Research map / Probabilistic and scientific modeling / Causal discovery
A Critical Look at the Consistency of Causal Estimation with Deep Latent Variable Models
Severi Rissanen, Pekka Marttinen
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 full neural CEVAE instead of linear CEVAE with 1D latent (ablation) | CEVAE · Synthetic data | AID | No clear effect |
| add a redundant latent dimension to linear CEVAE (2D latent) (ablation) | linear CEVAE · Synthetic data | causal effect estimate | Hurt |
| prevent posterior collapse with KL divergence annealing (ablation) | 10D linear CEVAE · Synthetic data | causal effect estimate | Hurt |
| use 2D instead of 1D latent variable (ablation) | full CEVAE · Synthetic data | AID | Improved |
| add repeated noisy proxies to the data | full CEVAE · Synthetic data with noisy proxies | c_yt estimate | Hurt |
| scale proxy reconstruction loss to one-half (ablation) | full CEVAE · Synthetic data with repeated proxies, n=20000 | c_yt estimate | Improved |
| scale proxy reconstruction loss below one-half toward zero (ablation) | full CEVAE · Synthetic data with repeated proxies, n=20000 | c_yt estimate | Hurt |
| scale proxy loss on IHDP-derived data (ablation) | full CEVAE · Modified IHDP | E[y|do(t=1)] estimate | Mixed |
| scale image reconstruction loss to 0.05 (ablation) | CEVAE · Proxy MNIST | causal effect estimate | Improved |
| scale image reconstruction loss to 0.1 or 0.01 (ablation) | CEVAE · Proxy MNIST | causal effect estimate | No clear effect |
| no scaling of image reconstruction loss (ablation) | CEVAE · Proxy MNIST | causal effect estimate | Hurt |
Closest papers by meaning
- Causal Inference with Conditional Front-Door Adjustment and Identifiable Variational Autoencoder (ICLR 2024 · similarity 0.91)
- Identifiable Latent Polynomial Causal Models through the Lens of Change (ICLR 2024 · similarity 0.89)
- Causal Effect Identification in LiNGAM Models with Latent Confounders (ICML 2024 · similarity 0.88)
- Covariate-Adjusted Deep Causal Learning for Heterogeneous Panel Data Models (NeurIPS 2026 · similarity 0.88)
- Differentiable Causal Discovery for Latent Hierarchical Causal Models (ICLR 2025 · similarity 0.88)
- Identifiability Guarantees for Causal Disentanglement from Purely Observational Data (NeurIPS 2024 · similarity 0.88)
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.
