AI Paper Atlas

Research map / Label-efficient and robust learning / Self-supervised and contrastive

Self-Soupervision: Cooking Model Soups without Labels

Anthony Fuller, James R Green, Evan Shelhamer

ICML 2026 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Invariance-based self-supervised learning · 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
mix ingredients inter-trained with different SSL algorithms instead of supervised-only ingredientsuniform-mix soup · ImageNet-Ctop-1 accuracy: 30.91 -> 32.34Improved
mix ingredients inter-trained with different SSL algorithms instead of supervised-only ingredientsuniform-mix soup · LAION-Ctop-1 accuracy: 22.05 -> 22.92Improved
mix ingredients inter-trained with different SSL algorithmsuniform-mix soup · ImageNet-Valtop-1 accuracy: 79.12 -> 79.09No clear effect
add SSL inter-training on corrupted test distribution (disjoint samples) on ImageNet-CSelf-Soup · ImageNet-Ctop-1 accuracy: 32.34 -> 35.72Improved
add SSL inter-training on LAION-C distribution (disjoint samples)Self-Soup · LAION-Ctop-1 accuracy: 22.92 -> 29.48Improved
inter-train on the test samples themselves rather than disjoint samplesSelf-Soup on IN-C · ImageNet-Ctop-1 accuracy: 35.72 -> 36.02Improved
inter-train on the test samples themselves rather than disjoint samplesSelf-Soup on LA-C · LAION-Ctop-1 accuracy: 29.48 -> 30.32Improved
inter-train on both IN-C and LAION-C rather than a single corruption setSelf-Soup · ImageNet-Ctop-1 accuracy: 35.72 -> 34.43Hurt
replace continued SSL (single algorithm) soup with Self-Soup (varying algorithms)soup on LAION-C · LAION-Ctop-1 accuracy: 27.43 -> 29.48Improved
add test-time adaptation (SAR) to Self-Soup on ImageNet-CSelf-Soup on ImageNet-C · ImageNet-Ctop-1 accuracy: 35.72 -> 37.50Improved
use Self-Soup instead of supervised-only soup on mini-VTAB-Cuniform mix · mini-VTAB-C meantop-1 accuracy: 34.6 -> 35.1Improved
use Self-Soup instead of supervised-only soup on noise corruptionsSelf-Soup · VTAB noise corruptionstop-1 accuracy: -1% versus the supervised-only soupHurt

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