English

Learning Directed Acyclic Graphs with Penalized Neighbourhood Regression

Statistics Theory 2017-10-03 v3 Machine Learning Machine Learning Statistics Theory

Abstract

We study a family of regularized score-based estimators for learning the structure of a directed acyclic graph (DAG) for a multivariate normal distribution from high-dimensional data with pnp\gg n. Our main results establish support recovery guarantees and deviation bounds for a family of penalized least-squares estimators under concave regularization without assuming prior knowledge of a variable ordering. These results apply to a variety of practical situations that allow for arbitrary nondegenerate covariance structures as well as many popular regularizers including the MCP, SCAD, 0\ell_{0} and 1\ell_{1}. The proof relies on interpreting a DAG as a recursive linear structural equation model, which reduces the estimation problem to a series of neighbourhood regressions. We provide a novel statistical analysis of these neighbourhood problems, establishing uniform control over the superexponential family of neighbourhoods associated with a Gaussian distribution. We then apply these results to study the statistical properties of score-based DAG estimators, learning causal DAGs, and inferring conditional independence relations via graphical models. Our results yield---for the first time---finite-sample guarantees for structure learning of Gaussian DAGs in high-dimensions via score-based estimation.

Keywords

Cite

@article{arxiv.1511.08963,
  title  = {Learning Directed Acyclic Graphs with Penalized Neighbourhood Regression},
  author = {Bryon Aragam and Arash A. Amini and Qing Zhou},
  journal= {arXiv preprint arXiv:1511.08963},
  year   = {2017}
}

Comments

54 pages, 1 figure

R2 v1 2026-06-22T11:56:23.194Z