Large-scale pre-trained language models have achieved impressive results on a wide range of downstream tasks recently.

One ofthe most important tasks invideo understanding istounderstand human actions, which is one of the representativetasks for video understanding [54].

Stein discrepancies (SDs) monitor convergence andnon-convergence inapprox-imate inference when exact integration and sampling are intractable. However,the computation of a Stein discrepancy can be prohibitive if the Stein operator - often a sum over likelihood terms or potentials - is expensive to evaluate.

Consequently,thenoisy-to-clean transformation learned from the training data cannot be suitably applied to handle the testing noise, resulting in limited enhancement performance.

L(fk(xi),yi), (1) wherefk()andθk are the local model and model parameter,respectively. However,theyintroduced apublic dataset to enhance training, which is not practical. Overall, none of the above methods can be practically applied. Zhu et al. [14] also proposed a data-free knowledge distillation approach for FL, which learns a generator derived from the prediction of local models. In2019 IEEE 25Th international conference on parallel and distributed systems (ICPADS),pages985-989.IEEE,2019.