This performance is about a coding dialogue between my current and past self. During the piece I will collaborate with a machine learning algorithm trained on code from my past performances, accepting (or rejecting) suggestions made by the computer throughout concert based on the code I type, resulting in both expected and unexpected suggestions. Both the code, as well as the suggestions made by the computer, will be made visible to the audience.
Over the past years I've collected a database of code, programmed during my live coding performances. This was done by saving a textfile every time I evaluate code. With this database I explored training various neural networks such as RNN and LSTM networks as well as more basic nth-order Markov-chains. They are not perfect, sometimes giving incorrect code, but in the form of a collaborator they can provide useful suggestions.
This work is part of the project ./drum.code and funded by the Creative Industries Fund NL.
|
