My PhD thesis introduced model prototyping as a design act that helps envision machine learning in situation with humans before engineering it.
Model prototypes may be considered as design artifacts for machine learning techniques in interactive systems. They enable to test interactive data workflows with concrete algorithmic implementations before starting the engineering of a final learning model. As such, model prototypes enable to generate new ideas and envision new designs of machine learning in situation with human users. This contrasts with engineering sciences approaches to machine learning, which often consider user interaction only after a model is optimized from a large data set.
The notion of model prototype extends that of software prototype to the case of statistical models in machine learning. For example, one may test several model prototypes—e.g., centroid- or density-based—to design the machine learning technique of clustering. In my thesis, focusing on model prototyping over model engineering enabled to study four machine learning techniques in relation to four musical activities: unsupervised learning, reinforcement learning, deep reinforcement learning, and active learning.
PhD thesis (2019)