November 8, 2017

Jules Webster

OK, so why have I posted a senseless quote from somebody that you have never heard of? Well, for one, the reason you've never head of Jules Webster is because he's not an actual person. Jules Webster is the name we have given to an experiment in Machine Learning.

island, her whereabouts could not be discovered to the island. its matter of the mountain of the reporter, and a man of the companion of the source of the engineer which the mountain spars, the construction of the railow; to so addanting and construction of the fool, was sometimes for the sea and the colonists and the sunks and some on the river was supposed the mountain would have not the departs of the sailor, later, exposed, the ship
- Jules Webster

OK, so why have I posted a senseless quote from somebody that you have never heard of? Well, for one, the reason you've never head of Jules Webster is because he's not an actual person. Jules Webster is the name we have given to an experiment in Machine Learning.

Machine Learning is a technique where you "teach" a computer to do a task instead of giving it exact instructions. For example, instead of writing a program to recognize human faces, you use thousands of pictures of faces to teach it to recognize faces.

The quote above is the output generated by a machine learning algorithm that was trained to write sentences. Obviously it isn't at all legible, but it is actually somewhat impressive when you consider that all that it was trained on was a huge list of letters without any context at all. I used an example script from the Keras library on GitHub. Instead of using the the default training material, though, I fed it a few of Jules Vern's books that I also found on GitHub[1]. I gave it Around the World in 80 Days, The Mysterious Island, A Journey into the Interior of the Earth, and some of Twenty Thousand Leagues under the Sea and Webster's Dictionary. Thus the name "Jules Webster".

It was just a novel experiment but it caught my interest. I'll be working on making it better, hopefully getting it to the point that it actually make sense. I'll be posting my developments here if I make any.


  1. I got it from the GITenberg collection. They have a huge collection of public domain works that can be used by anybody. ↩︎