This master's thesis can explore improvisation in several ways, including but not limited to:
- Human-AI social interaction
- Human-Robot social interaction
- Music making with AI
- Performance with AI
The project can make use of:
- sequence modelling and prediction
- training deep learning models such as Transformers, DRL
- studies of real-time interactions
- development of software and/or hardware for Human-AI interaction
Modeling and Analyzing Improvisational Strategies in Music or Dance
Goal: AI as a tool to analyze and visualize creativity
- Use deep sequence models (Transformers, Variational Autoencoders, Recurrent Neural Networks) to analyze large datasets of musical or dance improvisations.
- Model patterns of exploration vs. repetition, risk-taking, and emergent structure in improvisation.
- Build an interactive system that lets users visualize their own improvisational style in real time, supported by AI analysis.
Datasets:
- Music improvisation data (MAESTRO (MIDI + audio of piano performances, with timing and dynamics), Groove MIDI Dataset (drumming improvisations))
- Dance improvisation data (AIST, CMU Mocap, DeepDance).
- You could also record your own dataset of improvisation (music or dance) for personalized data.
To learn more, contact