Louis Bradshaw

Profile

Project title:

Neuro-symbolic music models

Abstract:

In recent years, neural music generation has seen rapid progress thanks to generative deep learning models. In this research project, I propose to investigate how we can utilise neuro-symbolic music models to aid with, and analyse, the compositional process. More specifically, given a corpus of music in the form of MIDI or sound files, how may we utilise modern advancements in generative deep learning (such as Transformers, GANs, Diffusion Models, etc.) to create an AI system capable of analysing symbolic representations of music, both in and out of distribution. Furthermore, I propose to investigate how neuro-symbolic AI can be used to add to musical culture in ways other than automatic music generation.

C4DM theme affiliation:

Music generation

Research

Research Interests:

Generative music, Transformers, Foundation models