Yan introduced SSL model training within FL for speech, video, and image data. For speech, Yan studied the feasibility of implementing SSL in FL, addressing hardware and algorithmic challenges. For video, Yan proposed a novel FL framework using stochastic weighted averaging and partial weight updates. For image data, Yan presented an aggregation scheme using angular divergence to address model divergence. If you’re interested in federated self-supervised learning, please check out our documentation.