Correction: Pre-training, personalization, and self-calibration: all a neural network-based myoelectric decoder needs

Chenfei Ma^*Xinyu JiangKianoush Nazarpour^*

• School of Informatics, University of Edinburgh, Edinburgh, United Kingdom

Error in figure/table Wrong content There was a mistake in figure 1 as published. The EMG traces in figure 1(a) are not visible, and the arrow in figure 1(d) is deformed. The corrected figure 1 appears below. The original version of this article has been updated. Abstract Adding/removing text In the abstract, an incorrect word (“However”) is used in “However, the characteristics of the EMG signal can change significantly over time, even for the same user, leading to performance degradation during extended use.”. This has been corrected to read: “Meanwhile, the characteristics of the EMG signal can change significantly over time, even for the same user, leading to performance degradation during extended use.”. The original version of this article has been updated. Text correction Adding/removing text The following sentences were not separated, and an incorrect word was used. “Each test block lasted about 2 min and the total duration of the experiment on each day was 40 min, including intervals labels are balanced in each day” A correction has been made to the section 2.5 Data Collection, Paragraph 2: “For each of the 28 participants, data from 2 days was collected. On the first day, a calibration session was first conducted, with one trial per hand gesture. Each trial is of 2 s duration, and participants shape their hand in the first second and holding the same hand gesture in the last second. A 2-s inter-trial interval was provided. In all data collection of the 28 participants, we followed the same trial duration and inter-trial interval. Data collected in the calibration session was used to personalize the model in one shot. After the calibration session on the first day, five test blocks were performed, with 30 trials per block. Therefore, we collected 150 trials for all 6 hand gestures, that is, 25 trials per gesture. Participants could take. flexible self-paced breaks between test blocks, typically 5 min. On the second day, participants directly started five more testing blocks without any calibration session. Each test block lasted about 2 min and the total duration of the experiment on each day was 40 min, including intervals. Labels are balanced for each day. By exploring the performance variation along all test blocks on the same day and on 2 days, we could compare the robustness of different models during long-term use.” The original version of this article has been updated.