Abstract:Estimating the structure of directed acyclic graphs (DAGs) of features (variables) plays a vital role in revealing the latent data generation process and providing causal insights in various applications. Although there have been many studies on structure learning with various types of data, the structure learning on the dynamic graph has not been explored yet, and thus we study the learning problem of node feature generation mechanism on such ubiquitous dynamic graph data. In a dynamic graph, we propose to simultaneously estimate contemporaneous relationships and time-lagged interaction relationships between the node features. These two kinds of relationships form a DAG, which could effectively characterize the feature generation process in a concise way. To learn such a DAG, we cast the learning problem as a continuous score-based optimization problem, which consists of a differentiable score function to measure the validity of the learned DAGs and a smooth acyclicity constraint to ensure the acyclicity of the learned DAGs. These two components are translated into an unconstraint augmented Lagrangian objective which could be minimized by mature continuous optimization techniques. The resulting algorithm, named GraphNOTEARS, outperforms baselines on simulated data across a wide range of settings that may encounter in real-world applications. We also apply the proposed approach on two dynamic graphs constructed from the real-world Yelp dataset, demonstrating our method could learn the connections between node features, which conforms with the domain knowledge.

On the sparse DAG structure learning based on adaptive Lasso

Learning directed acyclic graphs based on sparsest permutations

Directed Acyclic Graphs With Tears

Directed Acyclic Graph Structure Learning from Dynamic Graphs

Adaptive Skeleton Construction for Accurate DAG Learning

Non-negative Weighted DAG Structure Learning

DAG-GAN: Causal Structure Learning with Generative Adversarial Nets

Directed Adaptive Graphical Lasso for causality inference.

$ψ$DAG: Projected Stochastic Approximation Iteration for DAG Structure Learning

Learning DAGs without imposing acyclicity

CD- NOTEARS: Concept Driven Causal Structure Learning Using NOTEARS

Sparse Optimization Driven Deep Unfolding Network for DOA Estimation

Structure Learning with Adaptive Random Neighborhood Informed MCMC

Structure Learning of Sparse Directed Acyclic Graphs Incorporating the Scale-Free Property

Efficient Learning of Nonparametric Directed Acyclic Graph with Statistical Guarantee

Bayesian Network Structure Learning with a New Ensemble Weights and Edge Constraints Setting Mechanism

DAGOR: Learning DAGs via Topological Sorts and QR Factorization

Effective Structure Learning for EDA Via L1-regularizedbayesian Networks.

On Low Rank Directed Acyclic Graphs and Causal Structure Learning

NOTMAD: Estimating Bayesian Networks with Sample-Specific Structures and Parameters

Sparse‐Dyn: Sparse dynamic graph multirepresentation learning via event‐based sparse temporal attention network