Verbleibende Zeit $0$:$38$:$14$

Which of the following statements about LSTM applications are true?

a$.$ Part$-$of$-$speech tagging is an example for sequence$-$to$-$sequence learning with one$-$to$-$one correspondences.

b$.$ Alex Graves showed in his $2013$ paper how LSTM can be used to generate text sequences simply by predicting one character at a time.

c$.$ Hochreiter et al$.(2001)$ showed how RNNs can "learn to learn" by arranging labelled datasets as sequences. At time $t,$ the RNN takes an input $x_t$ along with the label $y_t$ and makes a prediction hat$(y{)}_t.$

d$.$ Sequence$-$to$-$sequence learning tasks without one$-$to$-$one correspondences can be tackled using an encoderdecoder architecture.

e$.$ Language translation is an example for sequence$-$to$-$sequence learning tasks with one$-$to$-$one correspondences.

f$.$ Karpathy and Fei$-$Fei $(2015)$ propose to perform image captioning using a CNN image encoder network in combination with an LSTM next$-$character text generation network.