Whilevariousdataaugmentation techniques havebeenactivelystudied intheFLenvironment, mostofthesemethods share input-level or feature-level data information over communication, posing potential privacyleakage.

A large body of research since the 1980s empirically observed that large neural networks can be compressed orsparsified toasmallfraction oftheiroriginal sizewhilemaintaining theirpredictive accuracy[14-16,20,23,29,45].