• Try the LSTM example for Alice in Wonderland to generate baby names using the datasets compiled by Mark Kantrowitz, available at: https://www.cs.cmu.edu/Groups/AI/areas/nlp/corpora/names/0.html.
• Music generation is also a sequential prediction problem, as we mentioned. You can take a look at the version of the Nottingham Music Database written in ABC notation at http://abc.sourceforge.net/NMD/ to teach the computer to write music, using a similar recurrent neural network.
• Refactor the code on our recurrent neural network from scratch (the R6 implementation) so that it uses the forward_step function within the bptt function.
• We ignored bias terms in our recurrent neural network implementations. How will that affect the performance, if at all? Try to modify the code to include bias terms.
• Modify the code for the network using the cross-entropy method to use natural evolution strategies: https://blog.openai.com/evolution-strategies/.