Data from: Long-term unsupervised recalibration of cursor BCIs

Data accompanying the manuscript "Long-term unsupervised
recalibration of cursor BCIs", consisting of closed-loop cursor
control datasets collected with participant T5. The dataset includes
historical sessions as well as new online tests of recalibration methods
collected specifically for this manuscript. It also contains results of
parameter optimizations and cursor control simulations. Personal use data
collected with participant T11 (Figure 6) is not included, as it may
contain PHI. The README.md file describes each of the four data formats
included. This data is meant to be used with the accompanying github
repository: guyhwilson/nonstationarities: Unsupervised recalibration
project. Original abstract from the manuscript: Intracortical
brain-computer interfaces (iBCIs) require frequent recalibration to
maintain robust performance due to changes in neural activity that
accumulate over time. Compensating for this nonstationarity would enable
consistently high performance without the need for supervised
recalibration periods, where users cannot engage in personal use of their
device. Here we introduce a hidden Markov model (HMM) to infer what
targets users are moving toward during iBCI use. We then retrain the
system using these inferred targets, enabling unsupervised adaptation to
changing neural activity. Our approach outperforms distribution alignment
methods in large-scale, closed-loop simulations over two months, and in
closed-loop with a human iBCI user over one month. Leveraging an offline
dataset spanning five years of iBCI recordings, we further show how
recently proposed data distribution-matching approaches to recalibration
fail over long time scales. Only target-inference methods appear capable
of enabling long-term unsupervised recalibration, while
distribution-matching methods appear to accumulate compounding error over
time. Finally, we show offline that our approach also performs well on
freeform datasets of a person using a home computer with an iBCI. Our
results demonstrate how task structure can be used to bootstrap a noisy
decoder into a highly-performant one, thereby overcoming one of the major
barriers to clinically translating BCIs.