Likelihood-Free Overcomplete ICA and Applications In Causal Discovery

Inaddition, existingOICA algorithms rely on the Expectation Maximization (EM) procedure that requires computationally expensiveinference oftheposterior distribution ofindependent components.