What is the difference between Variational Inference (VI) and Markov Chain Monte Carlo (MCMC) for Bayesian posterior inference?

Machine Learning Hard

Machine Learning — Hard

What is the difference between Variational Inference (VI) and Markov Chain Monte Carlo (MCMC) for Bayesian posterior inference?

Key points

  • MCMC asymptotically samples true posterior
  • VI trades exactness for speed and scalability
  • VI minimizes KL divergence to approximate posterior
  • MCMC is computationally expensive and slow to converge

Ready to go further?

Related questions