Research map / Probabilistic and scientific modeling / Causal discovery
Learning Linear Causal Representations from Interventions under General Nonlinear Mixing
Simon Buchholz, Goutham Rajendran, Elan Rosenfeld, Bryon Aragam, Bernhard Schölkopf, Pradeep Kumar Ravikumar
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 |
|---|---|---|---|
| use nonlinear MLP encoder h instead of linear h on linear mixing (ablation) | Contrastive · linear f (d=5, d'=10) | MCC: 0.90 -> 0.05 | Hurt |
| nonlinear contrastive vs linear baseline for graph recovery on linear mixing | Contrastive · linear f (d=5, d'=10) | SHD: 7.0 -> 4.6 | Improved |
| contrastive linear vs linear baseline for graph recovery on linear mixing | Contrastive Linear · linear f (d=5, d'=10) | AUROC: 0.64 -> 0.80 | Improved |
| replace linear disentanglement baseline with nonlinear contrastive method | Contrastive · nonlinear f, ER(5,2), d'=20 | MCC: 0.32 -> 0.97 | Improved |
| replace linear baseline with contrastive method on graph recovery | Contrastive · nonlinear f, ER(10,2), d'=20 | SHD: 29.6 -> 3.6 | Improved |
| replace VAE with contrastive method for latent recovery | Contrastive · nonlinear f, ER(10,2), d'=100 | R2: 0.78 -> 0.97 | Improved |
| increase observed dimension d' from 20 to 100 (ablation) | Contrastive · nonlinear f, ER(5,2) | MCC: 0.97 -> 0.99 | Improved |
| increase observed dimension d' from 20 to 100 (ablation) | Linear baseline · nonlinear f, ER(5,2) | MCC: 0.32 -> 0.18 | Hurt |
| increase latent dimension d from 5 to 10 (ablation) | Contrastive · nonlinear f, d'=20 | SHD: 1.8 -> 3.6 | Hurt |
| replace VAE with contrastive learning on image data | Contrastive Learning · image data, 2 balls | MCC: 0.19 -> 0.87 | Improved |
| replace VAE with contrastive learning on image data | Contrastive Learning · image data, 5 balls | SHD: 18.6 -> 2.0 | Improved |
| increase number of balls from 5 to 10 (ablation) | Contrastive Learning · image data | SHD: 2.0 -> 11.0 | Hurt |
Closest papers by meaning
- Learning Linear Causal Representations from General Environments: Identifiability and Intrinsic Ambiguity (NeurIPS 2024 · similarity 0.93)
- Identifiable Latent Polynomial Causal Models through the Lens of Change (ICLR 2024 · similarity 0.91)
- Identifying General Mechanism Shifts in Linear Causal Representations (NeurIPS 2024 · similarity 0.91)
- A Sparsity Principle for Partially Observable Causal Representation Learning (ICML 2024 · similarity 0.90)
- Identifiability Guarantees for Causal Disentanglement from Purely Observational Data (NeurIPS 2024 · similarity 0.90)
- Identifying Representations for Intervention Extrapolation (ICLR 2024 · similarity 0.90)
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.
