The success of modern score-based generative models is largely driven by diffusion-based training, which learns a collection of time-indexed score fields corresponding to progressively noisier versions of the data distribution. While such models achieve impressive sample quality, leveraging them for sound probabilistic inference---particularly sampling from arbitrary conditionals of the model---remains challenging. This difficulty arises because conditioning information is only available for clean data and not for the higher noise levels required to generate exact conditional samples. In this paper, we introduce DIffusion-free SCOre matching (DISCO), a principled approach that sidesteps the need for time-dependent score fields altogether. Our method is based on an objective that, while reminiscent of diffusion-based training, estimates only a single score of the (slightly perturbed) data distribution, for which the exact conditional score is readily obtained. In experiments, DISCO is competitive with state-of-the-art diffusion models in terms of sample quality. More importantly, DISCO yields a more faithful representation of the underlying data distribution and---crucially---enables accurate sampling from arbitrary conditionals, outperforming standard heuristic samplers. This capability opens the door to principled and flexible probabilistic reasoning with score-based models.
Figures & tables
Figure 1: Conditional sampling with a diffusion model vs. DISCO (a single score field) trained on a 2D GMM ( p∗ , left), for the query x2∼pθ(x2∣x1=1.75) . Gradient Guidance [ 12 ] , the Replacement Heuristic [ 33 ] and the Twisted Diffusion Sampler (TDS) [ 42 ] deviate substantially from the diffusion model’s true conditional (via rejection sampling), whereas DISCO, using tempered SMC on the exact conditional score ( 7 ), matches it. Details in Section 5.1 and Appendix H .
Figure 2: DISCO Training: We pick a helper sample x′=x(3) from D:={x(i)}i=14 and perturb it at level t to obtain xt . Diffusion training regresses (x(3)−xt)/σ(t)2 ; DISCO regresses (x−xt)/σ(0)2 , with x drawn from the zero-level posterior p0(x∣xt) ( 10 ), a softmax with logits di and temperature σ(0)2 . Note that x(3) need not have high posterior probability.
Figure 3
CIFAR-10
Wide
Narrow
Super-Resolve 2x
Altern. Lines
Half
Expand
Method
LPIPS ↓
SSIM ↑
LPIPS ↓
SSIM ↑
LPIPS ↓
SSIM ↑
LPIPS ↓
SSIM ↑
LPIPS ↓
SSIM ↑
LPIPS ↓
SSIM ↑
Replacement
0.255
0.658
0.064
0.916
0.431
0.516
0.187
0.767
0.296
0.549
0.607
0.094
Grad. Guidance
0.203
0.732
0.036
0.943
0.196
0.727
0.082
0.860
0.262
0.600
0.581
0.125
DPS
0.190
0.752
0.032
0.948
0.130
0.793
0.065
0.882
0.261
0.614
0.552
0.210
RePaint
0.225
0.732
0.078
0.914
0.339
0.676
0.184
0.806
0.285
0.593
0.627
0.140
TDS
0.232
0.760
0.053
0.937
0.255
0.746
0.124
0.857
0.308
0.623
0.630
0.230
Table 2: Quantitative Inpainting Results: On both CIFAR-10 and FFHQ-64, DISCO often outperforms post-hoc inpainting heuristics (Replacement, Gradient Guidance, DPS, RePaint and TDS) applied to a pre-trained diffusion model with EDM architecture. EDM Masked (gray) learns separate conditional scores at each noise level and is a strong task-specific reference, but its conditionals need not be consistent with a single joint distribution. DISCO outperforms it on CIFAR-10 and comes close on FFHQ-64 while deriving all conditionals from a single score field (consistent with a joint in the limit of an exactly learned score), and has better unconditional samples on FFHQ-64 (see Table 3 ). Bold: best among methods that sample conditionals of a single joint model; EDM Masked is excluded.
Dataset
EDM
EDM Masked
DISCO
CIFAR-10
1.97
2.85
3.18
FFHQ-64
2.39
3.18
2.54
Table 3: FID scores on unconditional image generation.
Figure 4: Qualitative Inpainting Results: Comparison of inpainting methods on FFHQ-64 with the samplers of Table 2 ; Input dims the unobserved pixels, Truth shows the original image. Best viewed zoomed in. More results (conditional and unconditional) can be found in Appendix J .
Figure 5: Resampling Consistency Test: FID-10k (log scale) under repeated resampling of random masks, starting from each model’s samples.
Appendix figures & tables9 assets
Supplementary material from the paper’s appendix.
Appendix
Batch size
Moons
Checkerboard
Rings
64
0.085±0.031
0.038±0.011
0.340±0.040
128
0.081±0.018
0.032±0.008
0.230±0.039
256
0.111±0.046
0.039±0.011
0.170±0.030
512
0.093±0.039
0.032±0.007
0.117±0.021
Full dataset
0.068±0.022
0.029±0.008
0.066±0.026
Appendix
Table 4: Mini-batch posterior, model fit: W1(p0(x),pθ(x)) for DISCO trained with the zero-level posterior computed over mini-batches of the given size, or exactly over the full dataset (as in Table 3 ).
Batch size
Moons
Checkerboard
Rings
64
0.023±0.024
0.016±0.019
0.023±0.028
128
0.031±0.036
0.014±0.013
0.018±0.020
256
0.027±0.029
0.016±0.017
0.033±0.045
512
0.024±0.024
0.014±0.014
0.031±0.029
Full dataset
0.024±0.025
0.015±0.012
0.042±0.040
Appendix
Table 5: Mini-batch posterior, inference quality: Ep0(x1)[W1(pθ(x2∣x1),p^θ(x2∣x1))] for the same models as in Table 4 , with pθ the exact model conditional (via rejection sampling) and p^θ obtained by tempered SMC on the conditional score ( 7 ).
Figure 6: One chain of the resampling consistency test (seed 0) per method, shown every 10 rounds. DISCO and EDM Masked start from their own samples, the post-hoc samplers from the same sample of the pretrained EDM. Under a consistent sampler the image content may change from round to round (the chain mixes), but its quality should not degrade.
Figure 7: Unconditional samples from the DISCO model trained on FFHQ-64.
Figure 8: Unconditional samples from the diffusion model trained on FFHQ-64.
Figure 9: Qualitative inpainting results on FFHQ-64 (images 16–21 of the evaluation set, one per mask), with the samplers of Table 2 ; Input dims the unobserved pixels, Truth shows the original image.
Figure 10: Qualitative inpainting results on FFHQ-64 (images 10–15 of the evaluation set, one per mask), with the samplers of Table 2 ; Input dims the unobserved pixels, Truth shows the original image.
Figure 11: Unconditional samples from the DISCO model trained on CIFAR-10.
Figure 12: Unconditional samples from the diffusion model trained on CIFAR-10.
Score-based generative modeling learns data-distribution gradients to generate samples through reverse diffusion, flows, and related stochastic dynamics.
We establish finite-sample guarantees for conditional diffusion models learned from data and implemented with finitely many denoising steps. Our approach aligns the score-training objective with the sampling grid, allowing statistical estimation and numerical discretization errors to be analyzed together. Under a smooth conditional density model, we derive high-probability bounds on the generated distribution's error and quantify the required number of score evaluations. For every smoothness index β>0, a suitable grid yields averaged conditional Kullback--Leibler (KL) divergence Op(n−2β/(dx+dy+2β)) relative to the original target with or without early stopping. Both the full and early-stopped samplers attain average target total-variation (TV) error Op(n−β/(dx+dy+2β)). Here n is the sample size, dx,dy are the data and conditioning dimensions, and errors are averaged over the conditioning variable. The total-variation bound matches the rate established in existing conditional distribution-estimation theory, up to logarithmic factors.
Qiuyun Zhu
Department of Statistics and Actuarial Science University of Iowa
Despite the empirical success of score-based diffusion models, a complete theoretical understanding of how finite-sample learning, network parameterization, and numerical discretization jointly dictate generative quality remains underdeveloped. Existing sampling analyses often evaluate the generative performance conditional on an oracle score or a pre-specified error threshold. In this work, we establish a unified convergence and generalization framework for score-based diffusion models parameterized by practical ResNet-type architectures. We analyze the generalization and convergence properties from the practical finite-sample, discrete-time learning problem of the score function to the ideal continuous-time, population-level objective. Based on the generalization result of the learning problem of score function, we analyze the sampling process induced by the learned score function and provide an end-to-end total variation distance estimate for the generated terminal distribution. This estimate explicitly decomposes the overall generative error into four interpretable components: the truncation error of the forward process, the reverse-time discretization error, the generalization error incorporating both finite data and forward-time discretization, and the training optimization gap. Our results quantitatively characterize how the training sample size, temporal discretization grids, and optimization accuracy jointly control the final fidelity of samples generated by diffusion models.
Jinshu Huang, Yiming Jiang, Chunlin Wu
School of Mathematical Sciences, Nankai University, Tianjin, China
We propose a conditioning mechanism for diffusion models based on multi-speed joint diffusion of the target and the condition. The mechanism learns an unconditional joint score network and enforces conditioning at inference via a plug-in correction term. The plug-in term separates the conditioning contribution from the learned unconditional dynamics, offering a transparent view of how the condition steers generation of the target distribution. Building on this, we derive explicit conditional reverse-time SDEs and approximate probability-flow ODEs, enabling principled and directly comparable conditional samplers. To reduce the induced ODE--SDE discrepancy, we introduce a log-Fokker--Planck residual regularization that improves ODE sampling quality. Experiments on conditional image generation tasks demonstrate competitive performance and support the effectiveness of the plug-in conditioning view. Additional ODE--SDE comparison experiments show that the log-Fokker--Planck residual regularization improves deterministic ODE sampling.
Libo Chen, Souvik Ghosh, Teo Deveney +2
University of Bath, Bath, UK · International Institute of Information Technology Hyderabad, India