Interactive demo · Foundations of Deep Learning · Chapter 11

Linear-Gaussian graphs: ancestral sampling against the derived mean and covariance

Each node is Gaussian with a mean that is linear in its parents, xi = Σj∈pa(i) wij xj + bi + √vi εi (11.9). Sampling the nodes in order (ancestral sampling) gives points whose mean and covariance approach those of the recursions (11.12) and (11.13). A missing link shows up as a zero in the precision matrix Λ = Σ−1, not in Σ: it is a conditional independence, read off by d-separation. Set a weight to 0 to remove its link.

Link weights (0 = no link)

Offsets bi and noise variances vi

Samples

Plot

Try these

In the lecture notes