Classifier-Free Guidance is a Predictor-Corrector

• URL: http://arxiv.org/abs/2408.09000v2
• Date: Fri, 23 Aug 2024 17:21:35 GMT
• Title: Classifier-Free Guidance is a Predictor-Corrector
• Authors: Arwen Bradley, Preetum Nakkiran,
• Abstract summary: CFG is the dominant method of conditional sampling for text-to-image diffusion models. We disprove common misconceptions by showing that CFG interacts differently with DDPM and DDIM. We prove that in the SDE limit, CFG is actually equivalent to combining a DDIM predictor for the conditional distribution together with a Langevin dynamics corrector for a gamma-powered distribution.
• Score: 8.970133799609041
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: We investigate the theoretical foundations of classifier-free guidance (CFG). CFG is the dominant method of conditional sampling for text-to-image diffusion models, yet unlike other aspects of diffusion, it remains on shaky theoretical footing. In this paper, we disprove common misconceptions, by showing that CFG interacts differently with DDPM (Ho et al., 2020) and DDIM (Song et al., 2021), and neither sampler with CFG generates the gamma-powered distribution $p(x|c)^\gamma p(x)^{1-\gamma}$. Then, we clarify the behavior of CFG by showing that it is a kind of predictor-corrector method (Song et al., 2020) that alternates between denoising and sharpening, which we call predictor-corrector guidance (PCG). We prove that in the SDE limit, CFG is actually equivalent to combining a DDIM predictor for the conditional distribution together with a Langevin dynamics corrector for a gamma-powered distribution (with a carefully chosen gamma). Our work thus provides a lens to theoretically understand CFG by embedding it in a broader design space of principled sampling methods.

Related papers

• Contrastive CFG: Improving CFG in Diffusion Models by Contrasting Positive and Negative Concepts [55.298031232672734]
  As-Free Guidance (CFG) has proven effective in conditional diffusion model sampling for improved condition alignment. We present a novel method to enhance negative CFG guidance using contrastive loss.
  arXiv  Detail & Related papers  (2024-11-26T03:29:27Z)
• Rectified Diffusion Guidance for Conditional Generation [62.00207951161297]
  We revisit the theory behind CFG and rigorously confirm that the improper configuration of the combination coefficients (i.e., the widely used summing-to-one version) brings about expectation shift of the generative distribution. We propose ReCFG with a relaxation on the guidance coefficients such that denoising with ReCFG strictly aligns with the diffusion theory. That way the rectified coefficients can be readily pre-computed via traversing the observed data, leaving the sampling speed barely affected.
  arXiv  Detail & Related papers  (2024-10-24T13:41:32Z)
• Theory on Score-Mismatched Diffusion Models and Zero-Shot Conditional Samplers [49.97755400231656]
  We present the first performance guarantee with explicit dimensional general score-mismatched diffusion samplers. We show that score mismatches result in an distributional bias between the target and sampling distributions, proportional to the accumulated mismatch between the target and training distributions. This result can be directly applied to zero-shot conditional samplers for any conditional model, irrespective of measurement noise.
  arXiv  Detail & Related papers  (2024-10-17T16:42:12Z)
• Score-based generative models are provably robust: an uncertainty quantification perspective [4.396860522241307]
  We show that score-based generative models (SGMs) are provably robust to the multiple sources of error in practical implementation. Our primary tool is the Wasserstein uncertainty propagation (WUP) theorem. We show how errors due to (a) finite sample approximation, (b) early stopping, (c) score-matching objective choice, (d) score function parametrization, and (e) reference distribution choice, impact the quality of the generative model.
  arXiv  Detail & Related papers  (2024-05-24T17:50:17Z)
• Soft-constrained Schrodinger Bridge: a Stochastic Control Approach [4.922305511803267]
  Schr"odinger bridge can be viewed as a continuous-time control problem where the goal is to find an optimally controlled diffusion process. We propose to generalize this problem by allowing the terminal distribution to differ from the target but penalizing the Kullback-Leibler divergence between the two distributions. One application is the development of robust generative diffusion models.
  arXiv  Detail & Related papers  (2024-03-04T04:10:24Z)
• Broadening Target Distributions for Accelerated Diffusion Models via a Novel Analysis Approach [49.97755400231656]
  We show that a novel accelerated DDPM sampler achieves accelerated performance for three broad distribution classes not considered before. Our results show an improved dependency on the data dimension $d$ among accelerated DDPM type samplers.
  arXiv  Detail & Related papers  (2024-02-21T16:11:47Z)
• Adaptive Guidance: Training-free Acceleration of Conditional Diffusion Models [44.58960475893552]
  "Adaptive Guidance" (AG) is an efficient variant of computation-Free Guidance (CFG) AG preserves CFG's image quality while reducing by 25%. " LinearAG" offers even cheaper inference at the cost of deviating from the baseline model.
  arXiv  Detail & Related papers  (2023-12-19T17:08:48Z)
• Adaptive Annealed Importance Sampling with Constant Rate Progress [68.8204255655161]
  Annealed Importance Sampling (AIS) synthesizes weighted samples from an intractable distribution. We propose the Constant Rate AIS algorithm and its efficient implementation for $alpha$-divergences.
  arXiv  Detail & Related papers  (2023-06-27T08:15:28Z)
• A Convenient Infinite Dimensional Framework for Generative Adversarial Learning [4.396860522241306]
  We propose an infinite dimensional theoretical framework for generative adversarial learning. In our framework the Jensen-Shannon divergence between the distribution induced by the generator from the adversarial learning procedure and the data generating distribution converges to zero.
  arXiv  Detail & Related papers  (2020-11-24T13:45:17Z)
• Generative Modeling with Denoising Auto-Encoders and Langevin Sampling [88.83704353627554]
  We show that both DAE and DSM provide estimates of the score of the smoothed population density. We then apply our results to the homotopy method of arXiv:1907.05600 and provide theoretical justification for its empirical success.
  arXiv  Detail & Related papers  (2020-01-31T23:50:03Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.