27 October 2023
Hyperparameters strongly affect structure learning, incorrect values cause mistakes
Algorithms differ in performance even with optimal hyperparameters
No algorithm performs best under all conditions
Algorithms vary in robustness to misspecified hyperparameters
Fixed hyperparameters can work well, but risks remain
Hyperparameters affect causal graph learning
This paper investigates how hyperparameters influence the performance of algorithms that learn causal graph structures from data. It finds hyperparameters play a key role, with incorrect values leading to mistakes in recovered graphs. The paper also shows algorithms vary in robustness to misspecified hyperparameters.
No comments yet, be the first to start the conversation...
Sign up to comment on this paper