Fig. 1.

Learning the data. Depiction of maximum likelihood training of an RNN. $x^t$ are the target sequence tokens we are trying to learn by maximizing $P(x^t)$ for each step