REVIEW 5 cited by
Exploring Diffusion Time-steps for Unsupervised Representation Learning
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
Representation learning is all about discovering the hidden modular attributes that generate the data faithfully. We explore the potential of Denoising Diffusion Probabilistic Model (DM) in unsupervised learning of the modular attributes. We build a theoretical framework that connects the diffusion time-steps and the hidden attributes, which serves as an effective inductive bias for unsupervised learning. Specifically, the forward diffusion process incrementally adds Gaussian noise to samples at each time-step, which essentially collapses different samples into similar ones by losing attributes, e.g., fine-grained attributes such as texture are lost with less noise added (i.e., early time-steps), while coarse-grained ones such as shape are lost by adding more noise (i.e., late time-steps). To disentangle the modular attributes, at each time-step t, we learn a t-specific feature to compensate for the newly lost attribute, and the set of all 1,...,t-specific features, corresponding to the cumulative set of lost attributes, are trained to make up for the reconstruction error of a pre-trained DM at time-step t. On CelebA, FFHQ, and Bedroom datasets, the learned feature significantly improves attribute classification and enables faithful counterfactual generation, e.g., interpolating only one specified attribute between two images, validating the disentanglement quality. Codes are in https://github.com/yue-zhongqi/diti.
Forward citations
Cited by 5 Pith papers
-
MMDiff: Extending Diffusion Transformers for Multi-Modal Generation
Frozen DiT features fused across multiple denoising timesteps with concept attention and optional DINO yield competitive multi-modal dense predictions and strong synthetic training data.
-
CosmoFlow: Scale-Aware Representation Learning for Cosmology with Flow Matching
Flow matching on CDM fields yields an 8 number latent that reconstructs fields and estimates Omega_m and sigma_8 nearly as well as a raw-field network, with channels tied to spatial scales.
-
Harnessing Text-to-Image Diffusion Models for Point Cloud Self-Supervised Learning
PointSD uses a frozen Stable Diffusion model, conditioned on point clouds through rendered images, to generate training targets for point cloud self-supervised learning.
-
Controllable Coupled Image Generation via Diffusion Models
A cross-attention control method that couples backgrounds across multiple generated images by blending LLM-extracted background and entity prompts with a time-varying weight optimized for background similarity and tex...
-
On Denoising Walking Videos for Gait Recognition
DenoisingGait combines frozen Stable Diffusion features with learned direction-vector matching to create Gait Feature Fields, reporting new state-of-the-art rank-1 accuracy on CCPG and most settings of CASIA-B*, SUSTech1K.
Discussion (0). Sign in to comment.