The rapid growth in healthcare wearables demands reliable activity recognition algorithms. A study utilizing the UK Biobank's accelerometer dataset trained deep-learning models using self-supervision. Results showed a 24.4% median improvement over baseline, and the models demonstrated applicability across various cohorts and living environments. Pre-trained models have been released for broader application in the digital health research community.
The rapid growth in healthcare wearables demands reliable activity recognition algorithms. A study utilizing the UK Biobank's accelerometer dataset trained deep-learning models using self-supervision. Results showed a 24.4% median improvement over baseline, and the models demonstrated applicability across various cohorts and living environments. Pre-trained models have been released for broader application in the digital health research community.