Widespread in the 1980s, Bayesian networks are simple probabilistic graphical models, commonly used in classification, as well as in computer science in the field of artificial intelligence machine learning. They are both models of knowledge representation ; “calculating machines” conditional probabilities but also decision support systems. Automatic learning of a Bayesian network from observations of random variables makes it possible to extract knowledge from the data.
Until the 2000s, the methods used were limited to around thirty variables. In the last ten years, great progress has been made with integer linear programming and the optimal construction of cuts, pushing the limit to a hundred variables.
Faster, more efficient
Scientists from the Applied Mathematics and Informatics Toulouse (MIAT) laboratories of the INRAE Occitanie-Toulouse center and Applied Mathematics and Informatics (MIA) from the Versailles-Grignon Paris center have developed a new method with enhanced efficiency for a shorter calculation time . Called ELSA, for Exact Learning of bayesian network Structure using Acyclicity reasoningthis method uses the idea of cuts by integrating it into CPBayes, a constraint programming tool dedicated to this learning problem.
“Using another approach, we have developed a high-quality but above all much faster calculation method,” explains Simon de Givry, research fellow at the MIAT laboratory. On a test set composed of about forty problems, having at most 500 random variables and over the same allotted time of 90 hours, ELSA solves 23 at the optimum, while the GOBNILP method only solves 9 and CPBayes only 4”.
Among the perspectives for improving work, it is envisaged to use decision diagrams to handle very large domains of values more efficiently. This would further accelerate the operations carried out by ELSA.
One of the tools for designing future sunflower varieties
As part of the Sunrise program, studies have been carried out by the Plant-Microorganism-Environment Interactions Laboratory (LIPME) to identify genes of interest for drought tolerance and to model the agronomic characteristics of future sunflower varieties, bearing these genes. The ELSA method has contributed to a better understanding of the genetic and molecular bases controlling the physiology and development of the plant to predict the characteristics of hybrids.
We want to thank the writer of this article for this amazing material
More efficient calculation models at the service of agronomic research
Our social media pages here and other pages on related topics here.https://www.ai-magazine.com/related-pages/