Research map / Probabilistic and scientific modeling / Molecules and proteins
Symmetry-induced Disentanglement on Graphs
Giangiacomo Mercatali, Andre Freitas, Vikas Garg
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 |
|---|---|---|---|
| conditional SID (SIDC) instead of unconditional (SIDU) | SID · Watts-Strogatz | MIG: 28.5±2.8 -> 49.1±3.1 | Improved |
| conditional SID (SIDC) instead of unconditional (SIDU) | SID · Erdos-Renyi | MIG: 53.2±0.8 -> 52.4±3.1 | No clear effect |
| SID with Lie-algebra subgroup structure instead of GCN VAE | graph VAE · Watts-Strogatz | FL: 84.1±0.5 -> 92.3±0.9 | Improved |
| SIDC instead of best disentangled baseline DIS | graph VAE · Erdos-Renyi | FVM: 68.1±1.8 -> 83.3±0.7 | Improved |
| increase group size (ablation) | SIDU · Watts-Strogatz | all disentanglement metrics | Improved |
| add Hessian penalty (ablation) | SIDU · Watts-Strogatz | all disentanglement metrics | Improved |
| add commutative penalty (ablation) | SIDU · Watts-Strogatz | all disentanglement metrics | No clear effect |
| SIDC instead of MIA autoencoder | graph autoencoder · FRANKENSTEIN few-shot | accuracy: 65±1 -> 69±2 | Improved |
| SIDC instead of MIA autoencoder | graph autoencoder · COLORS-3 few-shot | accuracy: 33±2 -> 38±3 | Improved |
| SIDC instead of MIA autoencoder | graph autoencoder · Mutagenicity few-shot | accuracy: 78±2 -> 84±2 | Improved |
| SIDC instead of MIA autoencoder | graph autoencoder · NCI1 few-shot | accuracy: 61±1 -> 66±2 | Improved |
| SIDC instead of MIA autoencoder | graph autoencoder · FRANKENSTEIN reconstruction | MSE: 1.6±0.5 -> 2.8±0.5 | Hurt |
Closest papers by meaning
- Orthogonality-Enforced Latent Space in Autoencoders: An Approach to Learning Disentangled Representations (ICML 2023 · similarity 0.90)
- Quantifying and Learning Linear Symmetry-Based Disentanglement (ICML 2022 · similarity 0.89)
- Disentanglement Analysis with Partial Information Decomposition (ICLR 2022 · similarity 0.89)
- Graph-based Unsupervised Disentangled Representation Learning via Multimodal Large Language Models (NeurIPS 2024 · similarity 0.86)
- Enriching Disentanglement: From Logical Definitions to Quantitative Metrics (NeurIPS 2024 · similarity 0.86)
- Improving Equivariant Networks with Probabilistic Symmetry Breaking (ICLR 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.
