AI Paper Atlas

Research map / Image generation / Image restoration

Parameter Manifold Purification

Jiacong Hu, Jinxun Wu, Shengxuming Zhang, Shunyu Liu, Haofei Zhang, Mingli Song, Zunlei Feng

ICML 2026 · Image generation › Image restoration · approach: Diffusion priors for denoising and deblurring

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 changedWhereMeasuredEffect
add local manifold positional information (ablation)IMAE · parameter reconstructionaccuracy: 69.64 -> 75.32Improved
use PCD loss instead of MSE (ablation)IMAE · parameter reconstructionaccuracy: 61.42 -> 75.32Improved
use PCD loss instead of WP loss (ablation)IMAE · parameter reconstructionaccuracy: 10.86 -> 75.32Improved
use PCD loss instead of LWLN loss (ablation)IMAE · parameter reconstructionaccuracy: 67.25 -> 75.32Improved
tune intra-cluster weight alpha (ablation)PCD loss · parameter reconstructionaccuracy: peak at alpha = 3Mixed
increase inter-cluster weight beta (ablation)PCD loss · parameter reconstructionaccuracy: highest at beta = 2Hurt
apply PMP to long-tailed ViT-B/16 parametersViT-B/16 · ImageNet-LToverall accuracy: 78.0 -> 82.6Improved
apply PMP to noisy-label ResNet-34 parametersResNet-34 · CIFAR-10 (80% symmetric noise)accuracy: 67.43 -> 94.20Improved
apply PMP to backdoored PreAct-ResNet-18PreAct-ResNet-18 · Tiny-ImageNet BadNets-A2Oattack success rate: 99.96 -> 0.21Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy