L2C: A Unified Framework for Causal Discovery with Latent Variables
A recent publication on arXiv introduces L2C (Local to Cluster Causal Abstraction), a comprehensive framework designed for causal discovery that accommodates latent variables. Traditional local approaches emphasize immediate neighbors but often miss broader macro-level perspectives, whereas cluster-based methods either presuppose known clusters or demand causal sufficiency. L2C autonomously identifies cluster partitions derived from local causal patterns, effectively linking local structure learning with cluster-level causal discovery. Utilizing a cluster reduction theorem, it can condense any cluster to a maximum of three nodes while preserving causal information. This research is available under arXiv:2604.22416v1.
Key facts
- L2C stands for Local to Cluster Causal Abstraction
- The framework bridges local structure learning and cluster-level causal discovery
- It automatically discovers cluster partitions from local causal patterns
- It uses a cluster reduction theorem to reduce any cluster to at most three nodes
- The paper is published on arXiv with ID 2604.22416v1
- Latent variables pose a fundamental challenge to causal discovery
- Conventional local methods fail to provide macro-level insights
- Cluster-level methods often require known clusters or causal sufficiency
Entities
Institutions
- arXiv