Tag
The paper introduces FAVoR, a method to measure and mitigate the loss of author-specific writing styles in federated learning for personalized language model adaptation, addressing the problem of author-style homogenization.