I. Introduction
Sensor-based recognition of Activities of Daily Living (SbrADL) is a hot topic in digital healthcare, and its research can provide valuable advice for better healthcare and lifestyle. Most SbrADL rely heavily on supervised learning, which requires many labeled sensor data. Usually, we achieve semantic annotation (recognize and label the sensor data) by observing or monitoring the daily life of the participants. In any case, these methods are manually labor-intensive and time-consuming [1]. The lack of labeled sensor data due to cost is known as annotation scarcity [2].