AI Paper Atlas

Research map / Optimization / Implicit bias and generalization

Shadowing Properties of Optimization Algorithms

Antonio Orvieto, Aurelien Lucchi

NeurIPS 2019 · Optimization › Implicit bias and generalization · approach: Edge-of-stability dynamics · 9 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
Allow the shadowing orbit to start from a slightly perturbed initial conditionGD and Heavy-ball ODE approximations · Strongly-convex objectives and hyperbolic regionsuniform trajectory closenessImproved
Start Gradient Descent at the exact ODE initial condition instead of perturbing itGradient Descent near an unstable saddle direction · Hyperbolic saddletrajectory deviationHurt
Increase regularizationGradient Descent on MNIST digits 3 and 5 · MNIST digits 3 and 5, regularized sigmoid losstrajectory approximation errorMixed
Use Heavy-ball momentum rather than Gradient DescentMNIST digits 3 and 5 · MNIST digits 3 and 5, regularized sigmoid lossconvergence speedImproved
Use Heavy-ball momentum rather than Gradient DescentMNIST digits 3 and 5 · MNIST digits 3 and 5, regularized sigmoid lossshadowing radius: 5 times biggerHurt
Use Heavy-ball instead of Gradient DescentMNIST digits 3 and 5 · MNIST digits 3 and 5, regularized sigmoid lossdecay speed of approximation errorImproved
Use stochastic gradients with bounded gradient errorGradient Descent ODE approximation · Strongly-convex objectivesshadowing: h≤2(εμ-R)/(ℓL)Improved

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