Gradient-Based Neural DAG Learning

About

We propose a novel score-based approach to learning a directed acyclic graph (DAG) from observational data. We adapt a recently proposed continuous constrained optimization formulation to allow for nonlinear relationships between variables using neural networks. This extension allows to model complex interactions while avoiding the combinatorial nature of the problem. In addition to comparing our method to existing continuous optimization methods, we provide missing empirical comparisons to nonlinear greedy search methods. On both synthetic and real-world data sets, this new method outperforms current continuous methods on most tasks, while being competitive with existing greedy search methods on important metrics for causal inference.

Sébastien Lachapelle, Philippe Brouillard, Tristan Deleu, Simon Lacoste-Julien · 2019

Related benchmarks

Task	Dataset	Result
DAG Structure Recovery	non-linear-1 5000 samples	SHD4	48
Causal Discovery	Sachs real-world data protein signaling network	SHD13.2	41
Causal Structure Learning	Sachs	SHD15	38
Causal Discovery	CausalMan Small	F1 Score0.2	33
Causal Discovery	Syntren	F1 Score34.4	22
Causal Discovery	Sachs flow cytometry observational	SHD20	19
Causal Discovery	ER5 (n=30, h=5)	FDR0.67	18
Causal Discovery	SF5 (n=30, h=5)	FDR72	18
Causal Discovery	Synthetic SF3 n=50, h=3 (test)	FDR34	17
Causal Discovery	Synthetic ER3 n=50, h=3 (test)	FDR74	17