AI Paper Atlas

Research map / Optimization / Nonconvex and smooth optimization

Projection Efficient Subgradient Method and Optimal Nonsmooth Frank-Wolfe Method

Kiran K Thekumparampil, Prateek Jain, Praneeth Netrapalli, Sewoong Oh

NeurIPS 2020 · Optimization › Nonconvex and smooth optimization · approach: Stochastic Frank-Wolfe methods · 10 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
combine Moreau-Yosida smoothing with accelerated gradient slidingNonsmooth convex optimization with a projection oracle · General nonsmooth convex optimizationFO calls: \Ord(\frac{(G^2+\sigma^2)\|x_0 - x^*\|^2}{\varepsilon^2})No clear effect
replace exact projections with Frank-Wolfe approximate projectionsMOPES, yielding MOLES · General nonsmooth convex optimizationLMO calls: \Ord(\sqrt{d}\varepsilon^{-2}) -> \Ord({\frac{G^2 D_\mathcal{X}^2}{\varepsilon^2}})Improved
replace exact projections with Frank-Wolfe approximate projectionsMOPES, yielding MOLES · General nonsmooth convex optimizationFO calls: \Ord(\frac{(G^2+\sigma^2)D_\mathcal{X}^2}{\varepsilon^2})No clear effect
use Moreau smoothing and ordinary accelerated gradient descent rather than joint optimization with gradient sliding (ablation)Moreau envelope optimization · General nonsmooth convex optimizationFO calls: \Ord({\varepsilon^{-3}}) vs. \Ord({\varepsilon^{-2}})Hurt
use a generic approximate proximal step with error \delta=\varepsilon instead of Prox-Slide (ablation)Inexact APGD · General nonsmooth convex optimizationFO calls: \Ord({G^3D_\mathcal{X}^3/\varepsilon^3}) vs. MOPES \Ord(\frac{(G^2+\sigma^2)\|x_0 - x^*\|^2}{\varepsilon^2})Hurt
use only the average of Prox-Slide iterates instead of the final iterate (ablation)MOPES · General nonsmooth convex optimizationFO calls: increases by a factor of \Ord(\varepsilon^{-1})Hurt
use MOPES instead of PGDLow-rank SVM on Imagewoof 2.0 · Imagewoof 2.0 low-rank SVMPO callsImproved
use MOLES instead of FW-PGD or RandFWLow-rank SVM on Imagewoof 2.0 · Imagewoof 2.0 low-rank SVMLMO callsImproved

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