AI Paper Atlas

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

Contrastive Learning Inverts the Data Generating Process

Roland S. Zimmermann, Yash Sharma, Steffen Schneider, Matthias Bethge, Wieland Brendel

ICML 2021 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Inductive bias and minimal representations · 27 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
Use image augmentations instead of sampling positive pairs from a specified conditional distribution (ablation)Contrastive learning on 3DIdent · 3DIdent test setR²: 96.73 ± 0.10 -> 45.51 ± 1.43Hurt
Use image augmentations instead of sampling positive pairs from a specified conditional distribution (ablation)Contrastive learning on 3DIdent · 3DIdent test setMCC: 98.31 ± 0.04 -> 46.34 ± 1.59Hurt
Map the Box latent domain to a Sphere model space (ablation)InfoNCE on 3DIdent · 3DIdent test setR²: 96.73 ± 0.10 -> 65.74 ± 0.01Hurt
Map the Box latent domain to a Sphere model space (ablation)InfoNCE on 3DIdent · 3DIdent test setMCC: 98.31 ± 0.04 -> 42.44 ± 3.27Hurt
Use a Normal rather than Laplace positive-pair conditional (ablation)Contrastive learning on KITTI Masks · KITTI Masks, Δt̄=0.05sMCC: 77.1 ± 1.0 -> 58.3 ± 5.4Hurt
Use a Normal rather than Laplace positive-pair conditional (ablation)Contrastive learning on KITTI Masks · KITTI Masks, Δt̄=0.15sMCC: 79.4 ± 1.9 -> 60.2 ± 8.7Hurt
Use Unbounded rather than Box model space (ablation)InfoNCE on 3DIdent · 3DIdent test setMCC: 98.31 ± 0.04 -> 54.94 ± 0.02Hurt
Use Laplace rather than Normal positive-pair conditional (ablation)InfoNCE on 3DIdent · 3DIdent test setMCC: 98.31 ± 0.04 -> 98.38 ± 0.03No clear effect

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