NeurIPS 2020

On the Role of Sparsity and DAG Constraints for Learning Linear DAGs

Meta Review

This paper studies structure learning for DAGs and examines the role of sparsity and acyclic constraints in this problem. The paper shows that a regularized MLE objective recovers a DAG that is quasi-equivalent to the ground truth. The algorithm is also significantly more scalable than prior methods. This is a good step toward making structure learning in DAGs highly practical, and as such we recommend acceptance.