Research map / Probabilistic and scientific modeling / Causal discovery
Entropic Causal Inference: Identifiability and Finite Sample Results
Spencer Compton, Murat Kocaoglu, Kristjan Greenewald, Dmitriy Katz
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 |
|---|---|---|---|
| tighten decision threshold from 1.2 to 0.7 x log support (5-state) | entropic causality exogenous-entropy test · Tubingen pairs | accuracy (%): 60.0 -> 85.7 | Improved |
| tighten decision threshold from 1.2 to 0.7 x log support (10-state) | entropic causality exogenous-entropy test · Tubingen pairs | accuracy (%): 56.5 -> 84.6 | Improved |
| tighten decision threshold from 1.2 to 0.7 x log support (20-state) | entropic causality exogenous-entropy test · Tubingen pairs | accuracy (%): 49.4 -> 75.0 | Improved |
| quantize to 20 states instead of 10 states (threshold 0.85) | entropic causality · Tubingen pairs | accuracy (%): 70.6 -> 53.7 | Hurt |
| quantize to 5 states instead of 10 states (threshold 0.9) | entropic causality · Tubingen pairs | accuracy (%): 63.0 -> 57.1 | Hurt |
| perturb quantization intervals and take majority of 5 decisions | entropic causality · Tubingen pairs | accuracy | No clear effect |
| compare exogenous entropies instead of observed entropies H(X) vs H(Y) | entropic causality · synthetic data, regimes 2-3 | direction accuracy | Improved |
| compare exogenous entropies instead of total entropies | entropic causality · synthetic data | direction accuracy | No clear effect |
| use greedy minimum entropy coupling criterion instead of conditional entropy test | entropic causality · synthetic data, finite samples | probability of correct direction: by approximately 5% | Improved |
| increase number of samples N when n is small (n in 20,30,40) | entropic causality · synthetic data | fraction of identifiable models | No clear effect |
| increase number of samples N when n > 40 | entropic causality · synthetic data | fraction of identifiable models: to 1 | Improved |
| add light latent confounder with entropy t (H(E)=2) | entropic causality · synthetic data | direction accuracy | No clear effect |
Closest papers by meaning
- Entropic Causal Inference (AAAI 2017 · similarity 0.95)
- Entropic Causal Inference: Graph Identifiability (ICML 2022 · similarity 0.95)
- Identifiability Guarantees for Causal Disentanglement from Purely Observational Data (NeurIPS 2024 · similarity 0.90)
- Causal learning with the invariance principle (NeurIPS 2026 · similarity 0.89)
- Identifiable Latent Polynomial Causal Models through the Lens of Change (ICLR 2024 · similarity 0.89)
- Learning Structural Causal Models from Ordering: Identifiable Flow Models (AAAI 2025 · 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.
