Research map / Label-efficient and robust learning / Recognition and long-tail
Learning Disentangled Representations via Mutual Information Estimation
Eduardo Hugo Sanchez, Mathieu Serrurier, Mathias Ortner
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 |
|---|---|---|---|
| switch the shared representations (SSR) (ablation) | shared representation learning · Colored-MNIST (S_X) | distance to ideal: 1.0209 -> 0.0563 | Improved |
| add the L1 distance between S_X and S_Y (ablation) | shared representation learning · Colored-MNIST (S_X) | distance to ideal: 0.0780 -> 0.0563 | Improved |
| add the global mutual information term (ablation) | shared representation learning · Colored-MNIST (S_X) | distance to ideal: 0.0795 -> 0.0563 | Improved |
| add the local mutual information term (ablation) | shared representation learning · Colored-MNIST (S_X) | distance to ideal: 0.1958 -> 0.0563 | Improved |
| switch the shared representations (SSR) (ablation) | shared representation learning · IAM (S_X) | distance to ideal: 0.6094 -> 0.4730 | Improved |
| add the L1 distance between S_X and S_Y (ablation) | shared representation learning · IAM (S_X) | distance to ideal: 0.5412 -> 0.4730 | Improved |
| add the global mutual information term (ablation) | shared representation learning · IAM (S_X) | distance to ideal: 0.4886 -> 0.4730 | Improved |
| add the local mutual information term (ablation) | shared representation learning · IAM (S_X) | distance to ideal: 0.6235 -> 0.4730 | Improved |
| increase adversarial weight lambda_adv from 0.0 to 0.025 (ablation) | exclusive representation · colored-MNIST / 3D Shapes | exclusive-representation leakage of shared attributes: 0.0 -> 0.025 | Improved |
| increase adversarial weight lambda_adv beyond 0.025 (to 0.05) (ablation) | exclusive representation · colored-MNIST / 3D Shapes | exclusive attribute classification accuracy: 0.025 -> 0.05 | Hurt |
| use mutual-information model instead of VAE-GAN of Gonzalez-Garcia et al. | disentangled representation · Colored-MNIST (background color) | distance to ideal: 0.9581 -> 0.0563 | Improved |
| use mutual-information model instead of VAE-GAN of Gonzalez-Garcia et al. | disentangled representation · 3D Shapes | distance to ideal: 2.6643 -> 0.0020 | Improved |
Closest papers by meaning
- Learning Disentangled Representation with Pairwise Independence (AAAI 2019 · similarity 0.88)
- Leveraging sparse and shared feature activations for disentangled representation learning (NeurIPS 2023 · similarity 0.87)
- On the Transfer of Disentangled Representations in Realistic Settings (ICLR 2021 · similarity 0.87)
- Transferring disentangled representations: bridging the gap between synthetic and real images (NeurIPS 2024 · similarity 0.87)
- Disentangled Representation Learning via Flow Matching (NeurIPS 2026 · similarity 0.86)
- Disentangled Representation Learning via Modular Compositional Bias (NeurIPS 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.
