Learning Large DAGs by Combining Continuous Optimization and Feedback Arc Set Heuristics

Bayesian networks represent relations between variables using a directed acyclic graph (DAG). Learning the DAG is an NP-hard problem and exact learning algorithms are feasible only for small sets of variables. We propose two scalable heuristics DAGs in linear structural equation case. Our methods learn by alternating unconstrained gradient descent-based step to optimize objective function solving maximum subgraph enforce acyclicity. Thanks this decoupling, our scale up beyond thousands