Learning personalized ADL recognition models from few raw data.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Human-Computer Interaction

Background:

• Actigraphy-based recognition of activities of daily living (ADL) is crucial for assisted living.
• Current machine learning models require extensive user data for training, raising privacy and time concerns.

Purpose of the Study:

• To develop personalized ADL recognition models using few-shot learning.
• To leverage matching networks for efficient and accurate ADL recognition with limited data.

Main Methods:

• Utilized matching networks, a type of neural network, for few-shot learning.
• Focused on learning personalized ADL models from minimal raw actigraphy data.

Main Results:

• The proposed personalized models demonstrated effectiveness compared to general neural network models.
• The approach showed strong performance even with limited training data.

Conclusions:

• Few-shot learning with matching networks offers a viable solution for personalized ADL recognition.
• This approach addresses data scarcity and privacy issues in assisted living applications.