Feature-guided score diffusion for sampling conditional densities

ICLR 2025 Conference Submission13356 Authors

Keywords: Conditional sampling, Representation learning, Score diffusion

Abstract:

Score diffusion methods can learn probability densities from samples. The score of the noise-corrupted density is estimated using a deep neural network, which is then used to iteratively transport a Gaussian white noise density to a target density. Variants for conditional densities have been developed, but correct estimation of the corresponding scores is difficult.
We avoid these difficulties by introducing an algorithm that operate by projecting the score onto the target class mean in a learned feature space.
The features and the projected score are computed using the same network, which is trained by optimizing a single denoising loss.
Learned feature vectors of same-class images are tightly clustered relative to those of different classes.
We show that feature class centroids provide a low-dimensional Euclidean embedding of the class conditional densities.
We demonstrate that, when trained on a dataset of mixed image classes,
this projected score can generate high quality and diverse samples from the conditioning class.
Conditional generation can be performed using feature vectors interpolated between those of the training set, demonstrating out-of-distribution generalization.

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.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

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: 13356