Get Started
Home
Topics
Search
Library
Research questionHow can wrist-IMU recognition adapt to new fine-grained activities for each user with few labeled examples?Wrist activity recognizers need labeled examples for each user, task, and level of activity detail. Collecting those labels becomes especially difficult when new manipulative, gestural, or procedural activities are introduced.
AI
Machine Learning
Research Paper
Technology
Latest papersRecent research connected to this question, newest first.TransfHAR: Self-Supervised Wrist Representations for On-Demand Activity RecognitionThe source studies TransfHAR, which pretrains on coarse unlabeled wrist-IMU activities and adapts a real-time smartwatch recognizer to user-defined fine-grained activities. Evidence includes three offline cross-dataset evaluations and an in-lab study with 10 participants performing seven novel activities; the reported adaptation uses five examples per class or one one-minute recording per class.research paper · Sep 2, 2026
Related questions
How can video models recognize unseen actions from only a few labeled examples?How can motion-style transfer learn new styles from few paired examples without artifacts on unseen motions?How can robots adapt dressing actions to moving arms from limited static demonstrations despite occlusion and garment contact?How can robots learn robust, generalist bimanual household manipulation from limited human demonstrations?