The Potential of Reinforcement Learning for Live Musical Agents

Collins, Nick (2008) The Potential of Reinforcement Learning for Live Musical Agents. In: International Workshop on Machine Learning and Music 2008, Helsinki, Finland.

Full text not available from this repository.


Reinforcement learning has great potential applicability in computer music, particularly for interactive scenarios and the production of effective control policies for systems. This paper considers in particular the case of interactive
music systems, where a software agent can be trained online during a rehearsal session with a musician. A small-scale system is described which uses the Sarsa(�) algorithm
to update state-action values to determine a policy for output behaviours over a modest feature space. It is hoped that issues arising can be discussed fruitfully in a workshop context.

Item Type: Conference or Workshop Item (Paper)
Schools and Departments: School of Engineering and Informatics > Informatics
Depositing User: Nick Collins
Date Deposited: 06 Feb 2012 20:28
Last Modified: 08 Jun 2012 09:25
📧 Request an update