AI Paper Atlas

Research map / Optimization / Implicit bias and generalization

LDT: Layer-Decomposition Training Makes Networks More Generalizable

Zaizuo Tang, Zongqi Yang, Yu-Bin Yang

ICLR 2026 · Optimization › Implicit bias and generalization · approach: Shallow ReLU network convergence

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 Layer-Decomposition Training (without DPU) (ablation)MambaIR · DRealSR SR, Olympus source, Canon targetPSNR: 30.93 -> 32.33Improved
add Layer-Decomposition Training (without DPU) (ablation)MambaIR · DRealSR SR, Olympus source, Sony targetPSNR: 30.81 -> 31.25Improved
add Layer-Decomposition Training (without DPU) (ablation)MambaIR · DRealSR SR, Olympus source, Pan targetSSIM: 0.8688 -> 0.8631Hurt
add Dynamic Parameter Update to LDT (ablation)MambaIR · DRealSR SR, Olympus source, Sony targetPSNR: 31.25 -> 32.15Improved
add Dynamic Parameter Update to LDT (ablation)MambaIR · DRealSR SR, Olympus source, Canon targetPSNR: 32.33 -> 32.80Improved
partition layers by gradient variance instead of randomly (ablation)MambaIR · DRealSR SR, Olympus source, Sony targetPSNR: 30.88 -> 32.15Improved
partition layers by gradient variance instead of mean gradient magnitude (ablation)MambaIR · DRealSR SR, Olympus source, Sony targetPSNR: 31.86 -> 32.15Improved
partition by normalized variance (Var/Mean) instead of variance (ablation)MambaIR · DRealSR SR, Olympus source, Sony targetPSNR: 32.15 -> 31.89No clear effect
reassign stable/unstable layers dynamically every 5,000 steps instead of static partition (ablation)MambaIR · DRealSR SR, Olympus source, Sony targetPSNR: 32.15 -> 31.85Mixed
project variance rank into update coefficients instead of normalized variance (ablation)MambaIR · DRealSR SR, Olympus source, Sony targetPSNR: 31.01 -> 32.15Improved
use single-branch LDT-S instead of dual-branch LDT (efficiency) (ablation)MambaIR · Training memory, 48x48 patchestraining memory (GB): 20.25 -> 15.21Improved
use single-branch LDT-S instead of dual-branch LDT (ablation)MambaIR · DRealSR SR, Olympus source, Sony targetPSNR: 32.15 -> 31.46Hurt

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