Evaluation without ground-truth
By J. Pauwels
J. Pauwels talks to ADASP about evaluation without ground-truth in the data-driven age.
As music information retrieval has firmly entered the data-driven age, the availability of labeled data has become a major factor in making progress. Datasets are now a scarce resource and a race to harvest this resource is underway, in which academia can’t always keep up with industry. In this talk, we’ll explore ways to work around this reliance on ground-truth, by revisiting the basics of the scientific method.
Johan Pauwels is a researcher and lecturer in artificial intelligence, signal processing and data science applied to music. He received Master of Science degrees in electrical engineering (KU Leuven ‘06) and artificial intelligence (KU Leuven ‘07). In 2016, he obtained a PhD from Ghent University on the topic of automatic harmony recognition from audio. He has contributed to a variety of national and international research projects while working at Ghent University, IRCAM and Queen Mary University of London, and has taught at multiple universities.