AI Paper Atlas

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

IdEst: Assessing Self-Supervised Learning Representations via Intrinsic Dimension

Julie Mordacq, Vicky Kalogeiton, Steve Oudot

ICML 2026 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Data-efficient and robust self-supervision · 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
add IdEst-based selection instead of picking at random from the hyperparameter poolI-JEPA ViT-B · ImageNet, all hyperparameterstop-1 accuracy of selected model: 49.0 -> 66.4Improved
select with IdEst rather than RankMeI-JEPA ViT-B · ImageNet, all hyperparameterstop-1 accuracy of selected model: 61.9 -> 66.4Improved
select with IdEst rather than RankMeVICReg ResNet-50 · ImageNet, all hyperparameterstop-1 accuracy of selected model: 69.1 -> 65.5Hurt
select with IdEst rather than LiDARDINO ViT-S · ImageNet, all hyperparameterstop-1 accuracy of selected model: 65.5 -> 65.5No clear effect
select with IdEst rather than alpha-ReQDINO ResNet-50 · ImageNet, all hyperparameterstop-1 accuracy of selected model: 63.0 -> 67.6Improved
use dimMST (IdEst) instead of TwoNN and MLE for ID estimationintrinsic dimension estimator · 1-d helix, 200 pointsestimated IDImproved
lower IdEst (lower intrinsic dimension)SSL representations · ImageNet linear probe, 33 modelstop-1 accuracy: rho = -0.74, tau = -0.55Improved
evaluate IdEst on a single reference dataset (ImageNet) for transferSSL representations · iNat-18, iNat-21, CIFAR-10, CIFAR-100top-1 accuracy: rho -0.81 to -0.83Improved
monitor IdEst during pretrainingDINO, VICReg, I-JEPA · ImageNet offline probingIdEst vs accuracyImproved
use IdEst in first 10 epochs of ViT trainingDINO ViT-S · online probingIdEst informativenessHurt
replace IdEst with RankMe as proxyfoundation models · ImageNetSpearman rho with accuracy: -0.74 -> 0.49 (magnitude)Hurt
replace linear probe with IdEst for evaluationDINOv2 ViT-G · ImageNet-1kwall-clock minutes: 322.7 -> 12.8Improved

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