Online Streaming Features Causal Discovery Algorithm Based on Partial Rank Correlation | IEEE Journals & Magazine | IEEE Xplore

Online Streaming Features Causal Discovery Algorithm Based on Partial Rank Correlation


Impact Statement:Online local causal structure learning algorithms are a hot topic in the field of causal discovery. The goal of these algorithms is to solve the causal structure learning...Show More

Abstract:

Aimedat the problem of dynamic causal discovery in the era of artificial intelligence, this article combines partial rank correlation coefficients and streaming features ...Show More
Impact Statement:
Online local causal structure learning algorithms are a hot topic in the field of causal discovery. The goal of these algorithms is to solve the causal structure learning problem of multivariate discrete or continuous data under dynamic conditions (assuming that feature data are generated in a streaming way). Current learning algorithms based on stream features can effectively process dynamic causal structure learning of discrete data or continuous data subject to a linear Gaussian distribution. However, most algorithms require discrete processing of input data in dynamic causal structure learning of continuous data, which leads to data distortion. In the real world, most continuous data distributions do not conform to the Gaussian distribution and linear hypothesis, which poses a challenge to the processing of this kind of problem. This paper proposes an online streaming feature causal discovery algorithm based on partial rank correlation coefficients. The algorithm has no restriction...

Abstract:

Aimedat the problem of dynamic causal discovery in the era of artificial intelligence, this article combines partial rank correlation coefficients and streaming features in the field of Bayesian network structure learning and proposes a new online streaming feature causal discovery algorithm based on partial rank correlation named the partial rank casual discovery streaming feature based algorithm. This algorithm is not only suitable for Bayesian causal structure learning in dynamic feature spaces generated by sequential streams of features but can also effectively process multivariate linear Gaussian and nonlinear non-Gaussian data. We present three main contributions. First, for arbitrarily distributed datasets, which can be generated by the additive noise model, we proved that the partial rank correlation coefficient can be used as the criterion for the conditional independence test and explored the distribution of corresponding statistics. Second, the PCSDSF algorithm redefined the...
Published in: IEEE Transactions on Artificial Intelligence ( Volume: 4, Issue: 1, February 2023)
Page(s): 197 - 208
Date of Publication: 14 February 2022
Electronic ISSN: 2691-4581

Funding Agency:


Contact IEEE to Subscribe

References

References is not available for this document.