AI Paper Atlas

Research map / Optimization / Implicit bias and generalization

An algorithmic view of L2 regularization and some path-following algorithms

Yunzhang Zhu, Renxiong Liu

JMLR 2021 · Optimization › Implicit bias and generalization · approach: Implicit regularization of SGD

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
use proposed Newton path-following instead of Euler, Runge-Kutta or Rosset grid method at equal Newton-step budgetl2-regularized logistic regression path · Synthetic nonseparable and separable, (n,p) in (1000,500),(1000,1000),(1000,2000)global approximation error of solution pathImproved
choose grid points adaptively (Newton method) instead of equally spaced with C(t)=1/t (ablation)Rosset path-following method · Synthetic logistic regressionglobal approximation errorImproved
use Newton method instead of glmnet and gradient descentsolution path computation · Synthetic logistic regression, 100 simulationsruntime at small suboptimalityMixed
increase initial step size (ablation)Newton and gradient descent path methods · Synthetic logistic regression, (n,p) in (500,100),(500,500),(500,1000)approximate statistical risk curve (KL)Hurt
use second-order Runge-Kutta with larger constant step size (ablation)ODE-based path methods · Synthetic logistic regression risk curvesapproximate statistical risk curveNo clear effect

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