[1Win4-04] Adaptation of Prior Knowledge for the Causal Discovery Algorithm SCORE
Keywords:causal search
Causal discovery techniques are often used to identify the causes of anomalies in manufacturing equipment. However, calculating accurate causal relationships can be quite challenging.To address this, models incorporating domain knowledge from the factory floor are needed.
LiNGAM allows for the input of prior knowledge but is limited to linear data, making it difficult to use in environments with mixed linear and nonlinear data.
In contrast, SCORE can treat both linear and nonlinear data, but it does not have the capability to incorporate prior knowledge.
To address this limitation, we added labels indicating "no causal relationship," "causal relationship," and "unknown" to the causal discovery and pruning sections of SCORE.
Because of the challenges in evaluating with real data, we prepared five patterns of randomly masked toy data. We then executed SCORE with prior knowledge, which resulted in increased F-values and reduced computation time.
LiNGAM allows for the input of prior knowledge but is limited to linear data, making it difficult to use in environments with mixed linear and nonlinear data.
In contrast, SCORE can treat both linear and nonlinear data, but it does not have the capability to incorporate prior knowledge.
To address this limitation, we added labels indicating "no causal relationship," "causal relationship," and "unknown" to the causal discovery and pruning sections of SCORE.
Because of the challenges in evaluating with real data, we prepared five patterns of randomly masked toy data. We then executed SCORE with prior knowledge, which resulted in increased F-values and reduced computation time.
Authentication for paper PDF access
A password is required to view paper PDFs. If you are a registered participant, please log on the site from Participant Log In.
You could view the PDF with entering the PDF viewing password bellow.