Research map / Image generation / Fast diffusion sampling
Addressing Negative Transfer in Diffusion Models
Hyojun Go, Jinyoung Kim, Yunsung Lee, Seunghyun Lee, Shinhyeok Oh, Hyeongdon Moon, Seungtaek Choi
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 |
|---|---|---|---|
| add interval clustering with Uncertainty Weighting | ADM · FFHQ | FID: 24.95 -> 20.19 | Improved |
| add interval clustering with Uncertainty Weighting | ADM · CelebA-HQ | FID: 22.27 -> 17.74 | Improved |
| add interval clustering with PCgrad | LDM · FFHQ | FID: 10.56 -> 9.189 | Improved |
| add interval clustering with NashMTL (SNR) on LDM | LDM · FFHQ | FID: 10.56 -> 10.33 | Improved |
| use UW instead of NashMTL/PCgrad | ADM · FFHQ | Recall: 0.3996 -> 0.3635 | Hurt |
| compare ANT-UW to MinSNR loss weighting | DiT-L/2 · ImageNet | FID: 9.58 -> 6.17 | Improved |
| combine interval clustering (UW + PCgrad) with P2 objective | LDM · FFHQ | FID-50k: 7.21 -> 5.84 | Improved |
| increase number of clusters k from 2 to 5 (timestep) (ablation) | LDM UW+PCgrad · FFHQ | FID-10K: 9.563 -> 9.151 | Improved |
| increase number of clusters k from 5 to 8 (gradient) (ablation) | LDM UW+PCgrad · FFHQ | FID-10K: 9.033 -> 9.145 | No clear effect |
| use only two clusters versus vanilla (gradient) (ablation) | LDM UW+PCgrad · FFHQ | FID-10K: 10.56 -> 9.634 | Improved |
| replace HOA-based task grouping with interval clustering (ablation) | LDM UW+PCgrad · FFHQ | FID-10k: 9.873 -> 9.033 | Improved |
| replace sophisticated MTL with Random Loss Weighting (timestep) (ablation) | ADM · FFHQ | FID: 24.95 -> 38.06 | Hurt |
Closest papers by meaning
- Denoising Task Routing for Diffusion Models (ICLR 2024 · similarity 0.92)
- Denoising Task Difficulty-based Curriculum for Training Diffusion Models (ICLR 2025 · similarity 0.91)
- DiffusionMTL: Learning Multi-Task Denoising Diffusion Model from Partially Annotated Data (CVPR 2024 · similarity 0.90)
- Diffusion Models for Multi-Task Generative Modeling (ICLR 2024 · similarity 0.88)
- Diffusion Tuning: Transferring Diffusion Models via Chain of Forgetting (NeurIPS 2024 · similarity 0.88)
- Label-Noise Robust Diffusion Models (ICLR 2024 · similarity 0.87)
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.
