Unsuitability of NOTEARS for Causal Graph Discovery

2021/04/12 by Marcus Kaiser, Kaiser, Marcus, Maksim Sipos +1 · 2 citations
Computer Science · Decision Sciences · #Bayesian Modeling and Causal Inference #Data Quality and Management #FOS: Computer and information sciences #FOS: Mathematics #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Rough Sets and Fuzzy Logic #Statistics Theory (math.ST)

paper · pdf · doi:10.48550/arxiv.2104.05441

openalex publication_date 2021/04/12 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Causal Discovery methods aim to identify a DAG structure that represents causal relationships from observational data. In this article, we stress that it is important to test such methods for robustness in practical settings. As our main example, we analyze the NOTEARS method, for which we demonstrate a lack of scale-invariance. We show that NOTEARS is a method that aims to identify a parsimonious DAG from the data that explains the residual variance. We conclude that NOTEARS is not suitable for identifying truly causal relationships from the data.

Citations

• The Do-Calculus Revisited
• Algorithm 778: L-BFGS-B
• High-dimensional learning of linear causal networks via inverse covariance estimation
• DAGs with NO TEARS: Continuous Optimization for Structure Learning
• DAG-GNN: DAG Structure Learning with Graph Neural Networks
• Review of Causal Discovery Methods Based on Graphical Models
• Causality for Machine Learning
• On the Role of Sparsity and DAG Constraints for Learning Linear DAGs

Cited by

• Realizing LLMs' Causal Potential Requires Science-Grounded, Novel Benchmarks
• Causal inference of post-transcriptional regulation timelines from long-read sequencing in Arabidopsis thaliana

Related

• Evaluation of Induced Expert Knowledge in Causal Structure Learning by NOTEARS
• Causal DAG Summarization (Full Version)
• Causal Discovery in Mixtures of Populations
• Causal Discovery for Linear DAGs with Dependent Latent Variables via Higher-order Cumulants
• Amortized Causal Discovery: Learning to Infer Causal Graphs from Time-Series Data