Portfolio of compositions: Creating electroacoustic works through the sonification of recursive neural networks, and exploring the creative use of in

This research will explore the sonification of data from recursive neural networks as a way to generate new,
engaging musical materials, to create a portfolio of electroacoustic compositions that can provide some insights
into how these networks operate, including the rate of data transformation and the effect of recurrent cells on that
transformation. I hope this work can bring into focus the potential for human control in the use and design of deep learning networks for creative purposes, and in algorithmic sonification.
Research questions
- How can recursive neural networks' operating data be convincingly mapped as audio information to be
listened to?
- Of the different data streams obtainable from an RNN, which can be used to generate new, interesting
musical materials and structures that are intrinsically linked to the networks' algorithmic processes?
- How can this sonified output be used to create a portfolio of spatialised sound-art installations that provide
insights into the way RNNs operate to a non-expert audience?