AI Paper Atlas

Research map / Probabilistic and scientific modeling / Variational and generative sampling

Representation Uncertainty in Self-Supervised Learning as Variational Inference

Hiroki Nakamura, Masashi Okada, Tadahiro Taniguchi

ICCV 2023 · Probabilistic and scientific modeling › Variational and generative sampling · approach: Regularizing VAE latent information · 23 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
learn per-image uncertainty kappa with Power Spherical latent (VI-SimSiam) instead of SimSiamSimSiam · ImageNet100 linear eval, 50 epochsTop-1 accuracy: 41.07 -> 63.89Improved
VI-SimSiam instead of SimSiam at long trainingSimSiam · ImageNet100 linear eval, 500 epochsTop-1 accuracy: 81.61 -> 77.49Hurt
VI-SimSiam instead of SimSiam on CIFAR-10SimSiam · CIFAR-10 linear eval, 100 epochsTop-1 accuracy: 77.43 -> 89.92Improved
use predicted kappa as confidence score for correctnessVI-SimSiam (500 epochs) · ImageNet100AUROC: 0.5 (chance) -> 0.72 linear, 0.76 KNNImproved
use negative entropy of DINO representation as confidence scoreDINO (200 epochs) · ImageNet100AUROC: 0.5 (chance) -> 0.67Improved
random crop augmentation compared to other augmentations (ablation)VI-SimSiam kappa · ImageNet100 valmean kappa: 852.63 (normal) -> 812.26 (random crop)Hurt
select the highest-kappa view among 30 augmentations at test time (ablation)VI-SimSiam · ImageNet100 linear eval, 200 epochsTop-1 accuracy: 76.74 -> 77.36No clear effect
VI-SimSiam instead of SimSiam for transfer with 100-epoch pretrainingSimSiam · Flowers transfer linear evalTop-1 accuracy: 14.8 -> 64.5Improved
VI-SimSiam instead of SimSiam for transfer with 500-epoch pretrainingSimSiam · Food transfer linear evalTop-1 accuracy: 59.2 -> 46.1Hurt
VI-SimSiam instead of SimSiam with 1% labelsSimSiam · ImageNet100 semi-supervised, 1% labels, 100 epochsTop-1 accuracy: 33.8 -> 53.5Improved

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