Training – Baum-Welch (no maths)

By summing over all possible state sequences we marginalise away the state sequence: we don’t care what value it takes.