Research On Causal Inference Method Based On Schmidt Orthogonal Matrix Verification

With the development of machine learning,big data technology,and cloud computing technology,a series of research hotspots have emerged in the field of artificial intelligence.Causality discovery is one of them.The 2011 Turing Award winner,the Bayesian network proposer,Judea Pearl once said that artificial intelligence has fallen into the trap of correlation and neglected causation.It believes that scholars should focus their attention on causation because it may be artificial The only path of intelligence^[1].Although many methods for finding causality from observed data have been proposed in recent years,causal inference still has a low accuracy rate on mixed-type data and multivariate data,and the identified causality diagram and the real phenomenon are quite different.Based on this,this paper first discusses the related theories of causal inference and classic causal inference methods.Then,the main work of this paper is to propose a bidirectional addition based on neural network for the causal inference of binary variable mixed type observation data.Sexual noise model?Dual-ANM-MM?to identify the causal direction between variables.The model can better deal with the problem of causal inference based on the binary variable observation data generated by the mixed additive noise model.This method improves the loss function of the original mixed additive noise model,adds the Hilbert Schmidt independence of the result variables and distribution parameters,and then uses the gradient descent method to optimize the improved loss function,and finally Compare the distribution parameters with the causal variables,and the independence between the outcome variables to determine the causal direction of the mixed binary variables.This paper theoretically verifies the feasibility of the method and verifies it on simulated data and causal-effect public data sets.The experimental results show that the algorithm is more accurate than traditional IGCI,ANM,PNL,LiNGAM,SLOPE methods.Certainly improve.A framework for solving the causal inference problem among multiple variables?CIMV?is proposed on the observation data of multiple variables.The framework can infer a complete causal network structure diagram from the observation data.The framework needs to pre-set two thresholds?,?parameters?through experimental testing?,where?is used for mutual information independence test to identify direct nodes,and?is used for conditional mutual information to delete wrong direct nodes,thus forming a cause and effect Undirected graph;then use the CDI algorithm to identify the direction in the V-structure and the triangle structure on the logical structure,and then use the method of causal inference to mix the multiple binary variables?if the binary variable is The data is mixed type data,then use Dual-ANM-MM?until the direction of the edge around each node in the undirected graph is identified,that is,a complete causal network graph is obtained.In this paper,experiments were conducted on simulated data and real Bayesian networks.The results show that the proposed framework can better deal with the causal inference problem among multiple variables.