9 Kung Fu Levels in Understanding Expectation-Maximization

Translated from Wechat Official Account: AI2ML, by Shi Chunqi

Level 1: EM = E + M

Level 2: EM is to maximize a local lower bound

Level 3: EM is a probabilistic version of K-Means

Level 4: From EM to Generalized EM

Level 5: A special case of Generalized EM is VB-EM

Level 6: A special case of Generalized EM is Wake-Sleep Algorithm

Level 7: A special case of Generalized EM is Gibbs sampling

Level 8: Wake-Sleep algorithm is a simplication of combined VAE and GAN

Level 9: The unification of KL-divergence

kungfupanda