Joint graph learning and prediction

Abstract

Estimating the structure of directed acyclic graphs (DAGs, also known as Bayesian networks) is a challenging problem since the search space of DAGs is combinatorial and scales super exponentially with the number of nodes. Existing approaches rely on various local heuristics for enforcing the acyclicity constraint. Building upon a previous state-of-the-art algorithm for DAG learning from data sets, we incorporate a new constraint in order to see whether the existing algorithms can be improved upon in the context of survival analysis. The existing approach converts the combinatorial problem into a continuous optimization problem, based on equality constraints. We investigate the effect of finding the correlation between different features and integrating them with the learning process, as a new constraint.