Exploration of distributed self-supervised training optimization strategies in visual tasks

In response to the growing demand for handling unlabeled data in visual tasks, this paper introduces a novel federated self-supervised learning model (FedSSL), which employs a federated learning framework to conduct distributed model training across multiple decentralized datasets. This model effectively harnesses the unlabeled data that is discreetly distributed among various terminals, thereby collaboratively training high-performance models. Moreover, comparative experiments conducted under standard experimental parameters and on general datasets demonstrate the model's efficacy. FedSSL not only reduces the computational complexity of the model but also enhances classification accuracy.