Broadcast: Events

Making music with machine learning and dynamical systems

Tuesday 24 March 16:00 until 17:30
Arun 401
Speaker: Chris Keifer
Part of the series: COGS Seminars

Machine learning models are becoming a common part of the computer-musician's creative toolbox, for varied applications in composition, mapping, sound design and production. Many of these tools use only pre-trained models; there's an opportunity to open up the process of designing and training machine learning models to musicians for deeper creative engagement. This requires fresh approaches compared to current commonly-used deep learning techniques, with models that are lightweight and open to realtime manipulation.

This presentation will demonstrate such an approach, showing new reservoir computing techniques for sound synthesis, pattern generation and pattern recognition, with conceptor-controlled recurrent neural networks. These techniques will be demonstrated using Sema, a browser-based livecoding platform for machine learning and music, which has been made as part of the AHRC project 'Musically Intelligent Machines Interacting Creatively'.

By: Simon Bowes
Last updated: Wednesday, 12 February 2020