Research map / Probabilistic and scientific modeling / Causal discovery
Likelihood-Free Overcomplete ICA and Applications In Causal Discovery
Chenwei DING, Mingming Gong, Kun Zhang, Dacheng Tao
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 adversarial learning for over-complete ICA | ICA mixing-matrix recovery · Synthetic over-complete ICA | MSE: AICA: 4.61e-2 vs MFICA_biGauss: 1.82e-1; 5.95e-2 vs 6.56e-2; 6.96e-2 vs RICA: 9.03e-2; 5.92e-2 vs RICA: 7.54e-2 | Improved |
| jointly learn the dynamics and noise relation with a conditional model instead of estimating A^k first | Subsampled time-series causal discovery · Synthetic data, n=5 | MSE: 5.42e-3 -> 5.00e-4 (T=1000, k=2); 4.38e-3 -> 1.89e-4 (T=3000, k=2) | Improved |
| use the conditional model rather than the baseline model | Subsampled time-series causal discovery · Synthetic data, n=2 | MSE: At T=100, k=2: 1.03e-2 -> 5.25e-2; at T=300, k=2: 1.67e-3 -> 1.12e-3 | Mixed |
| use the divided model for aggregation instead of CDTA_finite | Aggregated time-series causal discovery · Synthetic data, n=2, k=2 or 3 | MSE: 2.10e-4 -> 1.78e-3 (T=150, k=2); 1.19e-4 -> 1.07e-3 (T=300, k=2); 8.17e-4 -> 8.00e-3 (T=150, k=3); 7.36e-4 -> 5.87e-3 (T | Hurt |
| vary divided-model piece length (ablation) | Divided model · Aggregated synthetic time series | MSE: At n=2, T=3000, k=2: l=8: 2.20e-04; l=10: 2.94e-04 | Mixed |
| use longer pieces when the total observed series is longer (ablation) | Divided model · Aggregated synthetic time series | MSE | Improved |
Closest papers by meaning
- Causal Discovery with Latent Confounders Based on Higher-Order Cumulants (ICML 2023 · similarity 0.92)
- Independence Testing-Based Approach to Causal Discovery under Measurement Error and Linear Non-Gaussian Models (NeurIPS 2022 · similarity 0.90)
- Robustifying Independent Component Analysis by Adjusting for Group-Wise Stationary Noise (JMLR 2019 · similarity 0.88)
- On the identifiability of causal graphs with multiple environments (ICLR 2026 · similarity 0.88)
- Identification of Causal Structure with Latent Variables Based on Higher Order Cumulants (AAAI 2024 · similarity 0.87)
- When Additive Noise Meets Unobserved Mediators: Bivariate Denoising Diffusion for Causal Discovery (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.
