Inter-subject Variance Transfer Learning for EMG Pattern Classification Based on Bayesian Inference.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Signal Processing

Background:

• Electromyogram (EMG)-based motion recognition typically requires extensive subject-specific labeled data.
• This data collection process is time-consuming and burdensome for individuals.

Purpose of the Study:

• To develop an efficient transfer learning method for EMG motion recognition.
• To reduce the need for extensive subject-specific data collection by leveraging inter-subject information.

Main Methods:

• Proposed an inter-subject variance transfer learning method using a Bayesian approach.
• Transferred variance patterns from pre-trained source subjects to a target subject.
• Introduced a coefficient to control the amount of transferred information.

Main Results:

• Demonstrated effective EMG motion recognition with limited target calibration data.
• Showcased the superiority of the proposed variance transfer strategy over existing methods.
• Validated the approach using two independent EMG datasets.

Conclusions:

• The proposed Bayesian variance transfer learning method significantly improves EMG-based motion recognition efficiency.
• This approach alleviates the burden of extensive data collection for individual subjects.
• Variance patterns offer a promising avenue for inter-subject transfer learning in EMG analysis.