The neural network was taught to think of Halloween costumes
A researcher from the USA taught a neural network to come up with ideas for Halloween costumes.
A researcher from the USA taught a neural network to come up with ideas for Halloween costumes. Reported by Business Insider.
To train the neural network Janelle Shane used the 4500 names of the costumes that helped her to collect Internet users. In the collection there was a lot to this classic holiday costumes: witches, ghosts, pirates, Batman, cats, vampires and pumpkins. 300 of the names of the suits contained the word “sexy”, including Dumbledore, Darth Vader, Benjamin Franklin, monkfish, eyeball and even the Dvorak keyboard.
Initially, the neural network gave a lot of absurd ideas and repeats the most frequently occurring in the original word list, and later began to appear more meaningful options, for example, the Queen butterfly, or a sad king-pumpkin, dragon or freedom. Was in the list also Professor Panda, space Batman, Martian devil disco monster.
Thanks for giving us Professor Panda and Shark Princess. pic.twitter.com/LDx66Vh7eU
— Spooky Herbie (@HerbLovesTech) 30 Oct 2017
During the download an error has occurred.I’m sure the neural network, coming up with costumes for Halloween will be very out of place, to create creative things that people love so much. It can output its own rules based on what he sees, not to what he remembers.Janelle Shane
In her view, a neural network can be useful for inventing names, when people do not have enough fresh ideas.
Previously Shane has already created a neural network that composes names for craft Beers and diseases.