grad_{X_t} log f_c(E[X_1|X_t,a_t])) inside the mean-zero coordinate simplex; with an equivariant graph-neural-network backbone and an invariant predictor, this makes the simulated flow SO(3)-invariant.

What would settle it

Run Algorithm 1 on a small known-data distribution where the guided marginal can be computed exactly (e.g., a two-atom Gaussian flow with a logistic classifier) and compare the distance between the algorithm's output distribution and the true p_1(x_1|c). If the distance grows with rho and does not shrink as N_iter increases, the gradient-ascent loop is not implementing the claimed guided flow.

Watch

Extended reading notes

Core claim

The paper claims that an unconditional multimodal flow model—one generating both continuous coordinates and discrete atom types—can be steered at inference time by an off-the-shelf time-independent predictor, without retraining. It constructs a joint distribution over the flow trajectory and the target condition such that the flow marginals stay exactly those of the unguided model, the condition aligns with the predictor at the clean sample, and the trajectory and condition are independent given the clean sample (Theorem 3.1); from this construction, the guided velocity and guided rate matrix follow as expectations conditioned on the current state and the target (Theorem 3.2). For discrete variables, exact computation of the guided rate matrix costs O(|A|^n) over atom-type assignments, but a self-normalised Monte-Carlo estimator using K = O((1/$epsilon^{2}$) log(n|A|/delta)) samples approximates it uniformly within epsilon with probability at least 1-delta (Theorem 3.4). For continuous coordinates, iterating a projected gradient-ascent step on the predictor's score gives an equivariant guided ODE whose simulated distribution is SO(3)-invariant (Theorem 3.5). On QM9, GEOM-Drug, and CrossDocked2020, this yields lower MAE than continuous training-free baselines on quantum properties, higher Tanimoto similarity to target structures, and the best Vina score among the compared pocket-based drug-design methods.

Load-bearing premise

The continuous guidance loop assumes that iterating projected gradient ascent on the predictor score produces a sample from the guided posterior used in Theorem 3.2, but the paper never proves this equivalence; if it fails at high guidance strength, the empirical gains are not explained by the theory.

Editorial extensions

If this is right

On QM9, TFG-Flow's MAE for six quantum properties is on average 20.3% better than the strongest continuous training-free baseline, and it is competitive with or better than a conditional flow that required labelled training data.
Target-structure guidance raises Tanimoto similarity over unconditional generation by 76.8% on QM9 and 22.4% on GEOM-Drug.
On CrossDocked2020, guided pocket-based design improves Vina score to -7.65, the best among the compared methods, while keeping QED and SA comparable.
Discrete guidance becomes practical: about 16 posterior samples already suffice empirically, consistent with the logarithmic sample-complexity theorem, and the method needs only four hyperparameters (rho, tau, K, N_iter).
Because the construction preserves flow marginals and the conditional-independence structure, the same guided velocity and rate-matrix recipe applies to any multimodal flow model with a factorised conditional flow, not only the molecular setting.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

One testable extension: in a synthetic Gaussian flow with a known classifier, the exact guided marginal p_1(x_1|c) proportional to p_data(x_1)f_c(c) is computable, so comparing Algorithm 1's output with that exact distribution would isolate whether the gradient-ascent loop actually implements the ideal guided velocity—a link the paper asserts but does not prove.
If the equivalence between gradient ascent and the guided posterior expectation fails at large guidance strength, the marginals-preservation property would break before alignment, which would show up as falling validity at high rho and tau; the paper's ablation already hints at this trade-off.
The logarithmic discrete estimator could transfer directly to other flow or diffusion settings with factorised conditional rate matrices and off-the-shelf scorers, such as discrete text generation or protein sequence-structure co-design.
The SO(3)-invariance guarantee assumes an equivariant backbone and an invariant predictor; using a non-equivariant predictor on invariant features would still guide the final sample but would not inherit the coordinate-marginal invariance proof.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Here's my take on TFG-Flow. The genuinely new thing is the discrete guidance: they construct a guided rate matrix via a weighted Monte Carlo estimator that needs only O(log(n|A|)) samples, and they prove a high-probability error bound. That part is solid and useful. The continuous guidance, however, does not do what the theory says. Theorem 3.2 defines the guided velocity as the conditional expectation under p1|t(x1|xt,c); Algorithm 1 instead nudges xt by gradient ascent on log f(E[X1|xt,at]), then calls the unguided network on the perturbed input. That's a reasonable heuristic in the spirit of DPS/FreeDoM, but it is not shown to sample from the guided flow of Theorem 3.1, and the claims about preserving marginals and exact alignment do not follow for the actual algorithm. The abstract's 'unbiased sampling' claim is also stronger than the ratio estimator in Eq. (13), which is consistent rather than unbiased.

The experiments are broad and show real gains over continuous-only training-free baselines, and the paper is honest about the gap to training-based methods and the weak Tanimoto scores. But there are internal inconsistencies: Table 1 lists α MAE as 2.32 while Appendix E Table 8 reports 3.52; εLUMO 941 vs 998; Δε 804 vs 914. That's the kind of thing that needs a correction pass.

Who should read it? Anyone working on guidance for flow matching or controllable molecular generation. The discrete part is a genuine contribution and the invariance analysis is clean. The continuous part should be treated as heuristic; the paper would be much better if it said so plainly and left the exact-guidance claim for the discrete component.

My recommendation: send it to review. The core idea is novel and the discrete analysis is worth refereeing. But the authors need to either prove something about the gradient-ascent update, or clearly mark it as an approximation and scale back the theoretical claims. Also fix the table mismatches. A major revision could make this a solid paper.

Referee Report

4 major / 3 minor

Summary. The paper proposes TFG-Flow, a training-free guidance method for multimodal (continuous + discrete) flow matching models, with applications to 3D molecular design. The method constructs a guided flow via a joint measure (Theorem 3.1), derives guided velocity and rate matrices (Theorem 3.2), handles discrete guidance with a Monte Carlo estimator whose sample complexity is logarithmic in the number of atoms and atom types (Theorem 3.4), and handles continuous guidance by a gradient-ascent update on the input coordinates with a claimed SO(3)-invariance guarantee (Theorem 3.5). Experiments on QM9, GEOM-Drug, and CrossDocked2020 report improved property alignment, structural similarity, and binding scores relative to several training-free and training-based baselines.

Significance. If the theoretical claims were fully supported, this would be a valuable extension of training-free guidance to discrete and multimodal data, and the logarithmic sample-complexity result for discrete guidance would be a genuine contribution. The paper also introduces a useful pocket-based drug-design benchmark and provides open-source code. However, the current manuscript contains a load-bearing gap between the ideal guided flow of Theorems 3.1-3.2 and the continuous guidance loop actually implemented, as well as a discrepancy between the discrete estimator analyzed in Theorem 3.4 and the estimator used in Algorithm 1. These issues, together with an inconsistency between the headline tables and the appendix, mean that the paper's central claims are not yet established as stated.

major comments (4)

[§3.3, Algorithm 1 (lines 16-19)] The continuous guidance implemented in Algorithm 1 is not shown to realize the guided flow of Theorem 3.2. Theorem 3.2 (Eq. 9) defines the guided velocity as vt(xt|c) = E_{p1|t(x1|xt,c)}[(x1−xt)/(1−t)], but the algorithm never evaluates this expectation. Instead, it perturbs the input by Xt ← ProjectΓ(Xt + ρt ∇Xt log f(E[X1|Xt,at])) and then reads the unguided network output E[X1|Xt',at] to advance the ODE. No statement in the paper shows that p1|t(·|Xt',at) approximates p1|t(·|Xt,c); for a Gaussian flow the Bayes-conditional mean has a different functional form from this gradient-ascent point. Consequently, the simulated continuous trajectory is not the Fokker-Planck flow of Theorem 3.2, and the claims of marginal preservation and alignment are unsupported for the algorithm that produced Tables 1-4. This is a load-bearing gap: either a proof of the approximation should be supplied, or the continuous loop should be explicitly reframed as a heuristic whose empirical behavior is not covered by Theorems 3.1-3.2.
[§3.2, Algorithm 1 (lines 5-10), Eq. (13)] The implemented discrete estimator does not match the estimator analyzed in Theorem 3.4. Eq. (13) requires i.i.d. samples G1|t,k = (X1|t,k, a1|t,k) from the joint posterior p1|t(G1|Gt), but Algorithm 1 samples only the discrete component a1|t,k and reuses the single point estimate X1|t = E[X1|Gt] in every evaluation f(X1|t, a1|t,k). Because the target predictor fc(G1) is nonlinear in X1, E[fc(X1,a1)] ≠ fc(E[X1],a1), so Theorem 3.4's high-probability guarantee does not apply to the implemented estimator. Moreover, the abstract's 'unbiased sampling in guiding discrete variables' is stronger than anything proven: the self-normalized ratio estimator in Eq. (13) is only consistent and high-probability accurate under the assumptions of Theorem 3.4, not unbiased in finite samples. Please either modify the algorithm to sample X1|t,k jointly (if the flow model can provide samples) or provide an analysis of the bias introduced by the point-estimate substitution.
[§3.1 and App. B.1] The relationship between the stated goal and the theorem needs to be corrected. Section 3 states that the construction yields p1(G1|c) = pdata(G1|c), but Theorem 3.1 (formal version, App. B.1) only constructs a joint measure satisfying P(c|G1) = fc(G1), from which P(G1|c) ∝ fc(G1)pdata(G1) follows. This equals pdata(G1|c) only when fc is exactly the true conditional pdata(c|G1), which is not guaranteed for an off-the-shelf training-free predictor. Relatedly, the informal Theorem 3.1 in the main text says marginals are preserved for t ∈ [0,1], while the formal statement preserves finite-dimensional marginals only for t < 1; the final-time marginal cannot be preserved and simultaneously satisfy P(c|G1) = fc(G1) with nontrivial fc. The proof of Theorem B.2 also takes expectations with respect to pdata(x1|c), which is not the final marginal of the constructed P. Please align the informal and formal statements and avoid claiming exact conditional sampling.
[Table 1 vs. App. E.2 (Table 8)] The headline results are inconsistent with the appendix. For TFG-Flow on QM9, Table 1 reports Cv = 1.75, α = 2.32, Δε = 804, and εLUMO = 941, whereas Table 8 (App. E.2) reports Cv = 1.48, α = 3.52, Δε = 914, and εLUMO = 998 for the same metric, with no explanation (the tables also differ in validity and novelty columns). The differences are far larger than the reported standard deviations (e.g., α: 0.0523), so either the appendix or the main table is wrong. Because the central quantitative claims (e.g., '+20.3% over TFG') rest on Table 1, the discrepancy must be resolved and the exact hyperparameter configuration for each reported number must be stated.

minor comments (3)

[App. B.5, proof of Theorem 3.5] In the proof of Theorem 3.5, the sentence 'X 7→ gθ(X, at)X is SO(3)-invariant' should read 'SO(3)-equivariant'; more importantly, the proof does not track the random discrete sample a1|t drawn at line 15 of Algorithm 1, so the invariance argument is incomplete as written.
[Algorithm 1 and §3.2] The notation for the target predictor is inconsistent: Algorithm 1 line 10 uses f in place of fc from Eq. (13), while line 15 uses fc; the temperature τ introduced in §3.4 also does not appear in the algorithm listing. Please make the notation uniform.
[Table 3] The 'Upper bound' row in Table 3 is not defined as clearly as in the quantum-property tables; please specify whether it is the shuffled-label baseline from EDM and why the method is allowed to exceed it.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the guided-flow construction and its velocity/rate consequences are self-contained; the continuous guidance loop is an unproven heuristic but not a circular reduction.

full rationale

The paper's central derivation chain is not circular. Theorem 3.1 explicitly constructs a joint measure with P(c|G1)=fc(G1) and verifies marginal preservation and conditional independence; Theorem 3.2 then derives the guided velocity and rate matrix by taking expectations in the Fokker-Planck and Kolmogorov equations. These are direct consequences of the construction, not hidden assumptions. Proposition 3.3 and Theorem 3.4 provide a self-contained Monte Carlo estimator for the discrete rate matrix with a Hoeffding-based concentration proof; no fitted constant enters the theorem statement. The continuous guidance in Sec. 3.3 (Eqs. 15-16) is presented as a heuristic adaptation of earlier training-free gradient methods, and the paper does not prove that the iterated gradient step equals the conditional expectation in Theorem 3.2; this is a validity gap between theorem and algorithm, not a circular reduction. Empirically, the QM9 evaluation uses a separate oracle predictor trained on a disjoint half to prevent reward hacking, and structure/drug metrics are computed by external software (RDKit, QVina), so the reported gains are not fitted by construction. The only overlapping-author citation, TFG (Ye et al., 2024), is used for baseline comparison and hyperparameter-search conventions, not to justify the theory. I therefore find no specific step where a prediction reduces to its own input.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The method introduces no new physical entities. It relies on trained models, two tuned guidance hyperparameters, and two fixed algorithmic constants. The theoretical claims are mostly self-contained, with domain assumptions about network equivariance and model accuracy.

free parameters (4)

rho (continuous guidance strength) = grid searched per task; reported as 0.02 for polarizability
Controls gradient update strength in Eq. (15). Grid searched in App D.2 with the objective of minimizing MAE while keeping validity above 75%.
tau (temperature for discrete predictor) = grid searched per task; reported as 10 for polarizability
Temperature in the softmax target predictor softmax(f(G; psi)/tau), controlling discrete guidance strength. Tuned via grid search.
K (Monte Carlo samples for discrete guidance) = 512 used in main experiments; ablation suggests 16 is sufficient
Number of samples in Eq. (13) for estimating the guided rate matrix. Fixed rather than tuned per task.
Niter (gradient ascent iterations) = 4
Number of continuous guidance update steps per flow time step. Fixed in experiments based on ablation.

assumptions (3)

domain assumption The target predictor fc is SO(3)-invariant and the flow model is SO(3)-equivariant.
Stated in Theorem 3.5 and used in its proof. These are structural assumptions about the learned networks, not proven properties.
domain assumption The learned flow model g_theta outputs sufficiently accurate samples or expectations of p1|t(G1|Gt) to serve as the proposal distribution for weighted Monte Carlo and gradient guidance.
Algorithm 1 and Eq. (7) rely on the model's predictions being close to the true conditional distribution. No error analysis covers model misspecification.
domain assumption The target predictor fc has a positive lower bound fc_min > 0 across the support of the generated samples.
Required by Theorem 3.4 for the concentration bound. Softmax outputs are positive when finite, but the lower bound magnitude affects the sample complexity.

how reviews work

0 comments

Cite this review

Pith. "Pith review of TFG-Flow: Training-free Guidance in Multimodal Generative Flow." pith.science (2026). https://pith.science/paper/BJ43L4CX

@misc{pith2026250114216,
  author       = {Pith},
  title        = {Pith review of: TFG-Flow: Training-free Guidance in Multimodal Generative Flow},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BJ43L4CX}},
  note         = {Machine review of arXiv:2501.14216}
}

read the original abstract

Given an unconditional generative model and a predictor for a target property (e.g., a classifier), the goal of training-free guidance is to generate samples with desirable target properties without additional training. As a highly efficient technique for steering generative models toward flexible outcomes, training-free guidance has gained increasing attention in diffusion models. However, existing methods only handle data in continuous spaces, while many scientific applications involve both continuous and discrete data (referred to as multimodality). Another emerging trend is the growing use of the simple and general flow matching framework in building generative foundation models, where guided generation remains under-explored. To address this, we introduce TFG-Flow, a novel training-free guidance method for multimodal generative flow. TFG-Flow addresses the curse-of-dimensionality while maintaining the property of unbiased sampling in guiding discrete variables. We validate TFG-Flow on four molecular design tasks and show that TFG-Flow has great potential in drug design by generating molecules with desired properties.

Figures

Figures reproduced from arXiv: 2501.14216 by the authors.

**Figure 1.** (a) Multimodal guided flow. The backward flow (blue arrow) is constructed with linear interpolation between observed data G1 and sampled noise G0; the forward flow (green arrows) is simulated by conditional velocity vt(x (i) t |c)and conditional rate matrix Rt(a (i) t , a (i) t+∆t |c). (b) The illustration of TFG-Flow. TFG-Flow guides the forward flow on each step Gt, which consists of a gradient-based guidance for … view at source ↗

**Figure 2.** Effect of varying hyperparameters (Niter, τ, ρ, K). The line plot indicates the validity (left y-axis, higher values are better), while the bar plot shows the MAE (right y-axis, lower values are better). Results Analysis. As binding affinity is the most important metric in target-aware drug design, we prioritize Vina score in the design of our quality score. The results show that TFG-Flow improves the drug quality o… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

22 extracted references · 10 canonical work pages

[1]

(17) Again, for each G′ 1 ∈ G, define a probability measurePG′ 1 C on C by PG′ 1 C (c′) = fc′(G′ 1)

for any EG ∈ FG. (17) Again, for each G′ 1 ∈ G, define a probability measurePG′ 1 C on C by PG′ 1 C (c′) = fc′(G′ 1). By definition of fc, PG1 C is a valid distribution. Then, we can define PG′ 1 on Ω as PG′ 1 (EG × EC) = PG′ 1 G (EG)PG′ 1 C (EC) for any EG ∈ FG; EC ∈ 2C. (18) We obtain the joint probability measure P on the product space (Ω, F ) by integ...

work page
[3]

Diffusion on the probability simplex

8 Griffin Floto, Thorsteinn Jonsson, Mihai Nica, Scott Sanner, and Eric Zhengyu Zhu. Diffusion on the probability simplex. arXiv preprint arXiv:2309.02530, 2023. 28 Paul G Francoeur, Tomohide Masuda, Jocelyn Sunseri, Andrew Jia, Richard B Iovanisci, Ian Snyder, and David R Koes. Three-dimensional convolutional neural networks and a cross-docked data set f...

arXiv 2023
[5]

6, 8, 32 Jonathan Ho and Tim Salimans

URL https://openreview.net/forum?id=o3BxOLoxm1. 6, 8, 32 Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance. arXiv preprint arXiv:2207.12598,

arXiv
[7]

Equivariant diffusion for molecule generation in 3d

28 Emiel Hoogeboom, Vıctor Garcia Satorras, Clément Vignac, and Max Welling. Equivariant diffusion for molecule generation in 3d. In International conference on machine learning, pp. 8867–8887. PMLR, 2022. 3, 7, 8, 31, 33 Emiel Hoogeboom, Jonathan Heek, and Tim Salimans. simple diffusion: End-to-end diffusion for high resolution images. In International C...

work page 2022
[8]

Learning joint 2d & 3d diffusion models for complete molecule generation

28 Han Huang, Leilei Sun, Bowen Du, and Weifeng Lv. Learning joint 2d & 3d diffusion models for complete molecule generation. arXiv preprint arXiv:2305.12347, 2023. 28 John J Irwin, Khanh G Tang, Jennifer Young, Chinzorig Dandarchuluun, Benjamin R Wong, Munkhzul Khurelbaatar, Yurii S Moroz, John Mayfield, and Roger A Sayle. Zinc20—a free ultralarge-scale ...

arXiv 2023
[9]

Optimizing diffusion noise can serve as universal motion priors

28 13 Published as a conference paper at ICLR 2025 Korrawe Karunratanakul, Konpat Preechakul, Emre Aksan, Thabo Beeler, Supasorn Suwajanakorn, and Siyu Tang. Optimizing diffusion noise can serve as universal motion priors. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 1334–1345, 2024. 29 Zixuan Ke, Yijia Shao, H...

arXiv 2025
[10]

Selecting large language model to fine-tune via rectified scaling law

29 Haowei Lin, Baizhou Huang, Haotian Ye, Qinyu Chen, Zihao Wang, Sujian Li, Jianzhu Ma, Xiaojun Wan, James Zou, and Yitao Liang. Selecting large language model to fine-tune via rectified scaling law. arXiv preprint arXiv:2402.02314, 2024. 28 Yaron Lipman, Ricky TQ Chen, Heli Ben-Hamu, Maximilian Nickel, and Matt Le. Flow matching for generative modeling....

arXiv 2024
[11]

Geometry-complete diffusion for 3d molecule generation and optimization

29 14 Published as a conference paper at ICLR 2025 Alex Morehead and Jianlin Cheng. Geometry-complete diffusion for 3d molecule generation and optimization. Communications Chemistry, 7(1):150, 2024. 28 Xingang Peng, Shitong Luo, Jiaqi Guan, Qi Xie, Jian Peng, and Jianzhu Ma. Pocket2mol: Efficient molecular sampling based on 3d protein pockets. InInternati...

arXiv 2025

Show all 22 references

[12]

Sparse training of discrete diffusion models for graph generation

28, 32, 33 Yiming Qin, Clement Vignac, and Pascal Frossard. Sparse training of discrete diffusion models for graph generation. arXiv preprint arXiv:2311.02142, 2023a. 28 Yiming Qin, Huangjie Zheng, Jiangchao Yao, Mingyuan Zhou, and Ya Zhang. Class-balancing diffusion models. I...

2014 arXiv
[14]

Tfg: Unified training-free guidance for diffusion models

28 Haotian Ye, Haowei Lin, Jiaqi Han, Minkai Xu, Sheng Liu, Yitao Liang, Jianzhu Ma, James Zou, and Stefano Ermon. Tfg: Unified training-free guidance for diffusion models. arXiv preprint, 2024. 1, 6, 8, 10, 28, 29, 30, 31, 32, 34 Jiasheng Ye, Zaixiang Zheng, Yu Bao, Lihua Qia...

2024 arXiv
[16]

Now we are ready to verify that the above joint probability measure P satisfies the desired properties

for any E ∈ F, (19) where PG,1(·) is the marginal distribution (law) of G1 in (ΩG, FG, PG). Now we are ready to verify that the above joint probability measure P satisfies the desired properties. Preservation of flow marginals. For any EG ∈ FG, P(EG × C) = Z G PG′ 1 (EG × C)PG,1(dG′
[17]

= Z G PG′ 1 G (EG)PG,1(dG′
[18]

(20) Thus, the marginal distribution of {Gt}t∈[0,1] under P is PG

= PG(EG). (20) Thus, the marginal distribution of {Gt}t∈[0,1] under P is PG. Specifically, for any 0 ≤ t1 < · · ·< tm < 1, P(Gt1 , · · ·, Gtm) = PG(Gt1 , · · ·, Gtm ). 20 Published as a conference paper at ICLR 2025 Alignment with Target Predictor. Note that in our definition,...

2025
[19]

(21) This establishes the specified alignment with the target predictor

= PG′ 1 (ΩG × {c′}) = PG′ 1 G (ΩG)PG′ 1 C ({c′}) = PG′ 1 C ({c′}) = fc(G′ 1). (21) This establishes the specified alignment with the target predictor. Conditional independence of trajectory and target. Let 0 ≤ t1 < t2 < · · ·< tm < 1, and let φG : Gm → R and φC : C →R be any b...

2025
[20]

present a more streamlined alternative to diffusion models (Sohl-Dickstein et al., 2015; Song & Ermon, 2019; Ho et al., 2020; Song et al., 2020), bypassing the need for forward and backward diffusion processes. Instead, they focus on noise-data interpolants (Albergo et al., 20...

2022
[21]

inference with external potentials

have demonstrated greater efficiency. Recent innovations, such as SEDD (Lou et al., 2023), extend score matching to discrete spaces, improving language modeling to a level competitive with autoregressive models. Additionally, DFM (Campbell et al., 2024) applies continuous-time...

2021
[22]

#Atoms” predicts the molecular properties by only using the number of atoms in the molecule. If a baseline outperforms “#Atoms

to use EGNN with kNN graph, where k = 32 and reduces the batch size to 16 and the hidden layer dimension to 128. For target predictors, we train a 6-layer discriminative EGNN by adding linear head on the average pooling of output node feature. In our implementation, we only mo...

2024
[2020]

Inverse design of 3d molecular structures with conditional generative neural networks

7, 28 Niklas W A Gebauer, Michael Gastegger, Stefaan SP Hessmann, Klaus-Robert Müller, and Kristof T Schütt. Inverse design of 3d molecular structures with conditional generative neural networks. Nature communications, 13(1):973, 2022. 9 Justin Gilmer, Samuel S Schoenholz, Pat...

2022 arXiv
[2021]

Self-conditioned embedding diffusion for text generation

28 Robin Strudel, Corentin Tallec, Florent Altché, Yilun Du, Yaroslav Ganin, Arthur Mensch, Will Grathwohl, Nikolay Savinov, Sander Dieleman, Laurent Sifre, et al. Self-conditioned embedding diffusion for text generation. arXiv preprint arXiv:2211.04236, 2022. 28 Zhicheng Sun,...

2022 arXiv
[2022]

Denoising diffusion probabilistic models

1, 30 Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. Advances in neural information processing systems, 33:6840–6851, 2020. 28 Jonathan Ho, Tim Salimans, Alexey Gritsenko, William Chan, Mohammad Norouzi, and David J Fleet. Video diffusion ...

2020
[2023]

Score-based diffusion models for accelerated mri

28 Hyungjin Chung and Jong Chul Ye. Score-based diffusion models for accelerated mri. Medical image analysis, 80:102479, 2022. 8, 29, 32 Valentin De Bortoli, Emile Mathieu, Michael Hutchinson, James Thornton, Yee Whye Teh, and Arnaud Doucet. Riemannian score-based generative m...

2022 arXiv
[2024]

Unigem: A unified approach to generation and property prediction for molecules

1, 10, 28 Shikun Feng, Yuyan Ni, Yan Lu, Zhi-Ming Ma, Wei-Ying Ma, and Yanyan Lan. Unigem: A unified approach to generation and property prediction for molecules. arXiv preprint arXiv:2410.10516,

arXiv

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.

[1] [1]

(17) Again, for each G′ 1 ∈ G, define a probability measurePG′ 1 C on C by PG′ 1 C (c′) = fc′(G′ 1)

for any EG ∈ FG. (17) Again, for each G′ 1 ∈ G, define a probability measurePG′ 1 C on C by PG′ 1 C (c′) = fc′(G′ 1). By definition of fc, PG1 C is a valid distribution. Then, we can define PG′ 1 on Ω as PG′ 1 (EG × EC) = PG′ 1 G (EG)PG′ 1 C (EC) for any EG ∈ FG; EC ∈ 2C. (18) We obtain the joint probability measure P on the product space (Ω, F ) by integ...

work page

[2] [3]

Diffusion on the probability simplex

8 Griffin Floto, Thorsteinn Jonsson, Mihai Nica, Scott Sanner, and Eric Zhengyu Zhu. Diffusion on the probability simplex. arXiv preprint arXiv:2309.02530, 2023. 28 Paul G Francoeur, Tomohide Masuda, Jocelyn Sunseri, Andrew Jia, Richard B Iovanisci, Ian Snyder, and David R Koes. Three-dimensional convolutional neural networks and a cross-docked data set f...

arXiv 2023

[3] [5]

6, 8, 32 Jonathan Ho and Tim Salimans

URL https://openreview.net/forum?id=o3BxOLoxm1. 6, 8, 32 Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance. arXiv preprint arXiv:2207.12598,

arXiv

[4] [7]

Equivariant diffusion for molecule generation in 3d

28 Emiel Hoogeboom, Vıctor Garcia Satorras, Clément Vignac, and Max Welling. Equivariant diffusion for molecule generation in 3d. In International conference on machine learning, pp. 8867–8887. PMLR, 2022. 3, 7, 8, 31, 33 Emiel Hoogeboom, Jonathan Heek, and Tim Salimans. simple diffusion: End-to-end diffusion for high resolution images. In International C...

work page 2022

[5] [8]

Learning joint 2d & 3d diffusion models for complete molecule generation

28 Han Huang, Leilei Sun, Bowen Du, and Weifeng Lv. Learning joint 2d & 3d diffusion models for complete molecule generation. arXiv preprint arXiv:2305.12347, 2023. 28 John J Irwin, Khanh G Tang, Jennifer Young, Chinzorig Dandarchuluun, Benjamin R Wong, Munkhzul Khurelbaatar, Yurii S Moroz, John Mayfield, and Roger A Sayle. Zinc20—a free ultralarge-scale ...

arXiv 2023

[6] [9]

Optimizing diffusion noise can serve as universal motion priors

28 13 Published as a conference paper at ICLR 2025 Korrawe Karunratanakul, Konpat Preechakul, Emre Aksan, Thabo Beeler, Supasorn Suwajanakorn, and Siyu Tang. Optimizing diffusion noise can serve as universal motion priors. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 1334–1345, 2024. 29 Zixuan Ke, Yijia Shao, H...

arXiv 2025

[7] [10]

Selecting large language model to fine-tune via rectified scaling law

29 Haowei Lin, Baizhou Huang, Haotian Ye, Qinyu Chen, Zihao Wang, Sujian Li, Jianzhu Ma, Xiaojun Wan, James Zou, and Yitao Liang. Selecting large language model to fine-tune via rectified scaling law. arXiv preprint arXiv:2402.02314, 2024. 28 Yaron Lipman, Ricky TQ Chen, Heli Ben-Hamu, Maximilian Nickel, and Matt Le. Flow matching for generative modeling....

arXiv 2024

[8] [11]

Geometry-complete diffusion for 3d molecule generation and optimization

29 14 Published as a conference paper at ICLR 2025 Alex Morehead and Jianlin Cheng. Geometry-complete diffusion for 3d molecule generation and optimization. Communications Chemistry, 7(1):150, 2024. 28 Xingang Peng, Shitong Luo, Jiaqi Guan, Qi Xie, Jian Peng, and Jianzhu Ma. Pocket2mol: Efficient molecular sampling based on 3d protein pockets. InInternati...

arXiv 2025