AI Paper Atlas

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

Optimal Positive Generation via Latent Transformation for Contrastive Learning

Yinqi Li, Hong Chang, Bingpeng Ma, Shiguang Shan, Xilin CHEN

NeurIPS 2022 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Inductive bias and minimal representations · 2 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
replace Gaussian latent transformation with learned optimal positivessynthetic-image contrastive learning · ImageNet-1K linear evaltop-1 (%): 48.73 -> 53.25Improved
replace steering with learned instance-specific positivessynthetic-image contrastive learning · ImageNet-1K linear evaltop-1 (%): 51.19 -> 53.25Improved
train on BigBiGAN synthetic images with COP-Gen instead of real images with SimCLRcontrastive learning · ImageNet-1K linear evaltop-1 (%): 49.44 -> 53.25Improved
remove data augmentation Tx (use only Tz) (ablation)COP-Gen · ImageNet-100 linear evaltop-1 (%): 63.96 -> 37.40Hurt
remove color augmentation (ablation)COP-Gen · ImageNet-100 linear evaltop-1 (%): 63.96 -> 55.26Hurt
remove crop augmentation (ablation)COP-Gen · ImageNet-100 linear evaltop-1 (%): 63.96 -> 43.84Hurt
remove all data space augmentation (ablation)COP-Gen · ImageNet-100 linear evaltop-1 (%): 63.96 -> 36.16Hurt
increase navigator training samples from 200K to 400K (ablation)COP-Gen · ImageNet-100 linear evaltop-1 (%): 63.96 -> 64.12No clear effect
increase navigator training samples from 200K to 600K (ablation)COP-Gen · ImageNet-100 linear evaltop-1 (%): 63.96 -> 63.56No clear effect
use linear latent trajectory instead of nonlinear MLP Tz (ablation)COP-Gen · ImageNet-100 linear evaltop-1 (%): 63.96 -> 61.18Hurt
learned optimal positives vs steering without data augmentationCOP-Gen · ImageNet-100 linear evaltop-1 (%): 28.68 -> 37.40Improved
COP-Gen vs real-image NNCLR on VOC detectionCOP-Gen · PASCAL VOCAP: 51.70 -> 53.09Improved

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