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
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 changed | Where | Measured | Effect |
|---|---|---|---|
| Use image augmentations instead of sampling positive pairs from a specified conditional distribution (ablation) | Contrastive learning on 3DIdent · 3DIdent test set | R²: 96.73 ± 0.10 -> 45.51 ± 1.43 | Hurt |
| Use image augmentations instead of sampling positive pairs from a specified conditional distribution (ablation) | Contrastive learning on 3DIdent · 3DIdent test set | MCC: 98.31 ± 0.04 -> 46.34 ± 1.59 | Hurt |
| Map the Box latent domain to a Sphere model space (ablation) | InfoNCE on 3DIdent · 3DIdent test set | R²: 96.73 ± 0.10 -> 65.74 ± 0.01 | Hurt |
| Map the Box latent domain to a Sphere model space (ablation) | InfoNCE on 3DIdent · 3DIdent test set | MCC: 98.31 ± 0.04 -> 42.44 ± 3.27 | Hurt |
| Use a Normal rather than Laplace positive-pair conditional (ablation) | Contrastive learning on KITTI Masks · KITTI Masks, Δt̄=0.05s | MCC: 77.1 ± 1.0 -> 58.3 ± 5.4 | Hurt |
| Use a Normal rather than Laplace positive-pair conditional (ablation) | Contrastive learning on KITTI Masks · KITTI Masks, Δt̄=0.15s | MCC: 79.4 ± 1.9 -> 60.2 ± 8.7 | Hurt |
| Use Unbounded rather than Box model space (ablation) | InfoNCE on 3DIdent · 3DIdent test set | MCC: 98.31 ± 0.04 -> 54.94 ± 0.02 | Hurt |
| Use Laplace rather than Normal positive-pair conditional (ablation) | InfoNCE on 3DIdent · 3DIdent test set | MCC: 98.31 ± 0.04 -> 98.38 ± 0.03 | No clear effect |
Closest papers by meaning
- The Power of Contrast for Feature Learning: A Theoretical Analysis (JMLR 2023 · similarity 0.91)
- A theoretical study of inductive biases in contrastive learning (ICLR 2023 · similarity 0.90)
- Improving Transformation Invariance in Contrastive Representation Learning (ICLR 2021 · similarity 0.89)
- Multi-modal contrastive learning adapts to intrinsic dimensions of shared latent variables (NeurIPS 2025 · similarity 0.88)
- Do Generated Data Always Help Contrastive Learning? (ICLR 2024 · similarity 0.87)
- Contrastive Identification and Generation in the Limit (NeurIPS 2026 · similarity 0.87)
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.
