AI Paper Atlas

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

Provable Guarantees for Self-Supervised Deep Learning with Spectral Contrastive Loss

Jeff Z. HaoChen, Colin Wei, Adrien Gaidon, Tengyu Ma

NeurIPS 2021 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Contrastive versus non-contrastive theory · 41 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
spectral contrastive loss instead of SimCLR/SimSiamResNet encoder · CIFAR-10/100, Tiny-ImageNet linear evaltop-1 accuracy: 90.60 -> 92.07 (CIFAR-10, 800 ep vs SimCLR)Mixed
spectral contrastive loss vs SimSiam on ImageNetResNet 100-epoch · ImageNet linear evaltop-1 accuracy (%): 68.1 -> 66.97No clear effect
increase projection radius hyperparameter mu from 1 to 10 (ablation)Spectral contrastive loss · Tiny-ImageNet 200 epochstop-1 accuracy: 28.76 -> 41.30Improved
increase mu from 3 to 10 with longer training (ablation)Spectral contrastive loss · CIFAR-100 800 epochstop-1 accuracy: 66.18 -> 65.16Hurt
increase mu from 1 to 10 at short training (ablation)Spectral contrastive loss · CIFAR-10 200 epochstop-1 accuracy: 86.47 -> 88.66Improved

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