We’ve said it before, but allows us to say it again: research scientist Janelle Shane has stumbled upon an entertainment goldmine with her creative use of neural networks, or computer software that is capable of machine learning. She’s previously user neural networks to make lists of fake Pokémon names and awful pickup lines, and now she’s back with a slew of computer-generated paint colors that, based on name alone, you might not want to paint your living room with.
Anybody who’s ever been to a Sherwin-Williams knows that different hues of paint have flowery and descriptive names like Dandelion Fields or Rainy Slate (we don’t know if those are real, we just made them up), and when Shane fed “about 7,700” of the store’s paint colors into her neural network, the results were even more abstract and bizarre.With that information, the network eventually learned how to produce valid RGB values (numbers that indicate what ratios of red, green, and blue make up a color), and much later on in the process, it learned how to give somewhat coherent names to these colors. Take a look at some of the hues the network eventually produced:
If you want to give your bathroom a fresh, dope look, “Dope” might be your best bet (although honestly, colors like “Turdly,” “Stanky Bean,” and “Bank Butt” might be more fitting). We can see why actual humans come up with paint color names: computers just aren’t up to the task quite yet.
What other neural network fun have you come across recently? Did you see the list of Dungeons & Dragons spell names and Magic: The Gathering cards? Share what else you’ve found in the comments below!
Featured image: United Soybean Board/Flickr