Research map / Probabilistic and scientific modeling / Molecules and proteins
Coordinated Disentanglement with Iterative Mode Discovery Under Hidden Correlations
Rong Hu, Ling Chen
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 mode-based CMI with iterative mode discovery and meta-coordination (CoDID) vs best DRL baseline | CoDID · seven datasets | accuracy: average +7.8% accuracy, +7.9% macro F1 | Improved |
| add mode prediction loss and discrimination loss (ablation) | CoDID-km · CMNIST test 3 | accuracy: 74.2 -> 79.0 (mode prediction); 73.0 -> 79.0 (discrimination) | Improved |
| share discriminator parameters among modes under the same attribute value (ablation) | CoDID-km · CMNIST test 3 | accuracy: 77.1 -> 79.0 (vs individual); 76.3 -> 79.0 (vs one shared) | Improved |
| naive iterative clustering (k-means or DPGMM) during DRL instead of fixed clustering (ablation) | CoDID-km · CMNIST test 3 | accuracy: 79.0 -> 76.5 (itkm); 79.0 -> 77.6 (itdpm) | Hurt |
| add weighted CMI minimization (meta-learned weights) (ablation) | CoDID-itdpm · CMNIST test 3 | accuracy: 77.6 -> 79.1 | Improved |
| add subcluster net with weight alignment loss for cluster splits (ablation) | CoDID-w/o-SC · CMNIST test 3 | accuracy: 79.1 -> 85.2 | Improved |
| add partial network reconfiguration when number of clusters changes (ablation) | CoDID-UA · CMNIST test 3 | accuracy: 82.0 -> 85.2 | Improved |
| use subcluster net instead of DPGMM subclustering (ablation) | CoDID · CMNIST subclustering | subcluster accuracy: 0.678 -> 0.892 | Improved |
| mode-based CMI with true mode labels vs attribute-based CMI | A-CMI · Toy data test 3 | a1 accuracy: 0.669 -> 0.926 | Improved |
| invariance/independence baselines vs plain supervised learning | BASE · CMNIST test 3 | accuracy: 79.2 -> 58.5 (A-CMI) | Hurt |
| CoDID-itkm and CoDID-km on image data despite ground-truth mode count (ablation) | CoDID · image datasets | accuracy: CMNIST 79.0 -> 85.2 (vs CoDID-km) | Improved |
Closest papers by meaning
- On Disentangled Representations Learned from Correlated Data (ICML 2021 · similarity 0.87)
- Learning Disentangled Representation with Pairwise Independence (AAAI 2019 · similarity 0.87)
- Graph-based Unsupervised Disentangled Representation Learning via Multimodal Large Language Models (NeurIPS 2024 · similarity 0.86)
- Disentangled Representation Learning via Modular Compositional Bias (NeurIPS 2025 · similarity 0.85)
- Sequential Disentanglement by Extracting Static Information From A Single Sequence Element (ICML 2024 · similarity 0.85)
- $\alpha$TC-VAE: On the relationship between Disentanglement and Diversity (ICLR 2024 · 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.
