Autoencoder clustering boosts federated sentiment analysis under label skew
A transformer-autoencoder method that groups federated clients without labels improved sentiment-classification accuracy across controlled label-skew conditions on Sentiment140 and IMDB.
