arXiv 2311.10263

Stable Differentiable Causal Discovery

By Achille Nazaret, Justin Hong, et al.

Published 2023-11-17

Mindmap

Browse the paper's core ideas, clusters, and relationships in a structured outline.

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…

View the original paper on arXiv