Research map / Label-efficient and robust learning / Recognition and long-tail
A Contrastive Objective for Learning Disentangled Representations
Jonathan Kahana, Yedid Hoshen
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 |
|---|---|---|---|
| add domain-wise negatives (negatives only from same domain) (ablation) | DCoDR · SmallNorb | invariance (lower better): 0.863 -> 0.143 | Improved |
| add domain-wise negatives (ablation) | DCoDR-norec · SmallNorb | invariance (lower better): 0.879 -> 0.071 | Improved |
| add MoCo-V2 pre-trained initialization (ablation) | DCoDR · SmallNorb | informativeness: 0.701 -> 0.785 | Improved |
| add MoCo-V2 pre-trained initialization (ablation) | DCoDR · SmallNorb | invariance (lower better): 0.253 -> 0.143 | Improved |
| add positive augmentations (ablation) | DCoDR · SmallNorb | informativeness: 0.555 -> 0.785 | Improved |
| add positive augmentations (ablation) | DCoDR · SmallNorb | invariance (lower better): 0.057 -> 0.143 | Hurt |
| add positive augmentations (ablation) | DCoDR-norec · SmallNorb | informativeness: 0.166 -> 0.730 | Improved |
| use selected augmentations instead of SimSiam augmentations (ablation) | DCoDR · SmallNorb | informativeness: 0.643 -> 0.785 | Improved |
| use Gaussian blur instead of horizontal flip augmentation | DCoDR-norec · Cars3D | informativeness: 0.725 -> 0.960 | Improved |
| add reconstruction constraint | DCoDR · SmallNorb | informativeness: 0.730 -> 0.785 | Improved |
| add reconstruction constraint | DCoDR · Edges2Shoes | retrieval accuracy: 0.53 -> 0.90 | Improved |
| add reconstruction constraint | DCoDR · CelebA | landmark L1 error (lower better): 1.127 -> 0.858 | Improved |
Closest papers by meaning
- What Should Not Be Contrastive in Contrastive Learning (ICLR 2021 · similarity 0.89)
- Detach and Adapt: Learning Cross-Domain Disentangled Deep Representation (CVPR 2018 · similarity 0.89)
- Disentangled Representation Learning via Modular Compositional Bias (NeurIPS 2025 · similarity 0.88)
- Transferring disentangled representations: bridging the gap between synthetic and real images (NeurIPS 2024 · similarity 0.87)
- $\mathbb{X}$-Sample Contrastive Loss: Improving Contrastive Learning with Sample Similarity Graphs (ICLR 2025 · similarity 0.87)
- Synergy Between Sufficient Changes and Sparse Mixing Procedure for Disentangled Representation Learning (ICLR 2025 · similarity 0.86)
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.
