Research map / Probabilistic and scientific modeling / Causal discovery
CARL: Preserving Causal Structure in Representation Learning
Yulong Li, Xiwei Liu, Feilong Tang, Zhixiang Lu, Ming Hu, Yichen Li, Haochen Xue, Peixin Guo, Jionglong Su, Yutong Xie, Eran Segal, Imran Razzak
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 conditional independence loss L_CI (ablation) | CARL · Synthetic | CSI: 0.25 -> 1.00 | Improved |
| add conditional independence loss L_CI (ablation) | CARL · Synthetic | Structural accuracy: 0.40 -> 0.61 | Improved |
| add L_CI (ablation) | CARL · Synthetic | MAC: 0.83 -> 0.55 | Hurt |
| add Markov boundary preservation loss L_MBR (ablation) | CARL · Synthetic | MBRI: 0.46 -> 0.63 | Improved |
| add Markov boundary preservation loss L_MBR (ablation) | CARL · Synthetic | CSI: 0.75 -> 1.00 | Improved |
| add monotonic alignment loss L_MAC (ablation) | CARL · Synthetic | MAC: 0.32 -> 0.55 | Improved |
| add structure losses to alignment-only training (ablation) | alignment-only · Synthetic | CSI: 0.25 -> 1.00 | Improved |
| use independent predictor heads instead of shared (ablation) | CARL · Synthetic | CSI: 0.75 -> 1.0 | Improved |
| add cross-validation (ablation) | CARL · Synthetic | CSI: 0.25 -> 1.0 | Improved |
| use more negative samples than K = 32 (ablation) | CARL · Synthetic | CSI: 0.75 -> 1.0 | Improved |
| representation dimension d = 16 instead of full (ablation) | CARL · Synthetic | all metrics: within 0.05 | No clear effect |
| increase noise sigma from 0.1 to 0.5 | CARL · Synthetic | MAC: 0.89 -> 0.42 | Hurt |
Closest papers by meaning
- Towards the Causal Complete Cause of Multi-Modal Representation Learning (ICML 2025 · similarity 0.88)
- Causal Representation Learning from Multimodal Biomedical Observations (ICLR 2025 · similarity 0.87)
- Multi-View Causal Representation Learning with Partial Observability (ICLR 2024 · similarity 0.87)
- Learning Linear Causal Representations from General Environments: Identifiability and Intrinsic Ambiguity (NeurIPS 2024 · similarity 0.87)
- Causal Representation Learning Made Identifiable by Grouping of Observational Variables (ICML 2024 · similarity 0.87)
- Causal Abstraction Learning based on the Semantic Embedding Principle (ICML 2025 · 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.
