Privacy-Preserving Motor Intent Classification via Feature Disentanglement

Recent studies have revealed that sensitive and private attributes could be decoded from surface electromyogram (sEMG) signals, which can incur privacy threat to the users of sEMG-based pattern recognition applications. Most studies so far focus on improving the accuracy and reliability of sEMG classifiers, but much less attention has been paid to their privacy. To fill this gap, this study implemented and evaluated a framework to optimize sEMG-based data-sharing mechanism. Our primary goal was to remove sensitive attributes (i.e., identity-relevant information) in the sEMG features before sharing them with primary pattern recognition tasks. We disentangled the identity-insensitive, task-relevant representations from original sEMG features. We shared it with the downstream pattern recognition tasks to reduce the chance of sensitive attributes being inferred by potential attackers. The proposed method was evaluated on data from twenty subjects, with training and testing data acquired 3-25 days apart. Our results showed that the disentangled representations significantly reduced the success rate of identity inference attacks compared to the original feature and its sparse representations generated by the state-of-the-art feature projection methods. The disentangled representation was then evaluated in hand gesture recognition tasks. Our results revealed that the disentangled representations led to higher classification accuracy across classifiers compared with other feature implementations. This work shows that disentangled representations of sEMG signals are a promising solution for privacy-preserving motor intent recognition applications.