arXiv 2311.10263
Stable Differentiable Causal Discovery
By Achille Nazaret, Justin Hong, et al.
Published 2023-11-17
Wiki summary
Explore the paper's summary, context, and related research on Papiers.
Inferring causal relationships as directed acyclic graphs (DAGs) is an important but challenging problem. Differentiable Causal Discovery (DCD) is a promising approach to this problem, framing the search as a continuous optimization. But existing DCD methods are numerically unstable, with poor performance beyond tens of variables. In this paper, we propose Stable Differentiable Causal Discovery (SDCD), a new method…