Training Diffusion Classifiers with Denoising Assistance

Chandramouli Shama Sastry, Sri Harsha Dumpala, Sageev Oore

Primary Area: generative models

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: Score-matching SDEs, Guided Diffusion, DDPMs, Semi-supervised Diffusion

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

TL;DR: A new training method for diffusion-guidance classifiers that improves both generative and discriminative performance. A semi-supervised guided-diffusion framework that we introduce shows additional benefits of the new training method.

Abstract: Score-matching and diffusion models have emerged as state-of-the-art generative models for both conditional and unconditional generation. Classifier-guided diffusion models are created by training a classifier on samples obtained from the forward-diffusion process (i.e., from data to noise). In this paper, we propose denoising-assisted (DA) classifiers wherein the diffusion classifier is trained using both noisy and denoised examples as simultaneous inputs to the model. We differentiate between denoising-assisted (DA) classifiers and noisy classifiers, which are diffusion classifiers that are only trained on noisy examples. Our experiments on Cifar10 and Imagenet show that DA-classifiers improve over noisy classifiers both quantitatively in terms of generalization to test data and qualitatively in terms of perceptually-aligned classifier-gradients and generative modeling metrics. We theoretically characterize the gradients of DA-classifiers to explain improved perceptual alignment. Building upon the observed generalization benefits of DA-classifiers, we propose and evaluate a semi-supervised framework for training diffusion classifiers and demonstrate improved generalization of DA-classifiers over noisy classifiers.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 8287