AI Paper Atlas

Research map / Optimization / Optimizers

Searching for Optimal Per-Coordinate Step-sizes with Multidimensional Backtracking

Frederik Kunstner, Victor S. Portella, Mark Schmidt, Nick Harvey

NeurIPS 2023 · Optimization › Optimizers · approach: Second-order and structured preconditioning · 0 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
Use multidimensional backtracking instead of diagonal AdaGrad, diagonal BB, RPROP, GD-HD with line-searchGradient descent with line-search · cpusmall and breast-cancer (ill-conditioned)convergence per function/gradient evaluationImproved
Use multidimensional backtracking instead of plain gradient descentGradient descent · News20 (d ~ 10^6)convergence per oracle callImproved
Use diagonal Hessian preconditioning instead of no preconditioningGradient descent · cpusmallcondition number: 5\cdot 10^{13} -> 300Improved
Use ellipsoid variant of multidimensional backtracking instead of the globally optimal preconditioner with line-searchGradient descent on ill-conditioned linear regression · Synthetic linear regressionconvergenceImproved
Use plain gradient updates on the step-size hyperparameters (GD-HD) instead of a robust methodHypergradient descent · Ill-conditioned problemsstabilityHurt
Use diagonal BB or RPROP instead of diagonal Hessian preconditioningGradient descent · Some ill-conditioned problemsconvergenceMixed
Use multidimensional backtracking instead of gradient descent on a less ill-conditioned problemGradient descent · News20relative advantageMixed
Lower GD-HD initial step-size from 10^-3 to 10^-10GD-HD · All experimentsdivergence: 10^{-3} -> 10^{-10}Improved
Add line-search to diagonal AdaGradDiagonal AdaGrad · Deterministic settingcompetitivenessImproved

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