Research map / Label-efficient and robust learning / Recognition and long-tail
Elastic-InfoGAN: Unsupervised Disentangled Representation Learning in Class-Imbalanced Data
Utkarsh Ojha, Krishna Kumar Singh, Cho-Jui Hsieh, Yong Jae Lee
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 |
|---|---|---|---|
| Learn the categorical prior and add full contrastive loss | InfoGAN · MNIST, YTF, 3D Cars, 3D Chairs, ShapeNet | NMI: 0.777 -> 0.889; 0.666 -> 0.792; 0.499 -> 0.850; 0.253 -> 0.650; 0.638 -> 0.790 | Improved |
| Add positive-pair contrastive loss to learned-prior InfoGAN (ablation) | Gumbel-Softmax InfoGAN · MNIST, YTF, 3D Cars, 3D Chairs, ShapeNet | NMI: 0.836 -> 0.878; 0.760 -> 0.765; 0.400 -> 0.582; 0.236 -> 0.454; 0.603 -> 0.724 | Improved |
| Add negative-pair contrastive constraints (ablation) | Gumbel-Softmax + pos-L_ntxent · MNIST, YTF, 3D Cars, 3D Chairs, ShapeNet | NMI: 0.878 -> 0.889; 0.765 -> 0.792; 0.582 -> 0.850; 0.454 -> 0.650; 0.724 -> 0.790 | Improved |
| Add contrastive loss to Ground-truth InfoGAN (ablation) | Ground-truth InfoGAN · MNIST, YTF, 3D Cars, 3D Chairs | NMI: 0.783 -> 0.801; 0.694 -> 0.742; 0.451 -> 0.784; 0.174 -> 0.592 | Improved |
| Use learned prior instead of fixed ground-truth prior (ablation) | InfoGAN · MNIST, YTF, 3D Cars, 3D Chairs, ShapeNet | NMI: 0.783 vs 0.836; 0.694 vs 0.760; 0.451 vs 0.400; 0.174 vs 0.236; 0.549 vs 0.603 | Mixed |
| Learn a categorical prior and use full contrastive loss | InfoGAN · MNIST, YTF, 3D Cars, 3D Chairs, ShapeNet | ENT: 0.457 -> 0.213; 1.031 -> 0.636; 1.108 -> 0.303; 1.663 -> 0.765; 0.531 -> 0.297 | Improved |
| Use Elastic-InfoGAN features rather than Uniform InfoGAN features | 1NN classification · MNIST, YTF, 3D Cars, 3D Chairs | accuracy: 0.8665 -> 0.9655; 0.9581 -> 0.9985; 0.7700 -> 0.9852; 0.6438 -> 0.9515 | Improved |
Closest papers by meaning
- Information Bottleneck Disentanglement for Identity Swapping (CVPR 2021 · similarity 0.87)
- Unsupervised Part-Based Disentangling of Object Shape and Appearance (CVPR 2019 · similarity 0.87)
- Disentangling by Factorising (ICML 2018 · similarity 0.87)
- An Adversarial Neuro-Tensorial Approach for Learning Disentangled Representations (IJCV 2019 · similarity 0.87)
- Transferring disentangled representations: bridging the gap between synthetic and real images (NeurIPS 2024 · similarity 0.86)
- Disentangled Representation Learning via Modular Compositional Bias (NeurIPS 2025 · similarity 0.85)
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.
