Weir R. F. (2018) The EMG Properties Limit Ultimate Classification Accuracy in Machine Learning for Prosthesis Control. Constructivist Foundations 13(2): 265–266. https://cepa.info/4617
The EMG Properties Limit Ultimate Classification Accuracy in Machine Learning for Prosthesis Control.
Constructivist Foundations 13(2): 265–266.
Fulltext at https://cepa.info/4617
Open peer commentary on the article “Applying Radical Constructivism to Machine Learning: A Pilot Study in Assistive Robotics” by Markus Nowak, Claudio Castellini & Carlo Massironi. Upshot: Machine learning (ML) has been applied in many forms and under many names over the years to the problem of mapping arrays of surface electromyogram (EMG) signals measured on the arm of a person with an amputation and then trying to correlate those signals to the control of multi-degree-of-freedom prosthetic arms. While being intrigued by the idea of the interactive machine learning (iML) component of the study, I am not surprised that iML did not do noticeably better than standard approaches. The issue, as demonstrated by many researchers, is not our ability to do ML but rather the fundamental problem associated with using EMG as the inputs to the ML system and the clinical issues associated with stable acquisition of those signals.