AI Paper Atlas

Research map / Optimization / Implicit bias and generalization

A Unified Approach to Controlling Implicit Regularization via Mirror Descent

Haoyuan Sun, Khashayar Gatmiry, Kwangjun Ahn, Navid Azizan

JMLR 2023 · Optimization › Implicit bias and generalization · approach: Implicit regularization of SGD · 3 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 normalized step size (divide gradient by loss, decaying step) in p-GDp-GD p=1.5, fully connected net · MNIST testtest accuracy (%): 97.65 -> 98.37Improved
use normalized step size in p-GDp-GD p=1.5, conv net · MNIST testtest accuracy (%): 98.68 -> 99.19Improved
use p=3 instead of p=2 (SGD) in p-GDResNet-18 · CIFAR-10 testtest accuracy (%): 93.90 -> 94.01Improved
use p=3 instead of p=2 (SGD) in p-GDVGG-11 · CIFAR-10 testtest accuracy (%): 90.15 -> 90.85Improved
use p=3 instead of p=2 (SGD) in p-GDMobileNet-v2 · CIFAR-10 testtest accuracy (%): 91.97 -> 93.23Improved
use p=3 instead of p=2 (SGD) in p-GDRegNetX-200mf · CIFAR-10 testtest accuracy (%): 92.75 -> 94.07Improved
use sparse-inducing p=1.1 instead of p=2 (SGD) in p-GDVGG-11 · CIFAR-10 testtest accuracy (%): 90.15 -> 88.19Hurt
use p=1.1 instead of p=2 (SGD) in p-GDResNet-18 · CIFAR-10 testtest accuracy (%): 93.90 -> 92.63Hurt
use p=10 instead of p=3 in p-GDVGG-11 · CIFAR-10 testtest accuracy (%): 90.85 -> 88.78Hurt
use p=10 instead of p=3 in p-GDRegNetX-200mf · CIFAR-10 testtest accuracy (%): 94.07 -> 92.97Hurt
run p-GD with p=1.1 instead of p=2linear classifier, 15 points in R^100 · l_1.1 norm of rescaled classifiernorm size: 6.161 -> 5.477Improved
run p-GD with p=10 instead of p=2linear classifier, 15 points in R^100 · l_10 norm of rescaled classifiernorm size: 0.382 -> 0.280Improved

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