We wondered if there was a way to leverage computers and hundreds of pre-existing recipes to create the most average chocolate chip cookie. Would it be bland and unremarkable? Or, perhaps like averaging human facial features, the results would be even better than each of its individual parts. Maybe an average cookie would be the most delicious of them all.
But what is an average cookie? We decided to interpret this idea using three different methods: a mathematical average, predictive text algorithms, and neural networks. After feeding each algorithm over 200 chocolate chip cookie recipes, they each generated something new. And, yes, we actually baked them.
The writer Michael Chabon likens the films of Wes Anderson to “scale models” or “boxed assemblages” built from “the bittersweet harvest of observation and experience.”
These “models” are carefully constructed out of wood and paint, text and image, long tracking shots and carefully framed subjects. Anderson is a meticulous world builder in both visual and thematic construction.
The Life Aquatic was the first Anderson movie I really fell in love with, and as I continue to watch more of them, I find myself pondering just what it is that makes an Anderson film Andersonian. Is it the carefully chosen color schemes or the symmetrical compositions? The recurring themes of family and fracture, of discovery and triumph? Or is it the brief magical flashes of the surreal?
Anderson certainly has a style, and his visual motifs are what I want to explore in this essay.
This is a fascinating look at how to perform machine learning on a data set: in this case, the visual motifs of Wes Anderson films. Nicely presented too. Better on desktop.
An important part of starting a new band is choosing an appropriate name. It is crucial that the name be unique, or you could risk at best confusion, and at worst an expensive lawsuit.
The neural network is here to help.
Prof. Mark Riedl of Georgia Tech, who recently provided the world a dataset of all the stories with plot summaries on Wikipedia, (enabling this post on neural net story names) now used his Wikipedia-extraction skills to produce a list of all the bands with listed discographies – about 84,000 in all.
I gave the list to the Char-rnn neural network framework, and it was soon producing unique band names for a variety of genres. Below are examples of its output at various temperature (i.e. creativity) settings.
Come for the funny names, stay for the bizarre shark influence.
A couple of years ago someone attempted to make a list of every video game ever made, and put it in a 6.5MB flat file. Like any sensible person, I used it to train a recurrent neural network.
- Metal Cat (2001, Sega) (Windows)
- Spork Demo (?, ?) (VIC-20)
- Black Mario (1983, Softsice) (Linux/Unix)
- Soccer Dragon (1987, Ange Software) (Amstrad CPC)
- Mutant Tycoon (2000, Konami) (GBC)
- Dick of the King (2007, Activision) (PC-9801)
- Spork Race (Universe) (1990, Atlus) (Arcade)
The ‘Spork’ franchise sounds like something I’d play, and ‘Black Mario’ seems sufficiently inclusive.
See also these wonderful recipes generated using a predictive text interface:
And Friends episodes: