Lab11: GRU, LSTM and Seq2SeqΒΆ

In this lab we introduce two more powerful variants of RNNs, namely, Gated Recurrent Units (GRUs) and Long Short Term Memorys (LSTMs). We then compare their performance with Elman RNN in the same square sequence direction binary classification task.

The second part of this lab focuses on the Sequence to Sequence models, first introduced by Ilya Sutskever, Oriol Vinyals, Quoc V. Le from Google at NeurIPS 2014. Then we will forcus on the decoder part for a Surname Generation task unconditioned and conditioned.

Credit: the notebooks are adapted from: