<ul><li>LSM-2 with Adaptive and Inherited Masking (AIM) is a novel self-supervised learning approach introduced for learning from incomplete wearable sensor data.</li><li>AIM uses learnable mask tokens to handle missing data and learn robust representations without explicit imputation, improving performance across various tasks.</li><li>Pre-trained on a large dataset of 40M hours of multimodal sensor data, LSM-2 with AIM achieves superior performance in tasks like classification, regression, and generative modeling.</li><li>LSM-2 with AIM demonstrates strong scaling performance and maintains high accuracy even under targeted missingness scenarios, making it suitable for real-world wearable data applications.</li></ul>

LSM-2: Learning from Incomplete Wearable Sensor Data

Discover more