For the first seminar of 2023 we are delighted to welcome Professor Devaney (Brooklyn College). She will be speaking about Integrating expert domain knowledge into computational models of musical understanding.
Humans are able to learn with greater efficiency than machine learning models, in large part because they learn not just from exposure, but also from domain knowledge, which includes codified knowledge and guided practice. Examples in the musical domain are established pedagogies found in music theory textbooks and the skilled interpretative practice found in musical performances. In music information retrieval (MIR), the music understanding process has been approximated by individual tasks such as chord recognition, rather than treating such single-label classification tasks as only one step in analysing musical signals. This talk discusses ongoing work of how, in the Western Art music domain, chord recognition can be used to parse the musical surface into structural and non-structural components using not just exemplar-based learning but also domain knowledge derived from music theory textbooks and music performance data.
Datasounds, datasets and datasense research network 