Auto-encoding adjacency matrix causality model with page rank

An auto-encoding adjacency matrix is learned with gradient descent for the iris dataset. The neural network is the form of: AX + B = X. Page rank is then applied to the adjacency matrix, and it is found that the petal length is the most important property of irises.