← All student projects · Artistic research projects

Training data as creative material

Artistic research treating dataset curation and model training as part of musical practice.

Level: Honours, Masters
Length: Honours projects run over two semesters; two semesters are preferred for Masters.
Prerequisites: An established musical practice and Python experience. COMP4350/COMP8350 or COMP1720/COMP6720 is a good background.

IMPSY models are trained on a musician’s own playing, so choosing what to record is a creative decision: should you record your best playing, your strangest, or just everything? Our previous research didn’t look at how these choices shape an instrument. In this project you’ll treat dataset curation and model training as part of your artistic practice. You’ll build several models from deliberately different datasets, perform with each, and document how your curation choices show up in the instrument’s behaviour and your music. Your material is the interaction logs IMPSY records (not audio), so curation means choosing which sessions and passages go into each dataset.

• Minimum: three or more models trained on curated datasets, a performance with each, and a reflective analysis linking dataset choices to musical outcomes. Alongside the reflection, report dataset sizes, training curves, and a simple comparison of each model’s behaviour (e.g., the timing and range of its generated gestures).
• Stretch: a public performance or composition that uses the differences between models as structural material.

Tags: #artistic-research #machine-learning #small-data #IMPSY

How to apply

Contact Charles Martin with your CV, your unofficial transcript (if you are an ANU student), and a brief statement (200 words) explaining how you would approach this project. Before applying, read the Join page and our project expectations.