Day 96: Neural Network Produces MJ Lyrics

October 6, 2020

So I have been working through a Recurrent Neural Network (RNN) that can predict any text. So of course I thought why not give it some famous lyrics and see if it can produce a cool new song (the lyrics at least).

The RNN was written by Andrej Karpathy in Python. It was shown as part of his Stanford lecture on RNNs which I am currently studying.

Do I understand everything that is going on here? Nope, but I wanted to start with running the code and working through the initialization. Basically, the RNN is fed a text file and then trains to be able to predict and produce a new text file (at least that’s my understanding so far).

Here’s the result below. I thought it would be a cool idea if new lyrics are written by artificial intelligence. This RNN is not limited to just lyrics. It can also work with a text file containing characters or symbols from any language. The possibilities are really interesting.

The lyric prediction output shown here is not perfect but its pretty good for a robot written in under 150 lines of code. Btw it trained with the data over 70000 times. It never stopped trying to improve and get the loss to 0.

The lyric prediction output shown here is not perfect but its pretty good for a robot written in under 150 lines of code. Btw it trained with the data over 70000 times. It never stopped trying to improve and get the loss to 0.

Anyways I still have a long way to go but I thought this was a fun thing to share. I have included the Stanford lecture below. Until next time!

YouTube player

Post Views: 240