GradCheck: Analyzing classifier guidance gradients for conditional diffusion sampling
This addresses a specific bottleneck in diffusion models for researchers and practitioners, but it is incremental as it focuses on improving existing methods rather than introducing a new paradigm.
The study tackled the problem of unstable gradients from classifiers in conditional diffusion sampling, finding that gradient stabilization techniques significantly improve sample quality for non-robust classifiers.
To sample from an unconditionally trained Denoising Diffusion Probabilistic Model (DDPM), classifier guidance adds conditional information during sampling, but the gradients from classifiers, especially those not trained on noisy images, are often unstable. This study conducts a gradient analysis comparing robust and non-robust classifiers, as well as multiple gradient stabilization techniques. Experimental results demonstrate that these techniques significantly improve the quality of class-conditional samples for non-robust classifiers by providing more stable and informative classifier guidance gradients. The findings highlight the importance of gradient stability in enhancing the performance of classifier guidance, especially on non-robust classifiers.