Unsuitability of NOTEARS for Causal Graph Discovery

Many popular causal discovery algorithms have significant limitations in applied settings. causaLens’ Marcus Kaiser and Maksim Sipos show how one such algorithm, NOTEARS, is not suitable for identifying truly causal relationships from data. The paper is published in Neural Processing Letters.

Our own Marcus Kaiser and Maksim Sipos demonstrate that the NOTEARS algorithm lacks scale-invariance by showing how a simple rescaling of the variables can lead to a large alteration of the derived DAG. Access the full article, published in Neural Processing Letters, here or download a copy by clicking the link below.