Incremental Learning of Human Activities in Smart Homes.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Sensor-based human activity recognition (HAR) is crucial for monitoring and understanding human behavior.
• Current HAR systems often struggle with adapting to evolving behaviors and identifying novel activities.
• Continuous learning and adaptation are essential for robust HAR systems.

Purpose of the Study:

• To propose a compression-based method for incremental learning in human activity recognition.
• To enable HAR systems to continuously adapt to new behaviors while retaining past knowledge.
• To develop a system capable of detecting novel and potentially risky human behaviors.

Main Methods:

• A novel compression-based approach for incremental learning was developed.
• The method allows for the continuous assimilation of new behavioral data.
• Prior knowledge is retained, and novel activities are identified.

Main Results:

• The proposed method demonstrated effective incremental learning on three public smart home datasets.
• The system successfully retained knowledge of previously learned activities.
• The approach showed promise in highlighting novel behaviors.

Conclusions:

• Compression-based incremental learning offers a viable solution for adaptive human activity recognition.
• This method enhances the ability of HAR systems to handle evolving human behaviors.
• The approach has significant implications for identifying emerging and potentially risky activities.