Research map / Graph learning / Graph algorithms and clustering
CauScale: Neural Causal Discovery at Scale
Bo Peng, Sirui Chen, Jiaguo Tian, Yu Qiao, Chaochao Lu
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 |
|---|---|---|---|
| add the reduction unit that average-pools data embeddings along the observation axis (ablation) | CauScale · synthetic, n up to 100 | mAP, inference time, GPU memory | Improved |
| use average pooling instead of max pooling in the reduction unit (ablation) | CauScale reduction unit · synthetic, n=50 | mAP: 71.2 -> 85.4 | Improved |
| use average pooling instead of strided pooling in the reduction unit (ablation) | CauScale reduction unit · synthetic, n=50 | mAP: 78.5 -> 85.4 | Improved |
| add the graph stream alongside the data stream (ablation) | CauScale · synthetic benchmark | mAP | Improved |
| add the inverse-covariance graph prior as graph input (ablation) | CauScale · synthetic benchmark | mAP | Improved |
| use tied attention weights instead of vanilla attention (ablation) | CauScale · synthetic, n=100, |E|=400 | mAP and inference time: mAP 0.67 -> 0.81; inference time 0.108 -> 0.038 s | Improved |
| use the pairwise decomposed prediction head instead of the vanilla AVICI head (ablation) | CauScale · synthetic, n up to 100 | cyclicity: 0.0-0.25% -> 0.0% | Improved |
| use the pairwise decomposed prediction head instead of the vanilla AVICI head (ablation) | CauScale · synthetic, n=100, |E|=400 | mAP: 0.81 vs 0.81 | No clear effect |
| apply cycle-breaking post-processing to predicted graphs (ablation) | CauScale at n=1000 · synthetic Polynomial, n=1000 | SHD: 3985.0 -> 3891.8 | Improved |
| estimate graph prior from fewer samples (ablation) | CauScale · n=1000, |E|=2000 | mAP: 64.9 (500 samples) -> 32.6 (10 samples) | Hurt |
| hide high-out-degree nodes as latent confounders (ablation) | CauScale · n=100, linear | mAP | Mixed |
| test on OOD noise and mechanism functions | CauScale · OOD generalization | mAP | Mixed |
Closest papers by meaning
- Since Faithfulness Fails: The Performance Limits of Neural Causal Discovery (ICML 2025 · similarity 0.88)
- CausalTab: Pretraining Across Causal Environments for Tabular Causal Discovery (NeurIPS 2026 · similarity 0.88)
- Integrating Background Knowledge for Scalable Causal Discovery (NeurIPS 2026 · similarity 0.87)
- CUTS+: High-Dimensional Causal Discovery from Irregular Time-Series (AAAI 2024 · similarity 0.87)
- CausalTime: Realistically Generated Time-series for Benchmarking of Causal Discovery (ICLR 2024 · similarity 0.87)
- Causal Discovery with Fewer Conditional Independence Tests (ICML 2024 · similarity 0.87)
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.
