The Limits of Closed-Form Couplings for Few-Step Generation
Organizations: Technical University of Munich · Helmholtz AI · Munich Center for Machine Learning · University of Texas at Austin ·
Abstract
Few-step generators via flow models require complicated technical solutions such as a pretrained flow model or a consistency loss. While rectified flows have gained popularity for their success in few-step sampling, they pay a cost of needing a pre-trained model and sampling a large number of couplings per dataset. Yet for a finite dataset and a Gaussian probability path, the probability-flow vector field of the empirical distribution is available in closed form, so one could use this field as the teacher and skip pretraining altogether. We demonstrate that, surprisingly, this approach is impossible for problems of practical relevance. In particular, the analytic solution of the diffusion process in high dimensions becomes almost a linear transformation. The integration returns the noise vector drawn at initialisation, plus a drift that is empirically a faint copy of the data point, almost independent of the noise. The coupling therefore differs from an independent pairing only through this drift. As a result, training a model on these couplings is almost equivalent to training the model on the independent draws of the noise and the data, which does not provide a useful teacher. We highlight that this is the direct consequence of the high-dimensionality of the data, and that for low enough dimensions or large enough datasets, analytic couplings become more efficient teachers. We quantify this effect by the fraction of endpoint energy carried by the drift, which scales as for a dataset of size in a -dimensional space. Finally, we provide an empirical study confirming all the theoretical results presented in the paper.
Figures & tables
Appendix figures & tables17 assets
Supplementary material from the paper’s appendix.
Appendix
| dataset | coupling | noise share | NN match | |
|---|---|---|---|---|
| CIFAR-10 | teacher | |||
| closed-form | ||||
| CelebA | teacher | |||
| closed-form | ||||
| ImageNet | teacher | |||
| closed-form |
| dataset | px | 64 steps | 128 steps | |
|---|---|---|---|---|
| CIFAR-10 | 16 | 768 | 2.6 | 2.1 |
| CIFAR-10 | 24 | 1728 | 3.4 | 3.1 |
| CIFAR-10 | 32 | 3072 | 3.6 | 3.1 |
| CelebA | 16 | 768 | 1.5 | 1.2 |
| CelebA | 32 | 3072 | 2.6 | 2.7 |
| CelebA | 64 | 12288 | 4.4 | 3.4 |
| dataset | px | 1 step | 2 steps | 1 step | 2 steps | 1 step | 2 steps |
|---|---|---|---|---|---|---|---|
| CIFAR-10 | 16 | 61.0 1.9 | 38.1 0.7 | 31.3 1.2 | 20.2 1.1 | 4.6 0.0 | 3.9 0.0 |
| CIFAR-10 | 24 | 72.7 0.9 | 49.1 0.6 | 51.5 1.5 | 36.5 1.1 | 9.3 0.2 | 8.3 0.3 |
| CIFAR-10 | 32 | 82.9 0.7 | 60.7 6.1 | 68.0 2.3 | 46.5 2.5 | 15.2 0.5 | 12.4 0.3 |
| CelebA | 16 | 38.3 1.0 | 29.2 0.8 | 21.3 0.1 | 14.7 0.1 | 2.0 0.1 | 2.0 0.0 |
| CelebA | 32 | 43.1 2.1 | 29.6 1.1 | 29.1 1.8 | 20.2 0.7 | 4.8 0.0 | 6.4 0.2 |
| dataset | px | |||||
|---|---|---|---|---|---|---|
| CIFAR-10 | 16 | 768 | 0.000 | 0.120 | 0.335 | 0.852 |
| CIFAR-10 | 24 | 1728 | 0.000 | 0.082 | 0.284 | 0.861 |
| CIFAR-10 | 32 | 3072 | 0.000 | 0.062 | 0.230 | 0.867 |
| CelebA | 16 | 768 | 0.000 | 0.103 | 0.242 | 0.703 |
| CelebA | 32 | 3072 | 0.000 | 0.054 | 0.171 | 0.730 |
| CelebA | 64 | 12288 | 0.000 | 0.028 | 0.108 | 0.746 |
| dataset | px | train | test | train | test | train | test | |
|---|---|---|---|---|---|---|---|---|
| CIFAR-10 | 16 | 768 | 0.01 | 0.00 | 0.54 | 0.52 | 0.86 | 0.85 |
| CIFAR-10 | 24 | 1728 | 0.03 | -0.01 | 0.52 | 0.48 | 0.84 | 0.82 |
| CIFAR-10 | 32 | 3072 | 0.05 | -0.01 | 0.52 | 0.45 | 0.81 | 0.78 |
| CelebA | 16 | 768 | 0.00 | 0.00 | 0.60 | 0.60 | 0.84 | 0.84 |
| CelebA | 32 | 3072 | 0.02 | -0.01 | 0.55 | 0.53 | 0.79 | 0.78 |
| dataset | px | train | test | train | test | train | test | |
|---|---|---|---|---|---|---|---|---|
| CIFAR-10 | 16 | 768 | 0.96 | 0.54 | 0.97 | 0.44 | 1.00 | 0.98 |
| CIFAR-10 | 24 | 1728 | 0.95 | 0.51 | 0.96 | 0.40 | 1.00 | 0.96 |
| CIFAR-10 | 32 | 3072 | 0.92 | 0.59 | 0.92 | 0.33 | 0.99 | 0.92 |
| CelebA | 16 | 768 | 0.95 | 0.52 | 0.97 | 0.61 | 1.00 | 0.99 |
| CelebA | 32 | 3072 | 0.89 | 0.50 | 0.92 | 0.52 | 1.00 | 0.98 |
| dataset | px | directions | variance | directions | variance | |
|---|---|---|---|---|---|---|
| CIFAR-10 | 16 | 768 | 5 | 238 | ||
| CIFAR-10 | 24 | 1728 | 3 | 174 | ||
| CIFAR-10 | 32 | 3072 | 3 | 154 | ||
| CelebA | 16 | 768 | 6 | 254 | ||
| CelebA | 32 | 3072 | 6 | 425 | ||
| dataset | pairs | 1 step | 2 steps | 1 step | 2 steps | 1 step | 2 steps | 1 step | 2 steps |
|---|---|---|---|---|---|---|---|---|---|
| ImageNet-16 | M | 430.1 | 158.0 | 97.5 | 44.2 | 2.3 | 1.6 | 0.78 | 0.73 |
| ImageNet-16 | k | 53.8 | 28.2 | 35.3 | 25.1 | 12.1 | 11.6 | 0.44 | 0.18 |
| latent | rFID | 1 step | 2 steps | 1 step | 2 steps | 1 step | 2 steps | 1 step | 2 steps | |
|---|---|---|---|---|---|---|---|---|---|---|
| 256 | 0.95 | 69.3 | 41.7 | 43.8 | 32.4 | 17.7 | 17.7 | 0.49 | 0.39 | |
| 512 | 0.55 | 71.1 | 40.3 | 49.4 | 34.9 | 20.8 | 20.9 | 0.43 | 0.28 | |
| 1024 | 0.37 | 71.2 | 56.0 | 64.8 | 48.2 | 24.1 | 23.1 | 0.14 | 0.24 | |
| px | width | parameters | per dimension | |||||
|---|---|---|---|---|---|---|---|---|
| 32 | 3072 | 128 | 56.4M | 18.4k | 85.1 | 69.2 | 14.7 | 0.23 |
| 24 | 1728 | 128 | 56.4M | 32.6k | 74.3 | 50.8 | 9.5 | 0.36 |
| 24 | 1728 | 96 | 31.7M | 18.4k | 84.8 | 58.5 | 9.1 | 0.35 |
| 16 | 768 | 128 | 56.4M | 73.4k | 62.9 | 33.0 | 4.7 | 0.51 |
| 16 | 768 | 64 | 14.1M | 18.4k | 93.9 | 52.1 | 4.7 | 0.47 |
| dataset | px | |||
|---|---|---|---|---|
| CIFAR-10 | 16 | 768 | ||
| CIFAR-10 | 24 | 1728 | ||
| CIFAR-10 | 32 | 3072 | ||
| CelebA | 16 | 768 | ||
| CelebA | 32 | 3072 | ||
| CelebA | 64 | 12288 |
Flow-Based Generative Modeling
Flow-based generative modeling learns probability transports from simple noise to data for fast sampling, density estimation, and controllable generation.
See all 610 papers