Research map / Label-efficient and robust learning / Self-supervised and contrastive
Evaluating the Representation Space of Diffusion Models via Self-Supervised Principles
Xiao Li, Yixuan Jia, Zekai Zhang, Xiang Li, Lianghe Shi, Jinxin Zhou, Zhihui Zhu, Liyue Shen, Qing Qu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| extract diffusion features at an intermediate noise level instead of very low or very high noise (ablation) | pretrained EDM (CIFAR10, CIFAR100) and SiT-XL/2 (ImageNet) · linear classification on diffusion features | test accuracy | Improved |
| train on full dataset longer (data-rich regime) | EDM on CIFAR10 and SiT-B/2 on ImageNet-256 · FID and ICR over training | ICR and FID | Improved |
| continue training past the ICR minimum on limited data | EDM on 4096-image CIFAR10 subset · CIFAR10 limited data | memorization ratio | Hurt |
| continue training past the ICR minimum on limited data | EDM on 10K ImageNet-64 and SiT-B/2 on 20K ImageNet-256 · ImageNet limited data | ICR and memorization ratio | Hurt |
| continue training on limited data (4096 images) | EDM on CIFAR10 · trace of invariant covariance | Tr(Sigma_s) and Tr(Sigma_xi) | Mixed |
| continue training on full CIFAR10 (50K images) | EDM on CIFAR10 · trace of invariant covariance | Tr(Sigma_s) and Tr(Sigma_xi) | Improved |
Closest papers by meaning
- Learning Diffusion Models with Flexible Representation Guidance (NeurIPS 2025 · similarity 0.90)
- Understanding Representation Dynamics of Diffusion Models via Low-Dimensional Modeling (NeurIPS 2025 · similarity 0.90)
- Sparsely Supervised Diffusion (NeurIPS 2026 · similarity 0.90)
- On the Edge of Memorization in Diffusion Models (NeurIPS 2025 · similarity 0.89)
- Why DiT Models Underperform as Representation Learners without Long Skip Connections (NeurIPS 2026 · similarity 0.89)
- Deconstructing Denoising Diffusion Models for Self-Supervised Learning (ICLR 2025 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
