EM two-example flow A two-lane diagram showing the E-step and M-step workflow for a height mixture example and a two-coin example. EM: E-step -> M-step in two examples Observed data stay fixed. The hidden assignment weights and parameters update each iteration. Example 1: mixed height data Example 2: two coins with unknown round source Observed X height values x1 ... xN hidden Z: which group theta0: pi, mu, sigma E-step compute gamma_iA, gamma_iB posterior group weights keep overlap uncertain M-step weighted mean update weighted variance update weighted mixture proportion theta1 repeat until stable Observed X 5 rounds of H/T counts hidden Z: coin A or B start pA=0.2, pB=0.7 E-step LA = pA^h (1-pA)^t rA = LA / (LA + LB) round source weights M-step weighted heads / flips pA -> 0.347 pB -> 0.529 next use new pA, pB again Validation point: if hidden labels were known, the weights collapse to 0/1 and M-step becomes ordinary grouped MLE.