pith. machine review for the scientific record. sign in

arxiv: 2604.03015 · v1 · submitted 2026-04-03 · 💻 cs.LG · math.PR· stat.ML

Recognition: 2 theorem links

· Lean Theorem

Generating DDPM-based Samples from Tilted Distributions

Achal Bassamboo, Agniv Bandyopadhyay, Dhruman Gupta, Himadri Mandal, Rushil Gupta, Sandeep Juneja, Sarvesh Ravichandran Iyer, Varun Gupta

Authors on Pith no claims yet

Pith reviewed 2026-05-13 20:02 UTC · model grok-4.3

classification 💻 cs.LG math.PRstat.ML
keywords diffusion modelstilted distributionsplug-in estimatorminimax optimalityWasserstein boundstotal variation accuracysample generation
0
0 comments X

The pith

A plug-in estimator from n original samples generates diffusion outputs close to a true tilted distribution.

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

The paper shows how to produce diffusion-based samples from a distribution obtained by tilting an original law by a vector parameter θ, when only n independent draws from the untitled distribution are given. It constructs a plug-in estimator for the tilted law and proves the estimator is minimax optimal while deriving explicit Wasserstein-distance bounds between the estimator's induced measure and the true tilted measure. Under further regularity assumptions the paper establishes that running a diffusion process on samples from the plug-in estimator produces output whose total variation distance to the desired tilted distribution vanishes at a controlled rate. The construction directly supports moment-constrained sampling tasks that arise in finance, weather, and climate applications.

Core claim

Given n independent samples from a d-dimensional probability distribution, the plug-in estimator for the θ-tilted distribution is minimax-optimal. Wasserstein bounds between the law of the plug-in estimator and the true tilted distribution are obtained as explicit functions of n and θ, identifying regimes in which the two are close. Under additional assumptions, diffusion models applied to samples drawn from the plug-in estimator achieve total-variation accuracy to the target tilted distribution.

What carries the argument

The plug-in estimator for the tilted distribution, which approximates the reweighted measure by modifying the empirical distribution according to the tilt parameter θ.

If this is right

  • Wasserstein distance between plug-in and true tilted laws shrinks with larger n for any fixed θ.
  • Diffusion sampling on the plug-in estimator produces total-variation-close draws to the target tilted law under the paper's assumptions.
  • The method supplies samples obeying practical moment constraints without requiring direct draws from the tilted measure.
  • Simulation experiments confirm the derived rates for both Wasserstein and total-variation metrics.

Where Pith is reading between the lines

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

  • The same plug-in construction could be paired with non-diffusion generators such as GANs or flow-based models.
  • In high dimensions the dependence of the bounds on d may require additional regularization or dimension reduction to remain practical.
  • The framework aligns with exponential tilting and could be used to enforce linear constraints in downstream optimization or risk-measure calculations.

Load-bearing premise

The total-variation guarantee for the diffusion step holds only under unspecified regularity conditions on the base distribution and the tilting map.

What would settle it

If the total-variation distance between diffusion-generated samples and the true tilted distribution does not decrease toward zero as n increases for fixed moderate θ, the accuracy claim is refuted.

Figures

Figures reproduced from arXiv: 2604.03015 by Achal Bassamboo, Agniv Bandyopadhyay, Dhruman Gupta, Himadri Mandal, Rushil Gupta, Sandeep Juneja, Sarvesh Ravichandran Iyer, Varun Gupta.

Figure 1
Figure 1. Figure 1: The left plot shows the empirical sliced Wasserstein distance (W2) between the reweighed esti￾mator and the true tilted distribution. The right plot shows the theoretical bound derived in Theorem 2 as a function of sample size N. As the theoretical bound van￾ishes, the empirical error decreases correspondingly. In this experiment, given a tilting parameter θ ∈ R d , we construct a reweighed distribution us… view at source ↗
Figure 2
Figure 2. Figure 2: For full construction details, refer to Appendix B. [PITH_FULL_IMAGE:figures/full_fig_p013_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: DDPM samples from P. Daily tempera￾ture fields (May–June, India, 5◦×5 ◦ ) [PITH_FULL_IMAGE:figures/full_fig_p032_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: DDPM samples from Pθ. Reweighted training targets the hotter, rarer slice with EPθ [g] = EP [g] + 1 [PITH_FULL_IMAGE:figures/full_fig_p033_4.png] view at source ↗
read the original abstract

Given $n$ independent samples from a $d$-dimensional probability distribution, our aim is to generate diffusion-based samples from a distribution obtained by tilting the original, where the degree of tilt is parametrized by $\theta \in \mathbb{R}^d$. We define a plug-in estimator and show that it is minimax-optimal. We develop Wasserstein bounds between the distribution of the plug-in estimator and the true distribution as a function of $n$ and $\theta$, illustrating regimes where the output and the desired true distribution are close. Further, under some assumptions, we prove the TV-accuracy of running Diffusion on these tilted samples. Our theoretical results are supported by extensive simulations. Applications of our work include finance, weather and climate modelling, and many other domains, where the aim may be to generate samples from a tilted distribution that satisfies practically motivated moment constraints.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes a plug-in estimator for generating DDPM samples from a theta-tilted version of an unknown d-dimensional distribution given n i.i.d. samples. It claims the estimator is minimax optimal, derives Wasserstein bounds between the law of the estimator and the true tilted measure as explicit functions of n and theta, and proves total-variation accuracy of the DDPM output under unspecified assumptions on the tilted density; the claims are illustrated by simulations and motivated by applications requiring moment constraints.

Significance. If the Wasserstein bounds and the passage to TV accuracy are made rigorous with explicit, verifiable regularity conditions, the results would supply a practical and theoretically grounded method for sampling from tilted distributions via diffusion models, directly relevant to constrained generation tasks in finance, climate modeling, and related domains.

major comments (2)
  1. [theoretical results on TV accuracy] The TV-accuracy guarantee for DDPM applied to the tilted plug-in samples (stated after the Wasserstein bounds) rests on regularity conditions on the score of the tilted density that are described only as 'some assumptions.' These conditions are load-bearing: standard DDPM convergence arguments require at least Lipschitz or bounded-gradient control on the score, which the tilt can violate when |theta| grows with d or n; the Wasserstein result alone does not imply the needed control.
  2. [definition and optimality of the plug-in estimator] The minimax-optimality claim for the plug-in estimator is stated without specifying the precise risk functional, the class of competing estimators, or the parameter regime (e.g., whether theta is fixed or may grow with n). Without these details the optimality statement cannot be verified against standard minimax lower bounds for density estimation or moment-constrained problems.
minor comments (2)
  1. [abstract] The abstract and introduction should explicitly list the regularity conditions required for the TV result rather than deferring them to 'some assumptions.'
  2. [experimental section] Simulations are described as 'extensive' but lack reported error bars, quantitative comparison to baselines, and explicit values of n, d, and |theta| regimes tested; these should be added to allow assessment of the practical range where the bounds hold.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive report. The comments highlight important points on the rigor of our assumptions and the precise statement of optimality. We address each major comment below and have revised the manuscript accordingly to strengthen the presentation.

read point-by-point responses

  1. Referee: [theoretical results on TV accuracy] The TV-accuracy guarantee for DDPM applied to the tilted plug-in samples (stated after the Wasserstein bounds) rests on regularity conditions on the score of the tilted density that are described only as 'some assumptions.' These conditions are load-bearing: standard DDPM convergence arguments require at least Lipschitz or bounded-gradient control on the score, which the tilt can violate when |theta| grows with d or n; the Wasserstein result alone does not imply the needed control.

    Authors: We agree that the regularity conditions must be stated explicitly rather than left as 'some assumptions.' In the revised manuscript we now specify that the score of the tilted density is assumed to be Lipschitz continuous with a constant that may depend on theta and d, and we add a discussion of the regimes in which this holds (in particular, when theta remains bounded independently of n and d). We also clarify that the Wasserstein closeness bounds control the deviation of the plug-in measure but do not by themselves guarantee the score regularity needed for standard DDPM TV bounds; the additional Lipschitz assumption is therefore stated separately. These changes appear in the statement of the TV-accuracy theorem and the surrounding discussion. revision: yes

  2. Referee: [definition and optimality of the plug-in estimator] The minimax-optimality claim for the plug-in estimator is stated without specifying the precise risk functional, the class of competing estimators, or the parameter regime (e.g., whether theta is fixed or may grow with n). Without these details the optimality statement cannot be verified against standard minimax lower bounds for density estimation or moment-constrained problems.

    Authors: We thank the referee for pointing out the need for precision. The minimax optimality is established with respect to the risk functional E[W_1(hat mu_theta, mu_theta)], where the expectation is over the n samples and W_1 denotes the 1-Wasserstein distance; the class of competing estimators consists of all measurable maps from n i.i.d. samples to probability measures on R^d. Theta is treated as fixed (independent of n), although the explicit Wasserstein bounds we derive hold uniformly over theta belonging to any fixed compact set. In the revision we have added a precise statement of the risk, the estimator class, and the fixed-theta regime immediately before the optimality theorem, together with a brief comparison to standard minimax lower bounds for density estimation under moment constraints. revision: yes

Circularity Check

0 steps flagged

No significant circularity: derivations use standard plug-in estimation, minimax proofs, and diffusion error bounds under explicit assumptions

full rationale

The paper defines a plug-in estimator for the tilted distribution, proves its minimax optimality via standard statistical arguments, derives Wasserstein bounds as functions of n and theta, and establishes TV-accuracy for DDPM sampling under regularity assumptions on the tilt. None of these steps reduce by construction to fitted parameters renamed as predictions, self-definitional loops, or load-bearing self-citations. The central claims rest on external properties of diffusion models and classical estimation theory rather than internal redefinitions or ansatzes smuggled via prior work by the same authors.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The central claim depends on the well-definedness of the tilted distribution via theta and the applicability of standard diffusion models to the plug-in samples; no new entities are postulated.

free parameters (2)
  • theta
    Tilt parameter in R^d that controls the degree of tilting and moment constraints.
  • n
    Number of independent samples from the original distribution used to build the plug-in estimator.
axioms (2)
  • domain assumption A tilted distribution can be obtained from the original by reweighting with parameter theta in a manner compatible with diffusion sampling.
    Core to the problem definition and the TV-accuracy claim.
  • domain assumption The plug-in estimator converges in a way that allows Wasserstein bounds and minimax optimality to hold.
    Invoked for the theoretical bounds and optimality result.

pith-pipeline@v0.9.0 · 5478 in / 1460 out tokens · 61418 ms · 2026-05-13T20:02:11.825336+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

Reference graph

Works this paper leans on

44 extracted references · 44 canonical work pages

  1. [1]

    Unified perspective on exponential tilt and bridge algorithms for rare trajectories of discrete Markov processes.Phys

    Javier Aguilar and Riccardo Gatto. Unified perspective on exponential tilt and bridge algorithms for rare trajectories of discrete Markov processes.Phys. Rev. E, 109:034113, Mar 2024

    work page 2024

  2. [2]

    Springer International Publishing, Cham, 2022

    Mayer Alvo.Exponential Tilting and Its Applications, pages 171–193. Springer International Publishing, Cham, 2022

    work page 2022

  3. [3]

    The accumulation of score estimation error in diffusion models

    Anonymous. The accumulation of score estimation error in diffusion models. InInternational Conference on Learning Representations (ICLR) 2026, 2026. Under review as ICLR 2026 submission

    work page 2026

  4. [4]

    Minimum entropy calibration of asset pricing models, internat

    M Avellaneda. Minimum entropy calibration of asset pricing models, internat. J. Theoret. Appl. Finance, 1:447472, 1998. GENERATING DDPM-BASED SAMPLES FROM TILTED DISTRIBUTIONS 15

    work page 1998

  5. [5]

    arXiv preprint arXiv:2308.03686 , year=

    Joe Benton, Valentin De Bortoli, Arnaud Doucet, and George Deligiannidis. Nearlyd-linear convergence bounds for diffusion models via stochastic local- ization.CoRR, abs/2308.03686, 2023

    work page arXiv 2023

  6. [6]

    Sliced and radon wasserstein barycenters of measures.Journal of Mathematical Imag- ing and Vision, 51(1):22–45, Jan 2015

    Nicolas Bonneel, Julien Rabin, Gabriel Peyr´ e, and Hanspeter Pfister. Sliced and radon wasserstein barycenters of measures.Journal of Mathematical Imag- ing and Vision, 51(1):22–45, Jan 2015

    work page 2015

  7. [7]

    The maximum entropy distribution of an asset inferred from option prices.Journal of Financial and Quantitative Analysis, 31(1):143–159, 1996

    Peter W Buchen and Michael Kelly. The maximum entropy distribution of an asset inferred from option prices.Journal of Financial and Quantitative Analysis, 31(1):143–159, 1996

    work page 1996

  8. [8]

    Hanqun Cao, Cheng Tan, Zhangyang Gao, Guangyong Chen, Pheng-Ann Heng, and Stan Z. Li. A survey on generative diffusion models.arXiv e-prints, abs/2209.02646, 2022

    work page arXiv 2022

  9. [9]

    The sample size required in importance sampling.The Annals of Applied Probability, 28(2):1099–1135, 2018

    Sourav Chatterjee and Persi Diaconis. The sample size required in importance sampling.The Annals of Applied Probability, 28(2):1099–1135, 2018

    work page 2018

  10. [10]

    Sitan Chen, Sinho Chewi, Jerry Li, Yuanzhi Li, Adil Salim, and Anru R. Zhang. Sampling is as easy as learning the score: theory for diffusion models with minimal data assumptions, 2023

    work page 2023

  11. [11]

    Mccann, Marc L

    Hyungjin Chung, Jeongsol Kim, Michael T. Mccann, Marc L. Klasky, and Jong Chul Ye. Diffusion posterior sampling for general noisy inverse problems, 2024

    work page 2024

  12. [12]

    Tail-gan: Learn- ing to simulate tail risk scenarios.Management Science, 2025

    Rama Cont, Mihai Cucuringu, Renyuan Xu, and Chao Zhang. Tail-gan: Learn- ing to simulate tail risk scenarios.Management Science, 2025

    work page 2025

  13. [13]

    Diffusion models in vision: A survey.arXiv e-prints, abs/2209.04747, 2022

    Florinel-Alin Croitoru, Vlad Hondru, Radu Tudor Ionescu, and Mubarak Shah. Diffusion models in vision: A survey.arXiv e-prints, abs/2209.04747, 2022

    work page arXiv 2022

  14. [14]

    On topology properties of f-divergences.Studia Scientifica Math- ematica Hungerica, 2:329–339, 1967

    Imre Csisz´ ar. On topology properties of f-divergences.Studia Scientifica Math- ematica Hungerica, 2:329–339, 1967

    work page 1967

  15. [15]

    Axiomatic characterizations of information measures.Entropy, 10(3):261–273, 2008

    Imre Csisz´ ar. Axiomatic characterizations of information measures.Entropy, 10(3):261–273, 2008

    work page 2008

  16. [16]

    Diffusion models beat GANs on image synthesis

    Prafulla Dhariwal and Alexander Quinn Nichol. Diffusion models beat GANs on image synthesis. In Marc’Aurelio Ranzato, Alina Beygelzimer, Yann N. Dauphin, Percy Liang, and Jennifer Wortman Vaughan, editors,Advances in Neural Information Processing Systems 34, pages 8780–8794, Virtual Confer- ence, 2021. Curran Associates, Inc

    work page 2021

  17. [17]

    Asymptotic minimax character of the sample distribution function and of the classical multinomial estimator.The Annals of Mathematical Statistics, 27(3):642–669, 1956

    Aryeh Dvoretzky, Jack Kiefer, and Jacob Wolfowitz. Asymptotic minimax character of the sample distribution function and of the classical multinomial estimator.The Annals of Mathematical Statistics, 27(3):642–669, 1956

    work page 1956

  18. [18]

    On the rate of convergence in wasserstein distance of the empirical measure, 2013

    Nicolas Fournier and Arnaud Guillin. On the rate of convergence in wasserstein distance of the empirical measure, 2013

    work page 2013

  19. [19]

    Nguyen, and Lingjiong Zhu

    Xuefeng Gao, Hoang M. Nguyen, and Lingjiong Zhu. Wasserstein convergence guarantees for a general class of score-based generative models.Journal of Machine Learning Research, 26(43):1–54, 2025. Published version with con- vergence bounds for SGMs in 2-Wasserstein distance

    work page 2025

  20. [20]

    Gerber and E.S.W

    H.U. Gerber and E.S.W. Shiu. Option pricing by esscher transforms, 1994

    work page 1994

  21. [21]

    Goll and Ludger Rueschendorf.Minimal distance martingale measures and optimal portfolios consistent with observed market prices

    T. Goll and Ludger Rueschendorf.Minimal distance martingale measures and optimal portfolios consistent with observed market prices. Taylor and Francis, 01 2002

    work page 2002

  22. [22]

    Estimation of non-normalized statistical models by score matching.Journal of Machine Learning Research, 6:695–709, 2005

    Aapo Hyv¨ arinen. Estimation of non-normalized statistical models by score matching.Journal of Machine Learning Research, 6:695–709, 2005

    work page 2005

  23. [23]

    Fundamental limits for weighted empirical approximations of tilted distribu- tions.CoRR, abs/2512.23979, 2025

    Sarvesh Ravichandran Iyer, Himadri Mandal, Dhruman Gupta, Rushil Gupta, Agniv Bandhyopadhyay, Achal Bassamboo, Varun Gupta, and Sandeep Juneja. Fundamental limits for weighted empirical approximations of tilted distribu- tions.CoRR, abs/2512.23979, 2025. 16 GENERATING DDPM-BASED SAMPLES FROM TILTED DISTRIBUTIONS

    work page arXiv 2025

  24. [24]

    Kingma, Tim Salimans, Ben Poole, and Jonathan Ho

    Diederik P. Kingma, Tim Salimans, Ben Poole, and Jonathan Ho. Varia- tional diffusion models. In Marc’Aurelio Ranzato, Alina Beygelzimer, Yann N. Dauphin, Percy Liang, and Jennifer Wortman Vaughan, editors,Advances in Neural Information Processing Systems 34, pages 21696–21707, Virtual Con- ference, 2021. Curran Associates, Inc

    work page 2021

  25. [25]

    Ro- bust optimization with diffusion models for green security.arXiv preprint arXiv:2503.05730, 2025

    Lingkai Kong, Haichuan Wang, Yuqi Pan, Cheol Woo Kim, Mingxiao Song, Alayna Nguyen, Tonghan Wang, Haifeng Xu, and Milind Tambe. Ro- bust optimization with diffusion models for green security.arXiv preprint arXiv:2503.05730, 2025

    work page arXiv 2025

  26. [26]

    Exponentially tilted thermodynamic maps (exptm): Predicting phase transitions across tem- perature, pressure, and chemical potential, 2025

    Suemin Lee, Ruiyu Wang, Lukas Herron, and Pratyush Tiwary. Exponentially tilted thermodynamic maps (exptm): Predicting phase transitions across tem- perature, pressure, and chemical potential, 2025

    work page 2025

  27. [27]

    Convergence and concentration of empirical measures under Wasser- stein distance in unbounded functional spaces.Bernoulli, 26(1), 2020

    Jing Lei. Convergence and concentration of empirical measures under Wasser- stein distance in unbounded functional spaces.Bernoulli, 26(1), 2020

    work page 2020

  28. [28]

    Ander- son

    Lizao Li, Robert Carver, Ignacio Lopez-Gomez, Fei Sha, and John R. Ander- son. Generative emulation of weather forecast ensembles with diffusion models. Science Advances, 10(13):eadk4489, 2024

    work page 2024

  29. [29]

    Behera, Dachao Jin, Baoxiang Pan, Huidong Jiang, Toshio Yamagata, et al

    Fenghua Ling, Zeyu Lu, Jing-Jia Luo, Lei Bai, Swadhin K. Behera, Dachao Jin, Baoxiang Pan, Huidong Jiang, Toshio Yamagata, et al. Diffusion model-based probabilistic downscaling for 180-year East Asian climate reconstruction.npj Climate and Atmospheric Science, 7:131, 2024

    work page 2024

  30. [30]

    Yaron Lipman, Ricky T. Q. Chen, Heli Ben-Hamu, Maximilian Nickel, and Matt Le. Flow matching for generative modeling. InInternational Conference on Learning Representations (ICLR) 2023, Kigali, Rwanda, 2023. OpenRe- view.net / ICLR

    work page 2023

  31. [31]

    Flow straight and fast: Learn- ing to generate and transfer data with rectified flow

    Xingchao Liu, Chengyue Gong, and Qiang Liu. Flow straight and fast: Learn- ing to generate and transfer data with rectified flow. InInternational Confer- ence on Learning Representations (ICLR) 2023, Kigali, Rwanda, 2023. Open- Review.net / ICLR

    work page 2023

  32. [32]

    Fully flexible views: Theory and practice, 2010

    Attilio Meucci. Fully flexible views: Theory and practice, 2010

    work page 2010

  33. [33]

    Art B. Owen. Safe and effective importance sampling.Journal of the American Statistical Association, 95(449):135–143, 2000

    work page 2000

  34. [34]

    Rubinstein.Simulation and the Monte Carlo Method

    Reuven Y. Rubinstein.Simulation and the Monte Carlo Method. John Wiley & Sons, 1981

    work page 1981

  35. [35]

    Springer New York, 2012

    Albert N Shiryaev.Problems in Probability. Springer New York, 2012

    work page 2012

  36. [36]

    Deep unsupervised learning using nonequilibrium thermodynamics

    Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. In Francis Bach and David M. Blei, editors,Proceedings of the 32nd International Con- ference on Machine Learning, volume 37 ofProceedings of Machine Learning Research, pages 2256–2265, Lille, France, 2015. PMLR

    work page 2015

  37. [37]

    Denoising Diffusion Implicit Models

    Jiaming Song, Chenlin Meng, and Stefano Ermon. Denoising Diffusion Implicit Models. InProceedings of the International Conference on Learning Represen- tations (ICLR) 2021, 2021

    work page 2021

  38. [38]

    Loss-guided diffusion models for plug-and-play controllable generation

    Jiaming Song, Qinsheng Zhang, Hongxu Yin, Morteza Mardani, Ming-Yu Liu, Jan Kautz, Yongxin Chen, and Arash Vahdat. Loss-guided diffusion models for plug-and-play controllable generation. In Andreas Krause, Emma Brunskill, Kyunghyun Cho, Barbara Engelhardt, Sivan Sabato, and Jonathan Scarlett, editors,Proceedings of the 40th International Conference on Mac...

    work page 2023

  39. [39]

    Kingma, Abhishek Kumar, Ste- fano Ermon, and Ben Poole

    Yang Song, Jascha Sohl-Dickstein, Diederik P. Kingma, Abhishek Kumar, Ste- fano Ermon, and Ben Poole. Score-based generative modeling through stochas- tic differential equations, 2021

    work page 2021

  40. [40]

    A simple nonparametric approach to derivative security val- uation.The Journal of Finance, 51(5):1633–1652, 1996

    Michael Stutzer. A simple nonparametric approach to derivative security val- uation.The Journal of Finance, 51(5):1633–1652, 1996

    work page 1996

  41. [41]

    A. W. van der Vaart.Asymptotic Statistics. Cambridge Series in Statistical and Probabilistic Mathematics. Cambridge University Press, 1998

    work page 1998

  42. [42]

    A connection between score matching and denoising autoen- coders.Neural Computation, 23(7):1661–1674, 2011

    Pascal Vincent. A connection between score matching and denoising autoen- coders.Neural Computation, 23(7):1661–1674, 2011

    work page 2011

  43. [43]

    Protein conformation generation via force-guided SE(3) diffusion models, 2024

    Yan Wang, Lihao Wang, Yuning Shen, Yiqun Wang, Huizhuo Yuan, Yue Wu, and Quanquan Gu. Protein conformation generation via force-guided SE(3) diffusion models, 2024

    work page 2024

  44. [44]

    1 1 n Pn i=1 exp(θT g(xi)) 2 # ≤E

    Ling Yang, Zhilong Zhang, Shenda Hong, Wentao Zhang, and Bin Cui. Dif- fusion models: A comprehensive survey of methods and applications.arXiv e-prints, abs/2209.00796, 2022. 18 GENERATING DDPM-BASED SAMPLES FROM TILTED DISTRIBUTIONS AppendixA.PROOFS In this section, we prove all the results in our article. A.1.TILTING AS MINIMIZATION OF ENTROPIC DIVER- G...

    work page arXiv 2022