Signal Processing Approaches to Minimize or Suppress Calibration Time in Oscillatory Activity-Based Brain–Computer Interfaces (2015)
Open in webOpen in zoteroOpen pdf
1 Abstract
One of the major limitations of Brain-Computer Interfaces (BCI) is their long calibration time, which limits their use in practice, both by patients and healthy users alike. Such long calibration times are due to the large between-user variability and thus to the need to collect numerous training electroencephalography (EEG) trials for the machine learning algorithms used in BCI design. In this paper, we first survey existing approaches to reduce or suppress calibration time, these approaches being notably based on regularization, user-to-user transfer, semi-supervised learning and apriori physiological information. We then propose new tools to reduce BCI calibration time. In particular, we propose to generate artificial EEG trials from the few EEG trials initially available, in order to augment the training set size. These artificial EEG trials are obtained by relevant combinations and distortions of the original trials available. We propose 3 different methods to do so. We also propose a new, fast and simple approach to perform user-to-user transfer for BCI. Finally, we study and compare offline different approaches, both old and new ones, on the data of 50 users from 3 different BCI data sets. This enables us to identify guidelines about how to reduce or suppress calibration time for BCI.
2 NOTES
In this paper, Lotte proposes three data augmentation techinques, two are very similar to each other (S&R and TF-S&R) and the third is a very different one, which is based on computing the signal power from some trials and using it as a ratio when reconstructing a third signal. While TF-S&R does seem interesting I haven’t seem any paper using it, instead I have seem plenty usage of S&R, and in all those it has been very efficient (mostly used with transformers). The plots are awful and there isn’t a single table with number so I can easily compare results, so I will base my conclusions on the text. “For several of the calibration time reduction methods analyzed we can observe that only 10 trials per class are enough to reach the same performances as that obtained with 30 trials per class with the baseline design, hence effectively reducing BCI calibration time by 3.” (Lotte, 2015, p. 22)