Optimizing steady-state visual evoked potential classifiers for high performance and low computational costs in brain-computer interfacing

Adapting to progressive paralysis: A tongue-brain hybrid robot interface for individuals with amyotrophic lateral sclerosis

“…To the reduce the need for a high-end computer to handle the online processing nodes, the BCI classifier was chosen for its relatively good performance at low computational costs [79]. However, other classifiers may achieve an even better performance but will likely also require longer training sessions and more electrodes.…”

“…The recursive STBF (R-STBF) is a computationally improved version of the STBF, and was chosen for this work due to its relatively high performance at a low computational cost [79]. It averages the segments recursively by applying an exponentially weighted moving average [79] rather than an equally-weighted average as used in earlier research [77], [80]. This reduced the computational cost and weighted the more recent data higher to allow a faster response [79].…”

“…It averages the segments recursively by applying an exponentially weighted moving average [79] rather than an equally-weighted average as used in earlier research [77], [80]. This reduced the computational cost and weighted the more recent data higher to allow a faster response [79].…”

Adapting to progressive paralysis: A tongue-brain hybrid robot interface for individuals with amyotrophic lateral sclerosis

“…The study showed the greatest advantage of classifiers using the SLIC method, as it can split the computations into smaller time segments and become independent of the classification time window. In the study, it was shown to reduce the maximum computational delay time was reduced from 0.33ms to 0.03ms [161]. The study also showed that the previously designed FB-STBF achieved a similar performance but with a slightly higher computational delay (0.07ms).…”

“…However, this would require a classification algorithm to calculate the activation status of each button at a high rate with low computational delay and with high accuracy. Therefore, the classifier was further improved with a focus on reducing the required computing power in our later study [161]. Here a new version of the TRCA was also developed (the SLIC-TRCA) by implementing the SLIC method (similar to the STBF).…”