AI Paper Atlas

Research map / Optimization / Implicit bias and generalization

Random Smoothing Regularization in Kernel Gradient Descent Learning

Liang Ding, Tianyang Hu, Jiahang Jiang, Donghao Li, Wenjia Wang, Yuan Yao

JMLR 2024 · Optimization › Implicit bias and generalization · approach: Low-rank and single-index recovery · 6 citations (OpenAlex)

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 Gaussian random smoothing (ablation)2-layer ReLU MLP trained with SGD · Synthetic D=1, early stopping, n=50test l2 loss: 2.2168e-03 -> 1.3468e-03Improved
add Laplace random smoothing (ablation)2-layer ReLU MLP trained with SGD · Synthetic D=1, weight decay, n=200test l2 loss: 1.1135e-03 -> 8.2806e-04Improved
add Gaussian random smoothing (ablation)2-layer ReLU MLP trained with SGD · Synthetic D=2, early stopping, n=50test l2 loss: 1.2628e-02 -> 6.7205e-03Improved
add Laplace random smoothing (ablation)2-layer ReLU MLP trained with SGD · Synthetic D=2, early stopping, n=200test l2 loss: 3.1396e-03 -> 1.7674e-03Improved
add Gaussian random smoothing (ablation)2-layer ReLU MLP trained with SGD · Synthetic D=3, weight decay, n=50test l2 loss: 2.0987e-02 -> 1.6498e-02Improved
add Laplace random smoothing (ablation)2-layer ReLU MLP trained with SGD · Synthetic D=3, early stopping, n=200test l2 loss: 4.9460e-03 -> 3.8615e-03Improved
use Gaussian instead of Laplace noise (ablation)2-layer ReLU MLP trained with SGD · Synthetic D=1, early stopping, n=50test l2 loss: 2.0638e-03 -> 1.3468e-03Improved
use Laplace instead of Gaussian noise (ablation)2-layer ReLU MLP trained with SGD · Synthetic D=1, weight decay, n=50test l2 loss: 1.7466e-03 -> 1.6765e-03Improved
vary smoothing scale (U-shaped) (ablation)2-layer ReLU MLP trained with SGD · Synthetic D=1test l2 loss: U-shaped curveMixed
increase training size from 50 to 200 (ablation)2-layer ReLU MLP trained with SGD · Synthetic D=1, optimal smoothing scaletest l2 loss: optimal scale decreasesImproved

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