Apr 04, 2017 12:00:00

Automatically generate Pokemon names & characteristics with neural networks

In the latest Pokemon series 'Pokemon Sun Moon' released in November 2016, there are 801 types of Pokemon appearing in the series. There are already quite a lot of types of Pokemon, but attempts to have neural networks automatically generate combinations of names and characteristics that existing Pokemon do not have are attracting attention on the Internet.

Postcards from the Frontiers of Science • Pokemon generated by neural network
http://lewisandquark.tumblr.com/post/147834883707/pokemon-generated-by-neural-network

Janelle Shane , a researcher at Boulder Non-Linear Systems, uses an open source neural network ' char-rnn ' created for Torch , a deep learning development environment, to automatically automate ' recipe ' and ' superhero name '. It seems that he was playing with it.

Following these, Shane's challenge was to 'train a neural network to randomly generate a combination of Pokemon names and characteristics.' After 50 trainings, the neural network seems to have learned every word about Pokemon and produced multiple interesting combinations.

Shane introduced an example of a combination of names and characteristics of Pokemon generated by neural networks on his blog, but the artist iguanamouth who saw it introduced from the combination of names and characteristics. An illustration of Pokemon is drawn and released, and an unprecedented original Pokemon is born through collaboration between neural networks and human power.

The names and characteristics of Pokemon generated by Shane's neural network and the illustrations drawn by iguanamouth inspired by them are as follows.

Name: Quincelax