Abstract
Brain-Machine Interfaces (BMIs) have recently emerged as a clinically viableoption to restore voluntary movements after paralysis. These devices are basedon the ability to extract information about movement intent from neural signalsrecorded using multi-electrode arrays chronically implanted in the motorcortices of the brain. However, the inherent loss and turnover of recordedneurons requires repeated recalibrations of the interface, which canpotentially alter the day-to-day user experience. The resulting need forcontinued user adaptation interferes with the natural, subconscious use of theBMI. Here, we introduce a new computational approach that decodes movementintent from a low-dimensional latent representation of the neural data. Weimplement various domain adaptation methods to stabilize the interface oversignificantly long times. This includes Canonical Correlation Analysis used toalign the latent variables across days; this method requires priorpoint-to-point correspondence of the time series across domains. Alternatively,we match the empirical probability distributions of the latent variables acrossdays through the minimization of their Kullback-Leibler divergence. These twomethods provide a significant and comparable improvement in the performance ofthe interface. However, implementation of an Adversarial Domain AdaptationNetwork trained to match the empirical probability distribution of theresiduals of the reconstructed neural signals outperforms the two methods basedon latent variables, while requiring remarkably few data points to solve thedomain adaptation problem.