The use of transfer learning to take estimation of running ground reaction forces via wearables from the treadmill to overground

This study found that a transfer learning approach using ground reaction force (GRF) data from treadmill running helped improve GRF estimation accuracy of a deep learning model during overground running. Leveraging a treadmill dataset 70 times larger than the overground dataset enabled the model to learn generalisable features that could be adapted to the target use case, resulting in more accurate overground<br/>GRF predictions than training on the smaller dataset alone.