Modern bionic hand prostheses feature unprecedented functionality, permitting motion in multiple degrees of freedom (DoFs). However, conventional user interfaces allow for contolling only one DoF at a time. An intuitive, direct and simultaneous control of multiple DoFs requires machine learning models. Unfortunately, such models are not yet sufficiently robust to real-world disturbances, such as electrode shifts. We propose a novel expectation maximization approach for transfer learning to rapidly recalibrate a machine learning model if disturbances occur.

In our experimental evaluation we show that even if few data points are available which do not cover all classes, our proposed approach finds a viable transfer mapping which improves classification accuracy significantly and outperforms all tested baselines.