Fig. 1
A factor graph with two variables. The probability function is p(x 1,x 2)∝f a (x 1)f b (x 1,x 2). It is customary to shade observed variables and leave latent variables unshaded. To calculate the marginal probability of the observed variables, we need to sum out the latent variables. The sum-product algorithm does that efficiently taking advantage of the conditional independence structure of the graph