AI Paper Atlas

Research map / Optimization / Implicit bias and generalization

What You See is What You Get: Principled Deep Learning via Distributional Generalization

Bogdan Kulynych, Yao-Yuan Yang, Yaodong Yu, Jarosław Błasiok, Preetum Nakkiran

NeurIPS 2022 · Optimization › Implicit bias and generalization · approach: Stability-based generalization analysis · 1 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
replace l2 regularization with Gaussian gradient noise (ablation)IS-SGD · CivilCommentsworst-group accuracy: 64.3 -> 71.9Improved
replace l2 regularization with Gaussian gradient noise (ablation)IS-SGD · iNaturalist2017worst-group accuracy: 70.6 -> 71.0Improved
replace l2 regularization with Gaussian gradient noise (ablation)IS-SGD · CelebAworst-group accuracy: 82.4 -> 84.9Improved
replace l2 regularization with Gaussian gradient noise (ablation)IW-SGD · CelebAworst-group accuracy: 89.0 -> 88.5No clear effect
replace l2 regularization with Gaussian gradient noise (ablation)IW-SGD · MultiNLIworst-group accuracy: 68.1 -> 69.7Mixed
replace l2 regularization with Gaussian gradient noise (ablation)gDRO · iNaturalist2017worst-group accuracy: 67.3 -> 56.4Hurt
replace l2 regularization with Gaussian gradient noise (ablation)gDRO · CelebAworst-group accuracy: 84.5 -> 83.3Hurt
replace l2 regularization with Gaussian gradient noise (ablation)gDRO · CivilCommentsworst-group accuracy: 67.3 -> 71.3Improved
add DP (clipping and noise) to importance sampling (ablation)IS-SGD · CelebAworst-group accuracy: 82.4 -> 86.0Improved
add DP (clipping and noise) to importance sampling (ablation)IS-SGD · MultiNLIworst-group accuracy: 70.4 -> 72.3Improved
add DP (clipping and noise) to importance sampling (ablation)IS-SGD · iNaturalist2017worst-group accuracy: 70.6 -> 51.4Hurt
use importance sampling instead of DP-SGD uniform sampling (ablation)DP-SGD · CelebA, epsilon 2 to 12test accuracy: around 8 p.p. lowerHurt

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