AI Paper Atlas

Research map / Label-efficient and robust learning / OOD and anomaly detection

Winning Prize Comes from Losing Tickets: Improve Invariant Learning by Exploring Variant Parameters for Out-of-Distribution Generalization

Zhuo Huang, Muyang Li, Li Shen, Jun Yu, Chen Gong, Bo Han, Tongliang Liu

IJCV 2024 · Label-efficient and robust learning › OOD and anomaly detection · approach: Out-of-distribution generalization theory · 18 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
add EVIL sparse trainingERM · DomainBed average (ResNet-50)accuracy: 63.8 -> 66.2Improved
add EVIL sparse trainingDRO · DomainBed average (ResNet-50)accuracy: 61.9 -> 63.8Improved
add EVIL sparse trainingIRM · DomainBed average (ResNet-50)accuracy: 63.1 -> 64.6Improved
add EVIL sparse trainingSWAD · DomainBed average (ResNet-50)accuracy: 67.2 -> 67.7Improved
add EVIL sparse trainingSAM · DomainBed average (ResNet-50)accuracy: 65.6 -> 67.5Improved
add EVIL sparse trainingSAM · DomainBed TerraIncognitaaccuracy: 43.3 -> 50.5Improved
use EVIL instead of SparseIRM at 60% sparsitysparse invariant learning · DomainBed average (sparsity 60%)accuracy: 64.8 -> 66.3Improved
use EVIL-SAM instead of SparseIRM at 80% sparsitysparse invariant learning · DomainBed average (sparsity 80%)accuracy: 61.7 -> 65.4Improved
use domain-label loss instead of class-label loss to explore variant parameters (EVIL vs RigL) (ablation)EVIL · DomainBedaccuracyImproved
set alpha 0.2 and Delta T 300 instead of other values (ablation)EVIL · DomainBedaccuracyImproved
initialize mask by weight value instead of connection sensitivity or random (ablation)EVIL · DomainBedaccuracyImproved
add EVIL during CLIP ViT-B/16 finetuningCLIP ViT-B/16 · iWildCam OODaccuracy: 35.0 -> 37.4Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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